/** * Copyright (c) 2021 OceanBase * OceanBase CE is licensed under Mulan PubL v2. * You can use this software according to the terms and conditions of the Mulan PubL v2. * You may obtain a copy of Mulan PubL v2 at: * http://license.coscl.org.cn/MulanPubL-2.0 * THIS SOFTWARE IS PROVIDED ON AN "AS IS" BASIS, WITHOUT WARRANTIES OF ANY KIND, * EITHER EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO NON-INFRINGEMENT, * MERCHANTABILITY OR FIT FOR A PARTICULAR PURPOSE. * See the Mulan PubL v2 for more details. */ #define USING_LOG_PREFIX STORAGE #include "lib/utility/ob_tracepoint.h" #include "lib/hash/ob_hashset.h" #include "share/ob_force_print_log.h" #include "storage/ob_partition_migrator.h" #include "storage/blocksstable/ob_block_sstable_struct.h" #include "storage/ob_sstable.h" #include "storage/ob_partition_service.h" #include "storage/ob_partition_log.h" #include "storage/ob_partition_base_data_ob_reader.h" #include "observer/ob_server_struct.h" #include "observer/ob_server.h" #include "clog/ob_clog_history_reporter.h" #include "share/scheduler/ob_sys_task_stat.h" #include "share/partition_table/ob_united_pt_operator.h" #include "storage/ob_partition_base_data_restore_reader.h" #include "storage/ob_partition_base_data_backup.h" #include "storage/ob_partition_base_data_validate.h" #include "share/ob_task_define.h" #include "share/backup/ob_backup_struct.h" #include "ob_migrate_logic_row_writer.h" #include "ob_table_mgr.h" #include "ob_migrate_macro_block_writer.h" #include "share/ob_debug_sync_point.h" #include "ob_storage_struct.h" #include "observer/omt/ob_tenant_node_balancer.h" #include "share/ob_force_print_log.h" #include "storage/ob_partition_migrator_table_key_mgr.h" #include "storage/ob_partition_migration_status.h" #include "storage/ob_pg_all_meta_checkpoint_reader.h" #include "storage/blocksstable/ob_store_file_system.h" #include "storage/ob_data_macro_id_iterator.h" #include "share/partition_table/ob_united_pt_operator.h" #include "share/backup/ob_backup_path.h" #include "storage/ob_partition_base_data_physical_restore.h" #include "storage/ob_file_system_util.h" #include "storage/ob_partition_backup.h" #include "storage/ob_pg_storage.h" #include "storage/ob_partition_scheduler.h" namespace oceanbase { using namespace common; using namespace common::hash; using namespace share; using namespace obrpc; using namespace blocksstable; using namespace oceanbase::share::schema; using namespace omt; using namespace transaction; namespace storage { #define MIGRATOR (::oceanbase::storage::ObPartitionMigrator::get_instance()) typedef common::hash::ObHashMap SSTableMap; int ObPartMigrationTask::assign(const ObPartMigrationTask& task) { int ret = common::OB_SUCCESS; if (task.ctx_.is_valid()) { ret = OB_NOT_SUPPORTED; LOG_ERROR("cannot assign valid ctx", K(ret), K(task)); } else { arg_ = task.arg_; status_ = task.status_; result_ = task.result_; need_reset_migrate_status_ = task.need_reset_migrate_status_; during_migrating_ = task.during_migrating_; } return ret; } ObMacroBlockReuseMgr::ObMacroBlockReuseMgr() : full_map_(), reuse_map_(), allocator_(ObModIds::OB_PARTITION_MIGRATOR) {} ObMacroBlockReuseMgr::~ObMacroBlockReuseMgr() { reset(); } // TODO(): split reuse macro block not support yet int ObMacroBlockReuseMgr::build_reuse_macro_map( ObMigrateCtx& ctx, const ObITable::TableKey& table_key, const common::ObIArray& macro_block_list) { int ret = OB_SUCCESS; ObTablesHandle tables_handle; ObDataMacroIdIterator data_macro_iter; ObIPartitionGroup* pg = ctx.get_partition(); if (OB_ISNULL(pg)) { ret = OB_ERR_SYS; LOG_WARN("error sys, pg must not be null", K(ret)); } else if (ObITable::MAJOR_SSTABLE != table_key.table_type_) { // only major sstable can reuse macro block } else if (OB_FAIL(data_macro_iter.init(ObPartitionService::get_instance(), pg))) { LOG_WARN("fail to init pg iterator", K(ret)); } else if (OB_FAIL(full_map_.init(ObModIds::OB_PARTITION_MIGRATOR, pg->get_partition_key().get_tenant_id()))) { LOG_WARN("fail to init hash map", K(ret)); } else if (!reuse_map_.is_inited() && OB_FAIL(reuse_map_.init(ObModIds::OB_PARTITION_MIGRATOR, pg->get_partition_key().get_tenant_id()))) { LOG_WARN("fail to init hash map", K(ret)); } else { ObMacroBlockInfoPair info; ObTenantFileKey file_key; const ObMacroBlockMetaV2* meta = nullptr; while (OB_SUCC(ret)) { if (OB_FAIL(data_macro_iter.get_next_macro_info(info, file_key))) { if (OB_ITER_END == ret) { ret = OB_SUCCESS; break; } } else { ObMajorMacroBlockKey key; meta = info.meta_.meta_; key.table_id_ = meta->table_id_; key.partition_id_ = meta->partition_id_; key.data_version_ = meta->data_version_; key.data_seq_ = meta->data_seq_; if (OB_FAIL(full_map_.insert_or_update(key, info))) { LOG_WARN("fail to set macro block map", K(ret)); } } } for (int64_t i = 0; OB_SUCC(ret) && i < macro_block_list.count(); ++i) { ObMacroBlockMetaV2* dst_meta = nullptr; ObMacroBlockSchemaInfo* dst_schema = nullptr; ObMacroBlockInfoPair info; ObMajorMacroBlockKey key; key.table_id_ = table_key.table_id_; key.partition_id_ = table_key.pkey_.get_partition_id(); key.data_version_ = macro_block_list.at(i).data_version_; key.data_seq_ = macro_block_list.at(i).data_seq_; if (OB_FAIL(full_map_.get(key, info))) { if (OB_ENTRY_NOT_EXIST == ret) { ret = OB_SUCCESS; } else { LOG_WARN("fail to get from hash map", K(ret), K(key)); } } else { FLOG_INFO("reuse a macro block", K(key), K(*info.meta_.meta_)); if (OB_FAIL(ctx.add_major_block_id(info.block_id_))) { LOG_WARN("fail to add major block id", K(ret)); } else if (OB_FAIL(info.meta_.meta_->deep_copy(dst_meta, allocator_))) { LOG_WARN("fail to deep copy meta", K(ret)); } else if (OB_FAIL(info.meta_.schema_->deep_copy(dst_schema, allocator_))) { LOG_WARN("fail to deep copy macro schema", K(ret)); } else { ObFullMacroBlockMeta full_meta; full_meta.schema_ = dst_schema; full_meta.meta_ = dst_meta; if (OB_FAIL(reuse_map_.insert_or_update(key, ObMacroBlockInfoPair(info.block_id_, full_meta)))) { LOG_WARN("fail to set reuse map", K(ret)); } } if (OB_FAIL(ret)) { if (nullptr != dst_meta) { dst_meta->~ObMacroBlockMetaV2(); dst_meta = nullptr; } if (nullptr != dst_schema) { dst_schema->~ObMacroBlockSchemaInfo(); dst_schema = nullptr; } } } } } full_map_.destroy(); return ret; } void ObMacroBlockReuseMgr::reset() { full_map_.destroy(); RemoveFunctor functor; reuse_map_.remove_if(functor); reuse_map_.destroy(); allocator_.reset(); } bool ObMacroBlockReuseMgr::RemoveFunctor::operator()( const blocksstable::ObMajorMacroBlockKey& block_key, blocksstable::ObMacroBlockInfoPair& info) { UNUSED(block_key); if (nullptr != info.meta_.meta_) { info.meta_.meta_->~ObMacroBlockMetaV2(); } if (nullptr != info.meta_.schema_) { info.meta_.schema_->~ObMacroBlockSchemaInfo(); } return true; } int ObMacroBlockReuseMgr::get_reuse_macro_meta( const blocksstable::ObMajorMacroBlockKey& block_key, blocksstable::ObMacroBlockInfoPair& info) { int ret = OB_SUCCESS; if (OB_FAIL(reuse_map_.get(block_key, info))) { if (OB_ENTRY_NOT_EXIST != ret) { LOG_WARN("fail to get from reuse map", K(ret)); } } return ret; } ObPartitionGroupInfoResult::ObPartitionGroupInfoResult() : result_(), choose_src_info_() {} void ObPartitionGroupInfoResult::reset() { result_.reset(); choose_src_info_.reset(); } int ObPartitionGroupInfoResult::assign(const ObPartitionGroupInfoResult& result) { int ret = OB_SUCCESS; reset(); if (OB_FAIL(result_.assign(result.result_))) { STORAGE_LOG(WARN, "fail to assign result", K(ret), K(result)); } else { choose_src_info_ = result.choose_src_info_; } return ret; } ObMigrateCtx::ObMigrateCtx() : lock_(), replica_op_arg_(), macro_indexs_(NULL), action_(UNKNOWN), result_(OB_SUCCESS), doing_task_cnt_(0), total_task_cnt_(0), need_rebuild_(false), partition_guard_(), replica_state_(OB_UNKNOWN_REPLICA), local_publish_version_(0), local_last_replay_log_id_(0), last_check_replay_ts_(0), create_ts_(0), task_id_(), copy_size_(0), continue_fail_count_(0), rebuild_count_(-1), finish_ts_(0), wait_replay_start_ts_(0), wait_minor_merge_start_ts_(0), last_confirmed_log_id_(OB_INVALID_ID), last_confirmed_log_ts_(OB_INVALID_TIMESTAMP), group_task_(NULL), restore_info_(), during_migrating_(false), need_online_for_rebuild_(false), trace_id_array_(), major_block_id_array_ptr_(NULL), curr_major_block_array_index_(0), can_rebuild_(true), pg_meta_(), part_ctx_array_(), need_offline_(false), is_restore_(share::REPLICA_NOT_RESTORE), use_slave_safe_read_ts_(false), need_report_checksum_(false), data_statics_(), is_copy_cover_minor_(true), mig_src_file_id_(OB_INVALID_DATA_FILE_ID), mig_dest_file_id_(OB_INVALID_DATA_FILE_ID), src_suspend_ts_(0), is_takeover_finished_(false), is_member_change_finished_(false), local_last_replay_log_ts_(0), old_trans_table_seq_(-1), create_new_pg_(false), restore_meta_reader_(nullptr), fetch_pg_info_compat_version_(ObFetchPGInfoArg::FETCH_PG_INFO_ARG_COMPAT_VERSION_V2), physical_backup_ctx_(), recovery_point_ctx_() {} ObMigrateCtx::~ObMigrateCtx() { for (int64_t i = 0; i < 2; ++i) { for (int64_t j = 0; j < major_block_id_array_[i].count(); ++j) { if (major_block_id_array_[i].at(j).is_valid()) { get_partition()->get_storage_file()->dec_ref(major_block_id_array_[i].at(j)); major_block_id_array_[i].at(j).reset(); } } } if (NULL != MIGRATOR.get_cp_fty()) { if (NULL != restore_meta_reader_) { MIGRATOR.get_cp_fty()->free(restore_meta_reader_); } if (NULL != macro_indexs_) { MIGRATOR.get_cp_fty()->free(macro_indexs_); } } } bool ObMigrateCtx::is_valid() const { return replica_op_arg_.is_valid() && UNKNOWN != action_ && doing_task_cnt_ >= 0 && total_task_cnt_ >= 0 && NULL != group_task_ // replica_state_ maybe OB_ERROR_REPLICA in init // partition maybe null // migrate_src mybe invalid ; } int ObMigrateCtx::notice_start_part_task() { int ret = OB_SUCCESS; if (NULL != group_task_) { int64_t backup_set_id = -1; if (VALIDATE_BACKUP_OP == replica_op_arg_.type_) { backup_set_id = replica_op_arg_.validate_arg_.backup_set_id_; } else if (BACKUP_BACKUPSET_OP == replica_op_arg_.type_) { backup_set_id = replica_op_arg_.backup_backupset_arg_.backup_set_id_; } if (OB_SUCCESS != (ret = group_task_->set_part_task_start(replica_op_arg_.key_, backup_set_id))) { STORAGE_LOG(ERROR, "set_part_task_start failed", K(ret), K(*this)); } else { STORAGE_LOG(INFO, "set_part_task_start", K(ret), K(*this)); } } return ret; } int ObMigrateCtx::notice_finish_part_task() { int ret = OB_SUCCESS; if (NULL != group_task_) { SMART_VAR(ObReplicaOpArg, tmp_replica_op_arg) { tmp_replica_op_arg = replica_op_arg_; int64_t backup_set_id = -1; if (VALIDATE_BACKUP_OP == replica_op_arg_.type_) { backup_set_id = replica_op_arg_.validate_arg_.backup_set_id_; } else if (BACKUP_BACKUPSET_OP == replica_op_arg_.type_) { backup_set_id = replica_op_arg_.backup_backupset_arg_.backup_set_id_; } if (OB_SUCCESS != (ret = group_task_->set_part_task_finish( replica_op_arg_.key_, result_, task_id_, during_migrating_, backup_set_id))) { STORAGE_LOG(ERROR, "set_part_task_finish failed", K(ret), K(tmp_replica_op_arg)); } else { STORAGE_LOG(INFO, "set_part_task_finish", K(ret), K(tmp_replica_op_arg)); } } } return ret; } void ObMigrateCtx::set_result_code(const int32_t result) { const bool write_lock = true; ObMigrateCtxGuard guard(write_lock, *this); if (OB_SUCCESS == result_) { result_ = result; LOG_WARN("record result code", K(result), K(lbt())); } } ObMigrateCtxGuard::ObMigrateCtxGuard(const bool is_write_lock, ObMigrateCtx& ctx) : ctx_(ctx) { int tmp_ret = OB_SUCCESS; if (is_write_lock) { if (OB_SUCCESS != (tmp_ret = ctx_.lock_.wrlock())) { STORAGE_LOG(ERROR, "failed to wrlock", K(tmp_ret)); } } else { if (OB_SUCCESS != (tmp_ret = ctx_.lock_.rdlock())) { STORAGE_LOG(ERROR, "failed to rdlock", K(tmp_ret)); } } } ObMigrateCtxGuard::~ObMigrateCtxGuard() { ctx_.lock_.unlock(); } int ObMigrateCtx::update_partition_migration_status() const { int ret = OB_SUCCESS; const ObMigrateCtx& ctx = *this; if (!ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(ctx)); } else { ObPartitionMigrationStatusGuard status_guard; ObPartitionMigrationStatus* status = NULL; if (OB_FAIL(ObPartitionMigrationStatusMgr::get_instance().get_status(ctx.task_id_, status_guard))) { STORAGE_LOG(WARN, "failed to get partition status", K(ret), K(ctx)); } else if (NULL == (status = status_guard.getStatus())) { ret = OB_ENTRY_NOT_EXIST; STORAGE_LOG(WARN, "status not found", K(ret), K(ctx)); } else { // status->task_id_; should not change // status->pkey_; should not change status->src_ = ctx.replica_op_arg_.data_src_.get_server(); // status->dest_; should not change status->clog_parent_ = ctx.clog_parent_; status->result_ = ctx.result_; status->start_time_ = ctx.create_ts_; status->action_ = ctx.action_; status->replica_state_ = ctx.replica_state_; status->doing_task_count_ = ctx.doing_task_cnt_; status->total_task_count_ = ctx.total_task_cnt_; status->rebuild_count_ = ctx.rebuild_count_; status->continue_fail_count_ = ctx.continue_fail_count_; status->finish_time_ = ctx.finish_ts_; status->data_statics_.total_macro_block_ = ATOMIC_LOAD(&ctx.data_statics_.total_macro_block_); status->data_statics_.ready_macro_block_ = ATOMIC_LOAD(&ctx.data_statics_.ready_macro_block_); status->data_statics_.major_count_ = ATOMIC_LOAD(&ctx.data_statics_.major_count_); status->data_statics_.mini_minor_count_ = ATOMIC_LOAD(&ctx.data_statics_.mini_minor_count_); status->data_statics_.normal_minor_count_ = ATOMIC_LOAD(&ctx.data_statics_.normal_minor_count_); status->data_statics_.buf_minor_count_ = ATOMIC_LOAD(&ctx.data_statics_.buf_minor_count_); // allow comment truncation, no need to set ret (void)ctx.fill_comment(status->comment_, sizeof(status->comment_)); STORAGE_LOG(INFO, "update migration status", K(*status)); } } return ret; } bool ObMigrateCtx::is_only_copy_sstable() const { return COPY_GLOBAL_INDEX_OP == replica_op_arg_.type_ || COPY_LOCAL_INDEX_OP == replica_op_arg_.type_ || RESTORE_REPLICA_OP == replica_op_arg_.type_ || RESTORE_FOLLOWER_REPLICA_OP == replica_op_arg_.type_ || BACKUP_REPLICA_OP == replica_op_arg_.type_ || RESTORE_STANDBY_OP == replica_op_arg_.type_ || VALIDATE_BACKUP_OP == replica_op_arg_.type_ || LINK_SHARE_MAJOR_OP == replica_op_arg_.type_ || BACKUP_BACKUPSET_OP == replica_op_arg_.type_; } bool ObMigrateCtx::is_copy_index() const { return COPY_GLOBAL_INDEX_OP == replica_op_arg_.type_ || COPY_LOCAL_INDEX_OP == replica_op_arg_.type_; } int ObMigrateCtx::add_major_block_id(const MacroBlockId& macro_block_id) { int ret = OB_SUCCESS; if (OB_ISNULL(major_block_id_array_ptr_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "major block id array ptr should not be NULL", K(ret)); } else { common::SpinWLockGuard guard(lock_); if (OB_FAIL(major_block_id_array_ptr_->push_back(macro_block_id))) { STORAGE_LOG(WARN, "fail to add major block key into array", K(ret)); } else if (OB_FAIL(get_partition()->get_storage_file()->inc_ref(macro_block_id))) { major_block_id_array_ptr_->pop_back(); STORAGE_LOG(WARN, "fail to inc ref, ", K(ret), K(macro_block_id)); } } return ret; } void ObMigrateCtx::free_old_macro_block() { int ret = OB_SUCCESS; common::SpinWLockGuard guard(lock_); int64_t index = (curr_major_block_array_index_ + 1) % 2; for (int64_t i = 0; i < major_block_id_array_[index].count(); ++i) { if (OB_FAIL(get_partition()->get_storage_file()->dec_ref(major_block_id_array_[index].at(i)))) { STORAGE_LOG(WARN, "fail to dec ref, ", K(ret), K(major_block_id_array_[index].at(i))); } } major_block_id_array_[index].reuse(); } void ObMigrateCtx::calc_need_retry() { common::SpinWLockGuard guard(lock_); if (!need_rebuild_) { LOG_INFO("ObParGroupMigrationTask do not need rebuild", K(*this)); } else if (rebuild_count_ < GCONF.sys_bkgd_migration_retry_num) { rebuild_count_++; } else { need_rebuild_ = false; } } void ObMigrateCtx::rebuild_migrate_ctx() { common::SpinWLockGuard guard(lock_); action_ = ObMigrateCtx::INIT; result_ = OB_SUCCESS; doing_task_cnt_ = 1; total_task_cnt_++; need_rebuild_ = false; need_online_for_rebuild_ = false; part_ctx_array_.reset(); need_report_checksum_ = false; data_statics_.reset(); trans_table_handle_.reset(); old_trans_table_seq_ = -1; create_new_pg_ = false; if (BACKUP_REPLICA_OP != replica_op_arg_.type_) { pg_meta_.reset(); } local_publish_version_ = 0; local_last_replay_log_ts_ = 0; fetch_pg_info_compat_version_ = ObFetchPGInfoArg::FETCH_PG_INFO_ARG_COMPAT_VERSION_V2; local_last_replay_log_id_ = 0; physical_backup_ctx_.reset(); recovery_point_ctx_.reset(); } int ObMigrateCtx::generate_and_schedule_migrate_dag() { int ret = OB_SUCCESS; ObBaseMigrateDag* dag = NULL; ObMigrateTaskSchedulerTask* scheduler_task = NULL; if (BACKUP_REPLICA_OP == replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; LOG_WARN("generate and schedule migrate dag get unexpected replica op", K(ret), K(*this)); } else { { common::SpinWLockGuard guard(lock_); if (OB_FAIL(alloc_migrate_dag(dag))) { LOG_WARN("failed to alloc migrate dag", K(ret)); } else if (OB_FAIL(dag->alloc_task(scheduler_task))) { STORAGE_LOG(WARN, "Fail to alloc task, ", K(ret)); } else if (OB_FAIL(scheduler_task->init())) { STORAGE_LOG(WARN, "failed to init prepare_task", K(ret)); } else if (OB_FAIL(dag->add_task(*scheduler_task))) { STORAGE_LOG(WARN, "Fail to add task", K(ret), K(replica_op_arg_)); } else if (OB_FAIL(ObDagScheduler::get_instance().add_dag(dag))) { STORAGE_LOG(WARN, "failed to add dag", K(ret), K(*dag)); if (OB_SIZE_OVERFLOW != ret && OB_EAGAIN != ret) { LOG_WARN("Fail to add task", K(ret)); ret = OB_EAGAIN; } } else { LOG_INFO("succeed to schedule migrate dag", K(*dag)); } } } if (OB_FAIL(ret)) { result_ = ret; if (NULL != dag) { dag->clear(); // release dag ctx_ ObDagScheduler::get_instance().free_dag(*dag); } } return ret; } int ObMigrateCtx::generate_and_schedule_backup_dag(const ObBackupDataType& backup_data_type) { int ret = OB_SUCCESS; ObBackupDag* backup_dag = NULL; ObBackupPrepareTask* prepare_task = NULL; if (!backup_data_type.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("generate and schedule backup dag get invalid argument", K(ret), K(backup_data_type)); } else if (BACKUP_REPLICA_OP != replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; LOG_WARN("generate and schedule backup dag get unexpected replica op", K(ret), K(*this)); } else { { common::SpinWLockGuard guard(lock_); if (OB_FAIL(ObDagScheduler::get_instance().alloc_dag(backup_dag))) { LOG_WARN("failed to alloc backup dag", K(ret)); } else if (OB_FAIL(backup_dag->init(backup_data_type, *this))) { LOG_WARN("failed to init migrate dag", K(ret)); } else if (OB_FAIL(backup_dag->alloc_task(prepare_task))) { STORAGE_LOG(WARN, "Fail to alloc task, ", K(ret)); } else if (OB_FAIL(prepare_task->init( *MIGRATOR.get_cp_fty(), *MIGRATOR.get_bandwidth_throttle(), *MIGRATOR.get_partition_service()))) { STORAGE_LOG(WARN, "failed to init prepare_task", K(ret)); } else if (OB_FAIL(backup_dag->add_task(*prepare_task))) { STORAGE_LOG(WARN, "Fail to add task", K(ret), K(replica_op_arg_)); } else if (OB_FAIL(ObDagScheduler::get_instance().add_dag(backup_dag))) { STORAGE_LOG(WARN, "failed to add dag", K(ret), K(*backup_dag)); if (OB_SIZE_OVERFLOW != ret && OB_EAGAIN != ret) { LOG_WARN("Fail to add task", K(ret)); ret = OB_EAGAIN; } } else { LOG_INFO("succeed to schedule backup dag", K(*backup_dag)); } } } if (OB_FAIL(ret)) { result_ = ret; if (NULL != backup_dag) { backup_dag->clear(); // release dag ctx_ ObDagScheduler::get_instance().free_dag(*backup_dag); } } return ret; } bool ObMigrateCtx::is_need_retry(const int result) const { bool ret = false; if (OB_NOT_INIT != result && OB_INVALID_ARGUMENT != result && OB_ERR_SYS != result && OB_INIT_TWICE != result && OB_ERR_UNEXPECTED != result && OB_LOG_NOT_SYNC != result && OB_NO_NEED_REBUILD != result && OB_SRC_DO_NOT_ALLOWED_MIGRATE != result && OB_NOT_SUPPORTED != result && OB_PG_IS_REMOVED != result && can_rebuild_) { ret = true; } return ret; } // if local has partition full data, no need copy, return true int ObMigrateCtx::change_replica_with_data(bool& is_replica_with_data) { int ret = OB_SUCCESS; is_replica_with_data = false; if (CHANGE_REPLICA_OP != replica_op_arg_.type_) { is_replica_with_data = false; LOG_INFO("not change replica op"); } else if (OB_ISNULL(partition_guard_.get_partition_group())) { ret = OB_INVALID_ARGUMENT; LOG_WARN("partition should not be NULL", K(ret)); } else { const ObReplicaType& replica_type = partition_guard_.get_partition_group()->get_replica_type(); is_replica_with_data = ObReplicaTypeCheck::is_replica_with_ssstore(replica_type); LOG_INFO("get local replica type", K(is_replica_with_data), K(replica_type)); } return ret; } int ObMigrateCtx::get_restore_clog_info(uint64_t& log_id, int64_t& acc_checksum) const { int ret = OB_SUCCESS; if (OB_UNLIKELY(!replica_op_arg_.is_physical_restore_leader())) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid argument", K(ret), K(replica_op_arg_)); } else if (OB_UNLIKELY(pg_meta_.is_valid())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "restore pg meta is invalid", K(ret), K(pg_meta_)); } else { log_id = pg_meta_.storage_info_.get_clog_info().get_last_replay_log_id(); acc_checksum = pg_meta_.storage_info_.get_clog_info().get_accumulate_checksum(); } return ret; } bool ObMigrateCtx::is_leader_restore_archive_data() const { return replica_op_arg_.is_physical_restore_leader() && REPLICA_RESTORE_ARCHIVE_DATA == is_restore_; } // reset is_restore in add_replica when src is in restore process int ObMigrateCtx::set_is_restore_for_add_replica(const int16_t src_is_restore) { int ret = OB_SUCCESS; if (ADD_REPLICA_OP != replica_op_arg_.type_) { // do nothing } else { switch (src_is_restore) { case REPLICA_NOT_RESTORE: case REPLICA_LOGICAL_RESTORE_DATA: case REPLICA_RESTORE_DATA: case REPLICA_RESTORE_STANDBY: case REPLICA_RESTORE_CUT_DATA: case REPLICA_RESTORE_STANDBY_CUT: is_restore_ = src_is_restore; break; case REPLICA_RESTORE_ARCHIVE_DATA: case REPLICA_RESTORE_LOG: case REPLICA_RESTORE_DUMP_MEMTABLE: is_restore_ = REPLICA_RESTORE_ARCHIVE_DATA; break; case REPLICA_RESTORE_WAIT_ALL_DUMPED: is_restore_ = REPLICA_RESTORE_WAIT_ALL_DUMPED; break; case REPLICA_RESTORE_MEMBER_LIST: { ret = OB_CANNOT_ADD_REPLICA_DURING_SET_MEMBER_LIST; STORAGE_LOG(WARN, "cannot add replica during set member list", K(ret), K(src_is_restore)); break; } default: { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "unknown restore type", K(ret), K(src_is_restore)); break; } } } return ret; } int ObMigrateCtx::alloc_migrate_dag(ObBaseMigrateDag*& base_migrate_dag) { int ret = OB_SUCCESS; base_migrate_dag = NULL; if (!replica_op_arg_.is_valid()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "replica op arg is invlaid", K(ret), K(*this)); } else if (VALIDATE_BACKUP_OP == replica_op_arg_.type_) { ObValidateDag* dag = NULL; if (OB_FAIL(ObDagScheduler::get_instance().alloc_dag(dag))) { LOG_WARN("failed to alloc validate dag", K(ret)); } else if (FALSE_IT(base_migrate_dag = dag)) { } else if (OB_FAIL(dag->init(*this))) { LOG_WARN("failed to init validate dag", K(ret)); } } else if (BACKUP_BACKUPSET_OP == replica_op_arg_.type_) { ObBackupBackupsetDag* dag = NULL; if (OB_FAIL(ObDagScheduler::get_instance().alloc_dag(dag))) { LOG_WARN("failed to alloc backup backupset dag", K(ret)); } else if (FALSE_IT(base_migrate_dag = dag)) { } else if (OB_FAIL(dag->init(*this))) { LOG_WARN("failed to init validate dag", K(ret)); } } else if (BACKUP_ARCHIVELOG_OP == replica_op_arg_.type_) { ObBackupArchiveLogDag* dag = NULL; if (OB_FAIL(ObDagScheduler::get_instance().alloc_dag(dag))) { LOG_WARN("failed to alloc backup archivelog dag", K(ret)); } else if (FALSE_IT(base_migrate_dag = dag)) { } else if (OB_FAIL(dag->init(*this))) { LOG_WARN("failed to init validate dag", K(ret)); } } else { ObMigrateDag* dag = NULL; if (OB_FAIL(ObDagScheduler::get_instance().alloc_dag(dag))) { LOG_WARN("failed to alloc migrate dag", K(ret)); } else if (FALSE_IT(base_migrate_dag = dag)) { } else if (OB_FAIL(dag->init(*this))) { LOG_WARN("failed to init migrate dag", K(ret)); } } if (OB_FAIL(ret)) { if (NULL != base_migrate_dag) { base_migrate_dag->clear(); // release dag ctx_ ObDagScheduler::get_instance().free_dag(*base_migrate_dag); base_migrate_dag = nullptr; } } return ret; } ObPartitionMigrateCtx::ObPartitionMigrateCtx() : ObIPartitionMigrateCtx(), copy_info_(), handle_(), is_partition_exist_(true), lock_(), need_reuse_local_minor_(true) {} bool ObPartitionMigrateCtx::is_valid() const { bool bret = NULL != ctx_ && ctx_->is_valid() && copy_info_.is_valid(); if (bret) { if (ctx_->is_copy_index() && !handle_.empty()) { bret = false; } } return bret; } void ObPartitionMigrateCtx::reset() { ctx_ = NULL; copy_info_.reset(); handle_.reset(); is_partition_exist_ = true; need_reuse_local_minor_ = true; } int ObPartitionMigrateCtx::assign(const ObPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; if (!part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("assign partition migrate ctx get invalid argument", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(copy_info_.assign(part_migrate_ctx.copy_info_))) { LOG_WARN("failed to assign copy info", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(handle_.assign(part_migrate_ctx.handle_))) { LOG_WARN("failed to assign handle", K(ret)); } else { ctx_ = part_migrate_ctx.ctx_; is_partition_exist_ = part_migrate_ctx.is_partition_exist_; need_reuse_local_minor_ = part_migrate_ctx.need_reuse_local_minor_; } return ret; } int ObPartitionMigrateCtx::add_sstable(ObSSTable& sstable) { int ret = OB_SUCCESS; ObIPartitionGroup* pg = NULL; const int64_t max_kept_major_version_number = 0; const bool in_slog_trans = false; if (OB_ISNULL(ctx_)) { ret = OB_ERR_SYS; LOG_WARN("ctx must not null", K(ret), K(*this)); } else if (!ctx_->is_copy_index()) { common::SpinWLockGuard guard(lock_); if (OB_FAIL(handle_.add_table(&sstable))) { LOG_WARN("failed to add table", K(ret)); } else { FLOG_INFO("hold sstable in ctx", "pkey", copy_info_.meta_.pkey_, "table_key", sstable.get_key()); } } else { if (OB_ISNULL(pg = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret)); } else if (OB_FAIL(pg->get_pg_storage().add_sstable( copy_info_.meta_.pkey_, &sstable, max_kept_major_version_number, in_slog_trans))) { LOG_WARN("failed to add sstable", K(ret)); } else { FLOG_INFO("add sstable to pg", "pkey", copy_info_.meta_.pkey_, "table_key", sstable.get_key()); } } return ret; } ObMigrateRecoveryPointCtx::ObMigrateRecoveryPointCtx() : ObIPartitionMigrateCtx(), recovery_point_index_(0), recovery_point_key_array_(), tables_handle_map_(), lock_() {} ObMigrateRecoveryPointCtx::~ObMigrateRecoveryPointCtx() { reset(); } int ObMigrateRecoveryPointCtx::init(ObMigrateCtx& migrate_ctx) { int ret = OB_SUCCESS; if (is_valid()) { ret = OB_INIT_TWICE; LOG_WARN("init twice", K(ret)); } else if (OB_FAIL(tables_handle_map_.create(BUCKET_NUM, ObModIds::OB_PARTITION_MIGRATOR))) { LOG_WARN("failed to create tables handle map", K(ret)); } else { ctx_ = &migrate_ctx; } return ret; } void ObMigrateRecoveryPointCtx::reset() { reuse(); tables_handle_map_.reuse(); } // for inner retry // just reuse recovery_point_index_ and recovery_point_key_array_ void ObMigrateRecoveryPointCtx::reuse() { recovery_point_index_ = 0; recovery_point_key_array_.reset(); } bool ObMigrateRecoveryPointCtx::is_valid() const { return tables_handle_map_.created() && recovery_point_index_ >= 0 && recovery_point_key_array_.count() >= 0; } int ObMigrateRecoveryPointCtx::add_sstable(ObSSTable& sstable) { int ret = OB_SUCCESS; ObTableHandle table_handle; if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (!sstable.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("add sstable get invalid argument", K(ret), K(sstable)); } else if (OB_FAIL(table_handle.set_table(&sstable))) { LOG_WARN("failed to set table", K(ret), K(sstable)); } else if (OB_FAIL(add_sstable(table_handle))) { LOG_WARN("failed to add sstable", K(ret), K(table_handle)); } return ret; } int ObMigrateRecoveryPointCtx::get_recovery_point_info( const int64_t recovery_point_index, ObRecoveryPointKey& recovery_point_key) { int ret = OB_SUCCESS; recovery_point_key.reset(); if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (recovery_point_index < 0 || recovery_point_index >= recovery_point_key_array_.count()) { ret = OB_INVALID_ARGUMENT; LOG_WARN( "get recovery point info get invalid argument", K(ret), K(recovery_point_index), K(recovery_point_key_array_)); } else { recovery_point_key = recovery_point_key_array_.at(recovery_point_index); } return ret; } int ObMigrateRecoveryPointCtx::get_current_recovery_point_info(ObRecoveryPointKey& recovery_point_key) { int ret = OB_SUCCESS; recovery_point_key.reset(); if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (recovery_point_index_ >= recovery_point_key_array_.count()) { ret = OB_ITER_END; LOG_WARN("get current recovery point info failed", K(ret), K(recovery_point_index_), K(recovery_point_key_array_)); } else { recovery_point_key = recovery_point_key_array_.at(recovery_point_index_); } return ret; } int ObMigrateRecoveryPointCtx::add_sstable(ObTableHandle& table_handle) { int ret = OB_SUCCESS; ObITable* table = NULL; if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (OB_ISNULL(table = table_handle.get_table())) { ret = OB_INVALID_ARGUMENT; LOG_WARN("table is NULL", K(ret), K(table_handle)); } else { common::SpinWLockGuard guard(lock_); if (OB_FAIL(tables_handle_map_.set_refactored(table->get_key(), table_handle))) { if (OB_HASH_EXIST == ret) { ret = OB_SUCCESS; } else { LOG_WARN("failed to set table handle into map", K(ret), K(table_handle)); } } } return ret; } bool ObMigrateRecoveryPointCtx::is_recovery_point_iter_finish() const { return recovery_point_index_ >= recovery_point_key_array_.count(); } bool ObMigrateRecoveryPointCtx::is_recovery_point_index_valid() const { return recovery_point_index_ >= 0 && recovery_point_index_ <= recovery_point_key_array_.count(); } int ObMigrateRecoveryPointCtx::get_sstable(const ObITable::TableKey& table_key, ObTableHandle& table_handle) { int ret = OB_SUCCESS; table_handle.reset(); if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (!table_key.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("get sstable get invalid argument", K(ret), K(table_key)); } else { common::SpinRLockGuard guard(lock_); if (OB_FAIL(tables_handle_map_.get_refactored(table_key, table_handle))) { LOG_WARN("failed to get table handle from map", K(ret), K(table_key)); } } return ret; } bool ObMigrateRecoveryPointCtx::is_tables_handle_empty() const { return 0 == tables_handle_map_.size(); } int ObMigrateRecoveryPointCtx::get_all_recovery_point_key(ObIArray& recovery_point_key_array) { int ret = OB_SUCCESS; if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("recovery point ctx do not init", K(ret)); } else if (OB_FAIL(recovery_point_key_array.assign(recovery_point_key_array_))) { LOG_WARN("failed to assign recovery point key array", K(ret)); } return ret; } int ObMigrateRecoveryPointCtx::remove_unneed_table_handle(const ObHashSet& table_key_set) { int ret = OB_SUCCESS; int hash_ret = OB_SUCCESS; hash::ObHashMap::const_iterator iter; ObArray need_removed_table_keys; if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("recovery point ctx do not init", K(ret)); } else { common::SpinWLockGuard guard(lock_); for (iter = tables_handle_map_.begin(); OB_SUCC(ret) && iter != tables_handle_map_.end(); ++iter) { const ObITable::TableKey& table_key = iter->first; hash_ret = table_key_set.exist_refactored(table_key); if (OB_HASH_EXIST == hash_ret) { } else if (OB_HASH_NOT_EXIST == hash_ret) { if (OB_FAIL(need_removed_table_keys.push_back(table_key))) { LOG_WARN("failed to push table key into array", K(ret), K(table_key)); } } else { ret = OB_SUCCESS == hash_ret ? OB_ERR_UNEXPECTED : hash_ret; } } for (int64_t i = 0; OB_SUCC(ret) && i < need_removed_table_keys.count(); ++i) { const ObITable::TableKey& table_key = need_removed_table_keys.at(i); if (OB_FAIL(tables_handle_map_.erase_refactored(table_key))) { LOG_WARN("failed to erase table key from map", K(ret), K(table_key)); } } } return ret; } int64_t ObMigrateRecoveryPointCtx::to_string(char* buf, const int64_t buf_len) const { int64_t pos = 0; if (OB_ISNULL(buf) || buf_len <= 0) { // do nothing } else { J_OBJ_START(); J_KV(KP(this), K_(recovery_point_index), K_(recovery_point_key_array)); J_COMMA(); J_OBJ_END(); } return pos; } int ObMigrateFinishPhysicalTask::check_sstable_meta( const blocksstable::ObSSTableBaseMeta& src_meta, const blocksstable::ObSSTableBaseMeta& write_meta) { int ret = OB_SUCCESS; if (src_meta.index_id_ != write_meta.index_id_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "index_id_ not match", K(ret)); } else if (src_meta.row_count_ != write_meta.row_count_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "row_count_ not match", K(ret)); } else if (src_meta.occupy_size_ != write_meta.occupy_size_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "occupy_size_ not match", K(ret)); } else if (src_meta.data_checksum_ != write_meta.data_checksum_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "data checksum not match", K(ret)); } else if (src_meta.row_checksum_ != write_meta.row_checksum_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "row_checksum_ not match", K(ret)); } else if (src_meta.data_version_ != write_meta.data_version_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "data_version_ not match", K(ret)); } else if (src_meta.rowkey_column_count_ != write_meta.rowkey_column_count_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "rowkey_column_count_ not match", K(ret)); } else if (src_meta.table_type_ != write_meta.table_type_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "table_type_ not match", K(ret)); } else if (src_meta.index_type_ != write_meta.index_type_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "index_type_ not match", K(ret)); } else if (src_meta.macro_block_count_ != write_meta.macro_block_count_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "macro_block_count_ not match", K(ret)); } else if (src_meta.lob_macro_block_count_ != write_meta.lob_macro_block_count_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "lob_macro_block_count_ not match", K(ret)); } else if (src_meta.column_cnt_ != write_meta.column_cnt_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "column_cnt_ not match", K(ret)); } else if (src_meta.total_sstable_count_ != write_meta.total_sstable_count_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "total_sstable_count_ not match", K(ret)); } else if (src_meta.max_logic_block_index_ != write_meta.max_logic_block_index_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "max_logic_block_index_ not match", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < src_meta.column_cnt_; ++i) { const ObSSTableColumnMeta& src_col = src_meta.column_metas_.at(i); const ObSSTableColumnMeta& write_col = write_meta.column_metas_.at(i); if (src_col.column_id_ != write_col.column_id_ || src_col.column_default_checksum_ != write_col.column_default_checksum_ || src_col.column_checksum_ != write_col.column_checksum_) { ret = OB_INVALID_DATA; STORAGE_LOG(WARN, "column_metas_ not match", K(ret), K(i), K(src_meta), K(write_meta)); } } } return ret; } int ObMigratePrepareTask::create_new_partition(const ObAddr& src_server, ObReplicaOpArg& replica_op_arg, ObIPartitionGroupGuard& partition_guard, ObStorageFileHandle& file_handle) { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; clog::ObIPartitionLogService* pls = NULL; const int64_t is_restore = ctx_->is_restore_; ObTablesHandle sstables_handle; // empty ctx_->pg_meta_.storage_info_.set_pg_file_id(OB_INVALID_DATA_FILE_ID); // for create a new pg_file in ofs mode const ObSavedStorageInfoV2& saved_storage_info = ctx_->pg_meta_.storage_info_; const ObPartitionSplitInfo& split_info = ctx_->pg_meta_.split_info_; int64_t split_state = ctx_->pg_meta_.saved_split_state_; // 221 and 222 observer may set split_state = -1, upgrade to 224 may has problem, now set split_state = 1 directly if (-1 == split_state) { split_state = 1; LOG_WARN("split state was -1, should be rewritten to 1", "pkey", ctx_->pg_meta_.pg_key_); } const ObBaseStorageInfo& clog_info = saved_storage_info.get_clog_info(); common::ObReplicaType replica_type = replica_op_arg.dst_.get_replica_type(); const int64_t memstore_percent = replica_op_arg.dst_.get_memstore_percent(); ObCreatePGParam param; common::ObReplicaProperty replica_property; // TODO: need using RS appoint replica property ObMigrateStatus migrate_status; bool need_create_memtable = false; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "task has not initialized.", K(ret)); } else if (OB_FAIL(partition_service_->create_new_partition(replica_op_arg.key_, partition))) { STORAGE_LOG(WARN, "Fail to create new partition, ", K(replica_op_arg.key_), K(ret)); } else if (NULL == partition) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "The partition is NULL, ", K(ret), K(replica_op_arg.key_)); } else if (OB_FAIL(ObMigrateStatusHelper::trans_replica_op(replica_op_arg.type_, migrate_status))) { LOG_WARN("failed to trans migrate status", K(ret), K(replica_op_arg)); } else { partition_guard.set_partition_group(partition_service_->get_pg_mgr(), *partition); ObChangePartitionLogEntry entry; const int64_t subcmd = ObIRedoModule::gen_subcmd(OB_REDO_LOG_PARTITION, REDO_LOG_ADD_PARTITION_GROUP); const bool write_slog = true; const int64_t data_version = 0; const bool in_slog_trans = true; entry.partition_key_ = ctx_->replica_op_arg_.key_; entry.replica_type_ = replica_type; // pg and partition witch contains pg_key and partition_key are same entry.pg_key_ = ctx_->replica_op_arg_.key_; ObStorageLogAttribute log_attr; param.create_frozen_version_ = ctx_->pg_meta_.create_frozen_version_; param.last_restore_log_id_ = ctx_->pg_meta_.last_restore_log_id_; param.last_restore_log_ts_ = ctx_->pg_meta_.last_restore_log_ts_; param.restore_snapshot_version_ = ctx_->pg_meta_.restore_snapshot_version_; param.restore_schema_version_ = ctx_->pg_meta_.restore_schema_version_; if (NULL == (pls = partition->get_log_service())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "Unexpected error, pls is NULL, ", K(ret)); } else if (OB_FAIL(replica_property.set_memstore_percent(memstore_percent))) { STORAGE_LOG(WARN, "failed to set memstore percent", K(ret), K(replica_op_arg), K(memstore_percent)); } else if (FALSE_IT(param.create_timestamp_ = ctx_->pg_meta_.create_timestamp_)) { } else if (FALSE_IT(param.replica_property_ = replica_property)) { } else if (FALSE_IT(param.replica_type_ = replica_type)) { } else if (FALSE_IT(param.is_restore_ = is_restore)) { } else if (FALSE_IT(param.write_slog_ = write_slog)) { } else if (FALSE_IT(param.data_version_ = data_version)) { } else if (FALSE_IT(param.split_state_ = split_state)) { } else if (FALSE_IT(param.file_handle_ = &file_handle)) { } else if (FALSE_IT(param.file_mgr_ = &OB_SERVER_FILE_MGR)) { } else if (FALSE_IT(param.migrate_status_ = migrate_status)) { } else if (OB_FAIL(param.set_storage_info(saved_storage_info))) { LOG_WARN("failed to set storage info", K(ret), K(saved_storage_info)); } else if (OB_FAIL(param.set_split_info(split_info))) { LOG_WARN("failed to set split info", K(ret), K(split_info)); } else if (OB_FAIL(partition_service_->get_clog_mgr()->add_partition(replica_op_arg.key_, replica_type, replica_property, clog_info, ObVersion(0, 0) /*freeze version*/, src_server, is_restore, pls))) { STORAGE_LOG(WARN, "fail to initiate partition log service", K(ret), K(replica_op_arg)); // migrate partition will not refresh schema, need set region } else if (OB_FAIL(partition_service_->set_region(replica_op_arg.key_, pls))) { STORAGE_LOG(WARN, "fail to set region", K(ret), K(replica_op_arg)); } else { log_attr.tenant_id_ = entry.pg_key_.get_tenant_id(); log_attr.data_file_id_ = file_handle.get_storage_file()->get_file_id(); param.migrate_status_ = migrate_status; } if (OB_FAIL(ret)) { } else if (OB_FAIL(SLOGGER.write_log(subcmd, log_attr, entry))) { STORAGE_LOG(WARN, "fail to write add partition (group) log.", K(ret), K(subcmd), K(log_attr)); } else if (OB_FAIL(partition->create_partition_group(param))) { STORAGE_LOG(WARN, "fail to create partition group", K(ret)); } else if (FALSE_IT(need_create_memtable = !(partition->get_pg_storage().is_restoring_base_data() || partition->get_pg_storage().is_restoring_standby()))) { } else if (need_create_memtable && OB_FAIL(partition->create_memtable(in_slog_trans))) { STORAGE_LOG(WARN, "fail to create memtable", K(ret)); // pause it to make sure the sstable upper trans version is update after migration finished. } else if (need_create_memtable && OB_FALSE_IT(partition->get_pg_storage().pause())) { } else { file_handle.reset(); // create new partition need set schema_version in pg meta in order to prevent gc for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { const ObMigratePartitionInfo& copy_info = ctx_->part_ctx_array_.at(i).copy_info_; const ObPGPartitionStoreMeta& meta = copy_info.meta_; if (!meta.pkey_.is_trans_table() && OB_FAIL(create_pg_partition(meta, partition, in_slog_trans))) { LOG_WARN("failed to create pg partition", K(ret), K(meta)); } } } } return ret; } //////////////////////////////////////////////////////////////////// // ObPartitionMigrator // ObPartitionMigrator::ObPartitionMigrator() : is_inited_(false), is_stop_(false), bandwidth_throttle_(NULL), partition_service_(NULL), svr_rpc_proxy_(NULL), pts_rpc_(NULL), cp_fty_(NULL), location_cache_(NULL), lock_(ObLatchIds::MIGRATE_LOCK), schema_service_(NULL) {} ObPartitionMigrator::~ObPartitionMigrator() { if (is_inited_) { destroy(); } } void ObPartitionMigrator::destroy() { stop(); wait(); } void ObPartitionMigrator::stop() { common::SpinWLockGuard guard(lock_); is_stop_ = true; } void ObPartitionMigrator::wait() { STORAGE_LOG(INFO, "wait restore_macro_block_task_pool_ stop"); restore_macro_block_task_pool_.destroy(); STORAGE_LOG(INFO, "wait migrate task stop"); if (is_inited_) { bool has_running_task = true; while (has_running_task) { int64_t migrate_dag_count = ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_MIGRATE) + ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_BACKUP) + ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_VALIDATE) + ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_BACKUP_BACKUPSET) + ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_BACKUP_ARCHIVELOG); has_running_task = migrate_dag_count > 0; if (has_running_task) { usleep(OB_MIGRATE_SCHEDULE_SLEEP_INTERVAL_S); } } } is_inited_ = false; STORAGE_LOG(INFO, "ObPartitionMigrator stopped"); } int ObMigrateTaskSchedulerTask::add_migrate_status(ObMigrateCtx* ctx) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (NULL == ctx) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid arg", K(ret)); } else if (!ctx->is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(*ctx)); } else if (OB_FAIL(add_partition_migration_status(*ctx))) { STORAGE_LOG(WARN, "failed to add partition_migration_status", K(ret), K(*ctx)); } else { STORAGE_LOG(INFO, "succeed to add migrate ctx", KP(ctx), K(*ctx)); } return ret; } int ObMigratePrepareTask::add_partition_migration_status(const ObMigrateCtx& ctx) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(ctx)); } else { ObPartitionMigrationStatus task; task.task_id_ = ctx.task_id_; task.migrate_type_ = ctx.replica_op_arg_.get_replica_op_type_str(); task.pkey_ = ctx.replica_op_arg_.key_; task.clog_parent_.reset(); // we don't know it now task.src_.reset(); // we don't know it now task.dest_ = OBSERVER.get_self(); task.result_ = ctx.result_; task.start_time_ = ctx.create_ts_; task.action_ = ctx.action_; task.replica_state_ = OB_UNKNOWN_REPLICA; // we don't know it now task.doing_task_count_ = 0; task.total_task_count_ = 0; task.rebuild_count_ = 0; task.continue_fail_count_ = 0; task.data_statics_ = ctx.data_statics_; // allow comment truncation, no need to set ret (void)ctx.fill_comment(task.comment_, sizeof(task.comment_)); if (OB_SUCCESS != (tmp_ret = ObPartitionMigrationStatusMgr::get_instance().add_status(task))) { STORAGE_LOG(WARN, "failed to add partition migration status", K(tmp_ret), K(task)); } } return ret; } int ObMigrateCtx::fill_comment(char* buf, const int64_t buf_len) const { // static function, no need check is_inited_ int ret = OB_SUCCESS; if (NULL == buf || NULL == group_task_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), KP(group_task_), K(buf)); } else { int n = snprintf(buf, buf_len, "single partition migration: group_task_id=%s key=%s, op_type=%s, src=%s, dest=%s," "dest_memstore_percent=%s", to_cstring(group_task_->get_task_id()), to_cstring(replica_op_arg_.key_), replica_op_arg_.get_replica_op_type_str(), to_cstring(replica_op_arg_.data_src_.get_server()), to_cstring(replica_op_arg_.dst_.get_server()), to_cstring(replica_op_arg_.dst_.get_memstore_percent())); if (n < 0 || n >= buf_len) { ret = OB_BUF_NOT_ENOUGH; if (REACH_TIME_INTERVAL(10 * 1000 * 1000)) { // 10s STORAGE_LOG(WARN, "buf not enough, comment is truncated", K(ret), K(n), K(buf)); } } } return ret; } ObPartitionMigrator& ObPartitionMigrator::get_instance() { static ObPartitionMigrator migrator_instance_; return migrator_instance_; } int ObPartitionMigrator::init(obrpc::ObPartitionServiceRpcProxy& srv_rpc_proxy, ObPartitionServiceRpc& pts_rpc, ObIPartitionComponentFactory* cp_fty, ObPartitionService* partition_service, common::ObInOutBandwidthThrottle& bandwidth_throttle, share::ObIPartitionLocationCache* location_cache, share::schema::ObMultiVersionSchemaService* schema_service) { int ret = OB_SUCCESS; common::SpinWLockGuard guard(lock_); if (is_inited_) { ret = OB_INIT_TWICE; STORAGE_LOG(WARN, "The ObPartitionMigrator has inited.", K(ret)); } else if (!srv_rpc_proxy.is_inited() || NULL == cp_fty || NULL == partition_service || NULL == location_cache || NULL == schema_service) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid arguments.", "proxy valid", srv_rpc_proxy.is_inited(), KP(cp_fty), KP(partition_service), KP(location_cache), KP(schema_service), K(ret)); } else if (OB_FAIL(restore_macro_block_task_pool_.init("RstMacroTask"))) { STORAGE_LOG(WARN, "failed to init restore_macro_block_task_pool_", K(ret)); } else { bandwidth_throttle_ = &bandwidth_throttle; partition_service_ = partition_service; svr_rpc_proxy_ = &srv_rpc_proxy; pts_rpc_ = &pts_rpc; cp_fty_ = cp_fty; location_cache_ = location_cache; schema_service_ = schema_service; is_inited_ = true; } return ret; } const char* ObMigrateCtx::trans_action_to_str(const MigrateAction& action) { const char* str = "UNKNOWN"; switch (action) { case INIT: str = "INIT"; break; case PREPARE: str = "PREPARE"; break; case COPY_MINOR: str = "COPY_MINOR"; break; case COPY_MAJOR: str = "COPY_MAJOR"; break; case END: str = "END"; break; default: str = "UNKOWN"; STORAGE_LOG(WARN, "unknown action", K(action)); } return str; } ObPartGroupMigrator::DoingTaskStat::DoingTaskStat() : rebuild_count_(0), backup_count_(0), validate_count_(0), backup_backupset_count_(0), backup_archivelog_count_(0), normal_migrate_count_(0) {} void ObPartGroupMigrator::DoingTaskStat::reset() { rebuild_count_ = 0; backup_count_ = 0; validate_count_ = 0; backup_backupset_count_ = 0; backup_archivelog_count_ = 0; normal_migrate_count_ = 0; } ObPartGroupMigrator::ObPartGroupMigrator() : is_inited_(false), update_task_list_lock_(ObLatchIds::GROUP_MIGRATE_LOCK), task_list_(), partition_service_(NULL), cp_fty_(NULL), is_stop_(false) {} ObPartGroupMigrator::~ObPartGroupMigrator() { if (is_inited_) { destroy(); } } void ObPartGroupMigrator::destroy() { stop(); wait(); } void ObPartGroupMigrator::stop() { common::SpinWLockGuard guard(update_task_list_lock_); is_stop_ = true; } void ObPartGroupMigrator::wait() { bool need_wait = true; while (need_wait) { common::SpinRLockGuard guard(update_task_list_lock_); if (task_list_.count() > 0) { STORAGE_LOG(INFO, "not finish part group task, waiting", "total_count", task_list_.count()); usleep(1000 * 1000); // sleep 1s } else { need_wait = false; } } is_inited_ = false; } bool ObPartGroupMigrator::is_stop() const { return is_stop_; } ObPartGroupMigrator& ObPartGroupMigrator::get_instance() { static ObPartGroupMigrator group_migrator; return group_migrator; } int ObPartGroupMigrator::init(storage::ObPartitionService* partition_service, ObIPartitionComponentFactory* cp_fty) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; STORAGE_LOG(ERROR, "cannot init twice", K(ret)); } else if (NULL == partition_service || NULL == cp_fty) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), KP(partition_service), KP(cp_fty)); } else { is_inited_ = true; partition_service_ = partition_service; cp_fty_ = cp_fty; } return ret; } int ObPartGroupMigrator::schedule(const ObReplicaOpArg& arg, const share::ObTaskId& task_id) { int ret = OB_SUCCESS; // task will be executed directly when it is scheduled and will be release when it execute over, // so this task point address will not keep effective ObPartGroupTask* group_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!arg.is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(task_id), K(arg)); } else { ObSEArray task_list; const bool is_batch_mode = false; const bool is_normal_migrate = true; if (OB_FAIL(task_list.push_back(arg))) { STORAGE_LOG(WARN, "failed to add task list", K(ret), K(arg)); } else if (OB_FAIL(inner_schedule(task_list, is_batch_mode, task_id, is_normal_migrate, group_task))) { STORAGE_LOG(WARN, "failed to inner schedule task", K(ret), K(arg), K(task_id)); } } return ret; } int ObPartGroupMigrator::schedule(const ObIArray& arg_list, const share::ObTaskId& task_id) { int ret = OB_SUCCESS; const bool is_batch_mode = true; // task will be executed directly when it is scheduled and will be release when it execute over, // so this task point address will not keep effective ObPartGroupTask* group_task = NULL; bool is_normal_migrate = true; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (arg_list.count() <= 0) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(task_id), K(arg_list)); } else if (OB_FAIL(inner_schedule(arg_list, is_batch_mode, task_id, is_normal_migrate, group_task))) { STORAGE_LOG(WARN, "failed to inner schedule task", K(ret), K(arg_list), K(task_id)); } return ret; } int ObPartGroupMigrator::mark(const ObReplicaOpArg& arg, const share::ObTaskId& task_id, ObPartGroupTask*& group_task) { // only remove replica task finsih is control outside, so allo outside call group_task to finish int ret = OB_SUCCESS; group_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!arg.is_valid() || task_id.is_invalid() || REMOVE_REPLICA_OP != arg.type_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(task_id), K(arg)); } else { SMART_VAR(ObArray, task_list) { const bool is_batch_mode = false; const bool is_normal_migrate = false; if (OB_FAIL(task_list.push_back(arg))) { STORAGE_LOG(WARN, "failed to add task list", K(ret), K(arg)); } else if (OB_FAIL(inner_schedule(task_list, is_batch_mode, task_id, is_normal_migrate, group_task))) { STORAGE_LOG(WARN, "failed to inner schedule task", K(ret), K(arg), K(task_id)); } DEBUG_SYNC(AFTER_MIGRATION_MARK); } } return ret; } int ObPartGroupMigrator::get_not_finish_task_count(DoingTaskStat& stat) { // caller must hold lock int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { if (!task_list_[i]->is_finish() && REMOVE_REPLICA_OP != task_list_[i]->get_type()) { if (REBUILD_REPLICA_OP == task_list_[i]->get_type()) { ++stat.rebuild_count_; } else if (BACKUP_REPLICA_OP == task_list_[i]->get_type()) { ++stat.backup_count_; } else if (BACKUP_BACKUPSET_OP == task_list_[i]->get_type()) { ++stat.backup_backupset_count_; } else if (BACKUP_ARCHIVELOG_OP == task_list_[i]->get_type()) { ++stat.backup_archivelog_count_; } else if (VALIDATE_BACKUP_OP == task_list_[i]->get_type()) { ++stat.validate_count_; } else { ++stat.normal_migrate_count_; } STORAGE_LOG(DEBUG, "dump task list", K(i), K(*task_list_[i])); } } } return ret; } static inline bool task_need_data_copy(const ObIArray& task_list) { bool bool_ret = false; for (int64_t i = 0; i < task_list.count() && !bool_ret; ++i) { const ObReplicaOpArg& task = task_list.at(i); if (REMOVE_REPLICA_OP != task.type_) { bool_ret = true; } } return bool_ret; } int ObPartGroupMigrator::has_task(const ObPartitionKey& pkey, bool& has_task) { int ret = OB_SUCCESS; ObArray pkey_list; common::SpinRLockGuard guard(update_task_list_lock_); has_task = false; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(pkey_list.push_back(pkey))) { LOG_WARN("failed to add pkey", K(ret), K(pkey)); } else if (OB_FAIL(check_dup_task(pkey_list))) { if (OB_TASK_EXIST != ret) { LOG_WARN("failed to check dup task", K(ret), K(pkey)); } else { ret = OB_SUCCESS; has_task = true; } } return ret; } int ObPartGroupMigrator::task_exist(const share::ObTaskId& task_id, bool& exist) { int ret = OB_SUCCESS; exist = false; common::SpinRLockGuard guard(update_task_list_lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count() && !exist; ++i) { if (NULL != task_list_[i] && task_id.equals(task_list_[i]->get_task_id())) { exist = true; } } return ret; } int ObPartGroupMigrator::check_copy_limit_(const ObIArray& arg_list) { int ret = OB_SUCCESS; int64_t data_copy_in_limit = GCONF.server_data_copy_in_concurrency; DoingTaskStat stat; ObReplicaOpType type = UNKNOWN_REPLICA_OP; if (GCONF.migrate_concurrency == 0) { data_copy_in_limit = 0; } if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (arg_list.count() <= 0) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(arg_list)); } else if (FALSE_IT(type = arg_list.at(0).type_)) { } else if (OB_FAIL(get_not_finish_task_count(stat))) { STORAGE_LOG(WARN, "failed to get_not_finish_task_count", K(ret)); } else if (!task_need_data_copy(arg_list)) { // pass } else if (REBUILD_REPLICA_OP == type) { if (stat.rebuild_count_ >= data_copy_in_limit) { ret = OB_REACH_SERVER_DATA_COPY_IN_CONCURRENCY_LIMIT; LOG_WARN("rebuild count exceed the limit", K(ret), K(stat), K(data_copy_in_limit)); } } else if (BACKUP_REPLICA_OP == type) { data_copy_in_limit = OB_GROUP_BACKUP_CONCURRENCY; if (stat.backup_count_ >= data_copy_in_limit) { ret = OB_REACH_SERVER_DATA_COPY_IN_CONCURRENCY_LIMIT; LOG_WARN("backup count exceed the limit", K(ret), K(stat), K(data_copy_in_limit)); } } else if (VALIDATE_BACKUP_OP == type) { data_copy_in_limit = OB_GROUP_VALIDATE_CONCURRENCY; if (stat.validate_count_ >= data_copy_in_limit) { ret = OB_REACH_SERVER_DATA_COPY_IN_CONCURRENCY_LIMIT; LOG_WARN("validate count exceed the limit", K(ret), K(stat), K(data_copy_in_limit)); } } else if (BACKUP_BACKUPSET_OP == type) { data_copy_in_limit = OB_GROUP_BACKUP_CONCURRENCY; if (stat.backup_backupset_count_ >= data_copy_in_limit) { ret = OB_REACH_SERVER_DATA_COPY_IN_CONCURRENCY_LIMIT; LOG_WARN( "backup backupset count exceed the limit", K(ret), K(stat.backup_backupset_count_), K(data_copy_in_limit)); } } else if (BACKUP_ARCHIVELOG_OP == type) { data_copy_in_limit = OB_GROUP_BACKUP_CONCURRENCY; if (stat.backup_archivelog_count_ >= data_copy_in_limit) { ret = OB_REACH_SERVER_DATA_COPY_IN_CONCURRENCY_LIMIT; LOG_WARN( "backup backupset count exceed the limit", K(ret), K(stat.backup_archivelog_count_), K(data_copy_in_limit)); } } else { if (stat.normal_migrate_count_ >= data_copy_in_limit) { ret = OB_REACH_SERVER_DATA_COPY_IN_CONCURRENCY_LIMIT; LOG_WARN("migrator count exceed the limit", K(ret), K(stat), K(data_copy_in_limit)); } } return ret; } int ObPartGroupMigrator::inner_schedule(const ObIArray& arg_list, const bool is_batch_mode, const share::ObTaskId& in_task_id, const bool is_normal_migrate, ObPartGroupTask*& out_group_task) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObPartGroupTask* group_task = NULL; share::ObTaskId task_id = in_task_id; // copy it DEBUG_SYNC(BEFORE_OBSERVER_SCHEDULE_MIGRATE); common::SpinWLockGuard guard(update_task_list_lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(check_arg_list(arg_list))) { STORAGE_LOG(WARN, "arg_list is empty", K(ret), K(arg_list)); } else if (OB_FAIL(check_copy_limit_(arg_list))) { LOG_WARN("failed to check_copy_limit", K(ret)); } else if (OB_FAIL(get_group_task(arg_list, is_batch_mode, task_id, group_task))) { STORAGE_LOG(WARN, "failed to get group task", K(ret), K(arg_list)); } else if (OB_FAIL(check_dup_task(*group_task))) { STORAGE_LOG(WARN, "failed to check new task", K(ret), K(group_task)); } else if (OB_FAIL(task_list_.push_back(group_task))) { STORAGE_LOG(WARN, "failed to add task", K(ret), K(group_task)); } else if (is_normal_migrate && OB_FAIL(schedule_group_migrate_dag(group_task))) { LOG_WARN("failed to schedule_group_migrate_dag", K(ret)); if (OB_SUCCESS != (tmp_ret = task_list_.remove(task_list_.count() - 1))) { LOG_ERROR("failed to rollback failed group task, fatal error", K(ret)); ob_abort(); } } else { STORAGE_LOG(INFO, "succeed to schedule partition group migration task", "task_count", task_list_.count(), KP(group_task), K(*group_task)); out_group_task = group_task; group_task = NULL; } if (NULL != group_task && NULL != cp_fty_) { cp_fty_->free(group_task); } return ret; } // static func int ObPartGroupMigrator::check_arg_list(const ObIArray& arg_list) { int ret = OB_SUCCESS; ObReplicaOpType type = UNKNOWN_REPLICA_OP; const int64_t restore_concurrency = GCONF.restore_concurrency; if (arg_list.count() <= 0) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "arg list must not empty", K(ret), K(arg_list)); } else { type = arg_list.at(0).type_; if (!is_replica_op_valid(type)) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid type", K(ret), K(type), K(arg_list)); } else if (RESTORE_REPLICA_OP == type && restore_concurrency <= 0) { ret = OB_OP_NOT_ALLOW; STORAGE_LOG(WARN, "restore is disabled, cannot restore tenant", K(ret), K(restore_concurrency)); } for (int64_t i = 1; OB_SUCC(ret) && i < arg_list.count(); ++i) { if (type != arg_list.at(i).type_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(ERROR, "op type must be same", K(ret), K(type), K(i), K(arg_list)); } } } return ret; } int ObPartGroupMigrator::check_dup_task(const ObPartGroupTask& task) { // caller must hold update_task_list_lock_ int ret = OB_SUCCESS; ObArray pkey_list; ObReplicaOpType type = task.get_type(); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(task.check_self_dup_task())) { STORAGE_LOG(WARN, "failed to check_self_dup_task", K(ret)); } else if (VALIDATE_BACKUP_OP == type || BACKUP_BACKUPSET_OP == type || BACKUP_ARCHIVELOG_OP == type) { // skip } else if (OB_FAIL(task.get_pkey_list(pkey_list))) { STORAGE_LOG(WARN, "failed to get pkey list", K(ret)); } else if (OB_FAIL(check_dup_task(pkey_list))) { LOG_WARN("failed to check dup task", K(ret)); } return ret; } int ObPartGroupMigrator::check_dup_task(const ObIArray& pkey_list) { // caller must hold update_task_list_lock_ int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { if (OB_FAIL(task_list_[i]->check_dup_task(pkey_list))) { STORAGE_LOG(WARN, "failed to check_dup_task in task_list", K(ret), K(i)); } } return ret; } int ObPartGroupMigrator::schedule_group_migrate_dag(ObPartGroupTask*& group_task) { int ret = OB_SUCCESS; ObGroupMigrateDag* dag = NULL; ObGroupMigrateExecuteTask* task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_ISNULL(group_task)) { ret = OB_INVALID_ARGUMENT; LOG_ERROR("group_task must not null", K(ret), KP(group_task)); } else if (OB_FAIL(ObDagScheduler::get_instance().alloc_dag(dag))) { LOG_WARN("failed to alloc group migrate dag", K(ret)); } else if (OB_FAIL(dag->init(group_task, partition_service_))) { LOG_WARN("failed to init dag", K(ret)); } else if (OB_FAIL(dag->alloc_task(task))) { LOG_WARN("failed to alloc task, ", K(ret)); } else if (OB_FAIL(task->init(group_task))) { LOG_WARN("failed to init task", K(ret)); } else if (OB_FAIL(dag->add_task(*task))) { LOG_WARN("Failed to add task", K(ret)); } else if (OB_FAIL(ObDagScheduler::get_instance().add_dag(dag))) { if (OB_SIZE_OVERFLOW != ret && OB_EAGAIN != ret) { LOG_WARN("Fail to add group migrate dag", K(ret)); } } if (OB_FAIL(ret)) { if (NULL != dag) { dag->clear(); ObDagScheduler::get_instance().free_dag(*dag); } } return ret; } int ObPartGroupMigrator::remove_finish_task(ObPartGroupTask* group_task) { int ret = OB_SUCCESS; bool found = false; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_ISNULL(group_task)) { ret = OB_INVALID_ARGUMENT; LOG_ERROR("group_task must not null", K(ret), KP(group_task)); } else { { common::SpinWLockGuard guard(update_task_list_lock_); for (int64_t i = task_list_.count() - 1; OB_SUCC(ret) && !found && i >= 0; --i) { ObPartGroupTask* task = task_list_[i]; ; if (task == group_task) { found = true; if (OB_FAIL(task_list_.remove(i))) { STORAGE_LOG(ERROR, "failed to remove finished part group migration task", K(ret), K(i), K(task_list_)); } } } } if (OB_FAIL(ret)) { // do nothing } else if (!found) { ret = OB_ERR_SYS; LOG_ERROR("cannot find need remove finish task", K(ret), KP(group_task), K(*group_task)); } else { STORAGE_LOG( INFO, "remove finished part group migration task", K(task_list_.count()), KP(group_task), K(*group_task)); cp_fty_->free(group_task); } } return ret; } void ObPartGroupMigrator::wakeup() { common::SpinRLockGuard guard(update_task_list_lock_); for (int64_t i = 0; i < task_list_.count(); ++i) { ObPartGroupTask* task = task_list_[i]; if (NULL != task) { task->wakeup(); } } } int ObPartGroupMigrator::get_group_task(const ObIArray& arg_list, const bool is_batch_mode, const share::ObTaskId& in_task_id, ObPartGroupTask*& group_task) { int ret = OB_SUCCESS; group_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (arg_list.empty()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "replica op arg list should not be empty", K(ret), K(in_task_id), K(arg_list)); } else { const ObReplicaOpType type = arg_list.at(0).type_; if (BACKUP_REPLICA_OP == type) { ObPartGroupBackupTask* backup_group_task = NULL; if (NULL == (backup_group_task = cp_fty_->get_part_group_backup_task())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to alloc group task", K(ret)); } else if (FALSE_IT(group_task = backup_group_task)) { } else if (OB_FAIL(backup_group_task->init(arg_list, is_batch_mode, partition_service_, in_task_id))) { STORAGE_LOG(WARN, "failed to init group task", K(ret)); } } else { ObPartGroupMigrationTask* migration_group_task = NULL; if (NULL == (migration_group_task = cp_fty_->get_part_group_migration_task())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to alloc group task", K(ret)); } else if (FALSE_IT(group_task = migration_group_task)) { } else if (OB_FAIL(migration_group_task->init(arg_list, is_batch_mode, partition_service_, in_task_id))) { STORAGE_LOG(WARN, "failed to init group task", K(ret)); } } } return ret; } ObPartGroupTask::ObPartGroupTask() : is_inited_(false), is_batch_mode_(true), lock_(ObLatchIds::GROUP_MIGRATE_TASK_LOCK), task_list_(), partition_service_(NULL), first_error_code_(OB_SUCCESS), type_(UNKNOWN_REPLICA_OP), task_id_(), is_finished_(false), retry_job_start_ts_(0), tenant_id_(0), cond_(), need_idle_(true), report_list_(), change_member_option_(NORMAL_CHANGE_MEMBER_LIST) {} ObPartGroupTask::~ObPartGroupTask() {} bool ObPartGroupTask::is_finish() const { common::SpinRLockGuard guard(lock_); return is_finished_; } int ObPartGroupTask::get_pkey_list(ObIArray& pkey_list) const { int ret = OB_SUCCESS; common::SpinRLockGuard guard(lock_); pkey_list.reset(); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { if (OB_FAIL(pkey_list.push_back(task_list_.at(i).arg_.key_))) { STORAGE_LOG(WARN, "failed to add pkey", K(ret)); } } } return ret; } int ObPartGroupTask::check_dup_task(const ObIArray& pkey_list) const { // Check duplicate task just use pkey, so it do not has concurrency. No need lock int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (ObReplicaOpType::VALIDATE_BACKUP_OP == get_type() || ObReplicaOpType::BACKUP_BACKUPSET_OP == get_type() || ObReplicaOpType::BACKUP_ARCHIVELOG_OP == get_type()) { // skip } else { for (int64_t task_idx = 0; OB_SUCC(ret) && task_idx < task_list_.count(); ++task_idx) { for (int64_t pkey_idx = 0; OB_SUCC(ret) && pkey_idx < pkey_list.count(); ++pkey_idx) { if (task_list_[task_idx].arg_.key_ == pkey_list.at(pkey_idx)) { ret = OB_TASK_EXIST; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(WARN, "migration task is dup", K(ret), K_(task_id), K_(type), "pkey", pkey_list.at(pkey_idx), K(task_list_[task_idx])); } } } } return ret; } int ObPartGroupTask::check_self_dup_task() const { int ret = OB_SUCCESS; // Check duplicate task just use pkey, so it do not has concurrency. No need lock if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { for (int64_t j = i + 1; OB_SUCC(ret) && j < task_list_.count(); ++j) { if (task_list_[i].arg_.type_ == VALIDATE_BACKUP_OP && task_list_[j].arg_.type_ == VALIDATE_BACKUP_OP) { if (task_list_[i].arg_.validate_arg_.pg_key_ == task_list_[j].arg_.validate_arg_.pg_key_ && task_list_[i].arg_.validate_arg_.backup_set_id_ == task_list_[j].arg_.validate_arg_.backup_set_id_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "task is dup", K(ret), K(i), K(j), K(task_list_[i]), K(task_list_[j])); } } else if (task_list_[i].arg_.type_ == BACKUP_BACKUPSET_OP && task_list_[j].arg_.type_ == BACKUP_BACKUPSET_OP) { if (task_list_[i].arg_.backup_backupset_arg_.pg_key_ == task_list_[j].arg_.backup_backupset_arg_.pg_key_ && task_list_[i].arg_.backup_backupset_arg_.backup_set_id_ == task_list_[j].arg_.backup_backupset_arg_.backup_set_id_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "task is dup", K(ret), K(i), K(j), K(task_list_[i].arg_.backup_backupset_arg_), K(task_list_[j].arg_.backup_backupset_arg_)); } } else if (task_list_[i].arg_.type_ == BACKUP_ARCHIVELOG_OP && task_list_[j].arg_.type_ == BACKUP_ARCHIVELOG_OP) { if (task_list_[i].arg_.backup_archive_log_arg_.pg_key_ == task_list_[j].arg_.backup_archive_log_arg_.pg_key_ && task_list_[i].arg_.backup_archive_log_arg_.piece_id_ == task_list_[j].arg_.backup_archive_log_arg_.piece_id_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "task is dup", K(ret), K(i), K(j), K(task_list_[i].arg_.backup_archive_log_arg_), K(task_list_[j].arg_.backup_archive_log_arg_)); } } else if (task_list_[i].arg_.key_ == task_list_[j].arg_.key_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "task is dup", K(ret), K(i), K(j), K(task_list_[i]), K(task_list_[j])); } } } } return ret; } int ObPartGroupTask::set_part_task_start(const ObPartitionKey& pkey, const int64_t backup_set_id) { int ret = OB_SUCCESS; ObPartMigrationTask* task = NULL; { common::SpinWLockGuard guard(lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { for (int64_t i = 0; NULL == task && i < task_list_.count(); ++i) { if (VALIDATE_BACKUP_OP == get_type()) { if (task_list_[i].arg_.validate_arg_.pg_key_ == pkey && task_list_[i].arg_.validate_arg_.backup_set_id_ == backup_set_id) { task = &task_list_[i]; } } else if (BACKUP_BACKUPSET_OP == get_type()) { if (task_list_[i].arg_.backup_backupset_arg_.pg_key_ == pkey && task_list_[i].arg_.backup_backupset_arg_.backup_set_id_ == backup_set_id) { task = &task_list_[i]; } } else { if (task_list_[i].arg_.key_ == pkey) { task = &task_list_[i]; } } } if (NULL == task) { ret = OB_ENTRY_NOT_EXIST; STORAGE_LOG(ERROR, "cannot found pkey in task", K(ret), K(pkey), KP(this), K(*this)); } else if (ObPartMigrationTask::INIT == task->status_) { task->status_ = ObPartMigrationTask::DOING; STORAGE_LOG(INFO, "set_part_task_start", K(pkey), K(ret), K(*this)); dump_task_status(); } } } return ret; } int ObPartGroupTask::set_part_task_finish(const ObPartitionKey& pkey, const int32_t result, const share::ObTaskId& part_task_id, const bool during_migrating, const int64_t backup_set_id) { int ret = OB_SUCCESS; ObPartMigrationTask* task = NULL; { common::SpinWLockGuard guard(lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { for (int64_t i = 0; NULL == task && i < task_list_.count(); ++i) { if (VALIDATE_BACKUP_OP == get_type()) { if (task_list_[i].arg_.validate_arg_.pg_key_ == pkey && task_list_[i].arg_.validate_arg_.backup_set_id_ == backup_set_id) { task = &task_list_[i]; } } else if (BACKUP_BACKUPSET_OP == get_type()) { if (task_list_[i].arg_.backup_backupset_arg_.pg_key_ == pkey && task_list_[i].arg_.backup_backupset_arg_.backup_set_id_ == backup_set_id) { task = &task_list_[i]; } } else { if (task_list_[i].arg_.key_ == pkey) { task = &task_list_[i]; } } } if (NULL == task) { ret = OB_ENTRY_NOT_EXIST; STORAGE_LOG(ERROR, "cannot found pkey in task", K(ret), K(pkey), K(result), KP(this), K(*this)); } else { task->status_ = ObPartMigrationTask::FINISH; task->result_ = result; task->during_migrating_ = during_migrating; } } if (OB_SUCCESS == first_error_code_) { if (OB_SUCCESS != result) { if (BACKUP_REPLICA_OP == get_type() && OB_PG_IS_REMOVED == result) { // backup while partition may migrate out, skip set first error code // do nothing } else { first_error_code_ = result; } STORAGE_LOG(WARN, "record first_error_code_ with partition task", K(result), K(pkey), K(ret)); } else if (OB_SUCCESS != ret) { first_error_code_ = ret; STORAGE_LOG(WARN, "record first_error_code_", K(result), K(pkey), K(ret)); } } STORAGE_LOG(INFO, "set_part_task_finish", K(pkey), K(result), K(during_migrating), K(ret), K(part_task_id), K(*this), K(backup_set_id)); dump_task_status(); } ObPartGroupMigrator::get_instance().wakeup(); return ret; } int ObPartGroupTask::build_migrate_ctx(const ObReplicaOpArg& arg, ObMigrateCtx& migrate_ctx) { int ret = OB_SUCCESS; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "ObPartGroupMigrationTask has not been inited", K(ret)); } else { migrate_ctx.replica_op_arg_ = arg; migrate_ctx.action_ = ObMigrateCtx::INIT; migrate_ctx.result_ = OB_SUCCESS; migrate_ctx.doing_task_cnt_ = 1; migrate_ctx.total_task_cnt_ = 1; migrate_ctx.need_rebuild_ = false; if (0 == migrate_ctx.create_ts_) { migrate_ctx.create_ts_ = ObTimeUtility::current_time(); } migrate_ctx.task_id_.reset(); // filled in add_migrate_ctx_without_lock migrate_ctx.continue_fail_count_ = 0; migrate_ctx.rebuild_count_ = migrate_ctx.rebuild_count_ < 0 ? 0 : migrate_ctx.rebuild_count_; migrate_ctx.group_task_ = this; migrate_ctx.group_task_id_ = get_task_id(); migrate_ctx.part_ctx_array_.reset(); migrate_ctx.curr_major_block_array_index_ = (migrate_ctx.curr_major_block_array_index_ + 1) % 2; migrate_ctx.major_block_id_array_ptr_ = &migrate_ctx.major_block_id_array_[migrate_ctx.curr_major_block_array_index_]; migrate_ctx.is_copy_cover_minor_ = false; if (migrate_ctx.is_only_copy_sstable()) { // do nothing } else if (OB_FAIL(migrate_ctx.recovery_point_ctx_.init(migrate_ctx))) { STORAGE_LOG(WARN, "failed to init reocovery point ctx", K(ret), K(arg)); } } return ret; } void ObPartGroupTask::dump_task_status() { // caller must hold lock int tmp_ret = OB_SUCCESS; int64_t count[ObPartMigrationTask::MAX]; memset(count, 0, sizeof(count)); if (!is_inited_) { tmp_ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(tmp_ret)); } else { for (int64_t i = 0; OB_SUCCESS == tmp_ret && i < task_list_.count(); ++i) { ObPartMigrationTask& task = task_list_[i]; if (task.status_ < 0 || task.status_ >= ObPartMigrationTask::MAX) { tmp_ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "invalid task stauts", K(tmp_ret), K(i), K(task)); } else { count[task.status_]++; } } STORAGE_LOG(INFO, "dump group migration task status", "total_count", task_list_.count(), "INIT", count[ObPartMigrationTask::INIT], "DOING", count[ObPartMigrationTask::DOING], "FINISH", count[ObPartMigrationTask::FINISH]); } } int ObPartGroupTask::check_is_task_cancel() { int ret = OB_SUCCESS; bool is_cancel = false; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (OB_SUCCESS == first_error_code_) { common::SpinWLockGuard guard(lock_); if (OB_FAIL(SYS_TASK_STATUS_MGR.is_task_cancel(task_id_, is_cancel))) { STORAGE_LOG(ERROR, "failed to check is group task cancel", K(ret), K(*this)); } else if (is_cancel) { first_error_code_ = OB_CANCELED; STORAGE_LOG(WARN, "group part migration is cancelled", K(ret), K(*this)); } else if (ObPartitionMigrator::get_instance().is_stop()) { first_error_code_ = OB_CANCELED; STORAGE_LOG(WARN, "mirgator is stop, set group task is cancelled", K(ret), K(*this)); } else if (ObReplicaOpType::VALIDATE_BACKUP_OP == get_type() || ObReplicaOpType::BACKUP_BACKUPSET_OP == get_type() || ObReplicaOpType::BACKUP_ARCHIVELOG_OP == get_type()) { // do nothing } else if (!GCTX.omt_->has_tenant(tenant_id_)) { first_error_code_ = OB_TENANT_NOT_EXIST; STORAGE_LOG(WARN, "tenant not exists, set migration failed", K(ret), K(*this)); } } if (OB_SUCC(ret) && OB_SUCCESS == first_error_code_) { // TODO(jiage): remove it after fix GCTX.omt_->has_tenant bool is_tenant_exist = false; if (ObReplicaOpType::VALIDATE_BACKUP_OP == get_type() || ObReplicaOpType::BACKUP_BACKUPSET_OP == get_type() || ObReplicaOpType::BACKUP_ARCHIVELOG_OP == get_type()) { // do nothing } else if (OB_FAIL(check_is_tenant_exist(is_tenant_exist))) { LOG_WARN("failed to check is tenant exist", K(ret)); } else if (!is_tenant_exist) { common::SpinWLockGuard guard(lock_); first_error_code_ = OB_TENANT_NOT_EXIST; STORAGE_LOG(WARN, "tenant schema not exists, set migration failed", K(ret), K(*this)); } } return ret; } int ObPartGroupTask::check_partition_checksum(const common::ObPartitionKey& pkey) { int ret = OB_SUCCESS; ModulePageAllocator allocator(ObNewModIds::OB_PARTITION_MIGRATE); share::ObPartitionInfo partition_info; partition_info.set_allocator(&allocator); share::ObUnitedPtOperator united_operator; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "The migrator has not inited or stopped.", K(ret)); } else if (OB_ISNULL(GCTX.pt_operator_) || OB_ISNULL(GCTX.remote_pt_operator_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "pt_operator must not NULL.", K(ret)); } else if (OB_FAIL(united_operator.init(GCTX.pt_operator_, GCTX.remote_pt_operator_))) { LOG_WARN("fail to init united_operator", KR(ret)); } else if (OB_SUCCESS != (ret = united_operator.united_get(pkey.get_table_id(), pkey.get_partition_id(), partition_info))) { STORAGE_LOG(WARN, "fail to get partition info.", K(ret), K(pkey)); } else if (partition_info.replica_count() > 1) { const common::ObIArray& replicas = partition_info.get_replicas_v2(); int64_t lastIdx = -1; for (int64_t i = 0; OB_SUCC(ret) && i < replicas.count(); ++i) { if (replicas.at(i).replica_type_ == REPLICA_TYPE_LOGONLY || replicas.at(i).is_restore_ != 0 || (lastIdx >= 0 && replicas.at(lastIdx).is_restore_ != 0)) { continue; } else if (lastIdx == -1) { lastIdx = i; } else if (replicas.at(i).data_version_ > replicas.at(lastIdx).data_version_) { lastIdx = i; } else if (replicas.at(i).data_version_ == replicas.at(lastIdx).data_version_) { if (replicas.at(i).data_checksum_ != replicas.at(lastIdx).data_checksum_) { ret = OB_CHECKSUM_ERROR; STORAGE_LOG(ERROR, "replicas check sum is not match", K(pkey), "replica1", replicas.at(i), "replica2", replicas.at(lastIdx)); } } } } if (OB_SUCCESS != ret && OB_CHECKSUM_ERROR != ret) { STORAGE_LOG(WARN, "failed to check checksum for partition, overwrite ret", K(pkey), K(ret)); ret = OB_SUCCESS; } return ret; } int ObPartGroupTask::check_is_tenant_exist(bool& is_exist) { int ret = OB_SUCCESS; share::schema::ObSchemaGetterGuard schema_guard; share::schema::ObMultiVersionSchemaService& schema_service = share::schema::ObMultiVersionSchemaService::get_instance(); is_exist = true; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (OB_FAIL(schema_service.get_tenant_full_schema_guard(OB_SYS_TENANT_ID, schema_guard))) { LOG_WARN("failed to get schema guard", K(ret)); } else if (OB_FAIL(schema_guard.check_tenant_exist(tenant_id_, is_exist))) { LOG_WARN("failed to check tenant exist", K(ret), K_(tenant_id)); } return ret; } int ObPartGroupTask::check_is_partition_exist(const common::ObPartitionKey& pkey, bool& is_exist) { int ret = OB_SUCCESS; share::schema::ObSchemaGetterGuard schema_guard; share::schema::ObMultiVersionSchemaService& schema_service = share::schema::ObMultiVersionSchemaService::get_instance(); const uint64_t fetch_tenant_id = is_inner_table(pkey.get_table_id()) ? OB_SYS_TENANT_ID : pkey.get_tenant_id(); bool check_dropped_partition = true; is_exist = false; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (OB_FAIL(schema_service.get_tenant_full_schema_guard(fetch_tenant_id, schema_guard))) { LOG_WARN("failed to get schema guard", K(ret)); } else if (OB_FAIL(schema_guard.check_partition_exist( pkey.get_table_id(), pkey.get_partition_id(), check_dropped_partition, is_exist))) { LOG_WARN("failed to check partition exist", K(ret), K(pkey)); } return ret; } int ObPartGroupTask::check_can_as_data_source( const ObReplicaOpType& op, const ObReplicaType& src_type, const ObReplicaType& dst_type, bool& as_data_source) { int ret = OB_SUCCESS; as_data_source = false; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "ObPartGroupMigrationTask has not been inited", K(ret)); } else { if (COPY_GLOBAL_INDEX_OP == op || COPY_LOCAL_INDEX_OP == op) { as_data_source = ObReplicaTypeCheck::is_replica_with_ssstore(src_type) && ObReplicaTypeCheck::is_replica_with_ssstore(dst_type); } else { as_data_source = ObReplicaTypeCheck::can_as_data_source(dst_type, src_type); } } return ret; } int ObPartGroupTask::build_failed_report_list( const int32_t first_fail_code, ObIArray& report_list) { // static func int ret = OB_SUCCESS; if (OB_SUCCESS == first_fail_code) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(first_fail_code)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { report_list.at(i).result_ = first_fail_code; } } return ret; } int ObPartGroupTask::set_report_list_result( ObIArray& report_list, const ObPartitionKey& pkey, const int32_t result) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_SUCCESS == result) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "only failed result need set", K(ret), K(pkey), K(result)); } else { ret = OB_ENTRY_NOT_EXIST; for (int64_t i = 0; OB_ENTRY_NOT_EXIST == ret && i < report_list.count(); ++i) { if (report_list.at(i).arg_.key_ == pkey) { report_list.at(i).result_ = result; STORAGE_LOG(INFO, "set report list result", K(i), K(pkey), K(result)); ret = OB_SUCCESS; } } } return ret; } int ObPartGroupTask::fill_report_list(bool& is_batch_mode, ObIArray& report_list) { int ret = OB_SUCCESS; common::SpinRLockGuard guard(lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!is_finished_) { ret = OB_ERR_SYS; LOG_ERROR("part group migration task not finished, cannot fill report list", K(ret), K(*this)); } else if (OB_FAIL(report_list.assign(report_list_))) { LOG_WARN("failed to assign report list", K(ret)); } else { is_batch_mode = is_batch_mode_; } return ret; } bool ObPartGroupTask::has_error() const { return OB_SUCCESS != first_error_code_; } void ObPartGroupTask::wakeup() { ObThreadCondGuard guard(cond_); need_idle_ = false; cond_.broadcast(); } ObPartGroupMigrationTask::ObPartGroupMigrationTask() : ObPartGroupTask(), schedule_ts_(0), start_change_member_ts_(0), meta_indexs_(), meta_index_store_(), restore_version_(INT64_MAX), skip_change_member_list_(false) {} ObPartGroupMigrationTask::~ObPartGroupMigrationTask() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (is_inited_) { int64_t timestamp = ObTimeUtility::current_time(); common::SpinWLockGuard guard(lock_); bool is_all_finish = false; while (!is_all_finish) { if (ObPartGroupMigrator::get_instance().is_stop()) { STORAGE_LOG(WARN, "mirgator is stop, set group task is cancelled", K(ret), K(*this)); break; } else { // After the dfault loop is over, it should all be over is_all_finish = true; for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { if (task_list_[i].need_reset_migrate_status_) { DEBUG_SYNC(BEFORE_CLEAR_MIGRATE_STATUS); bool need_retry = false; ObIPartitionGroup* partition = NULL; ObPGStorage* pg_storage = NULL; ObMigrateStatus cur_migrate_status = OB_MIGRATE_STATUS_NONE; ObMigrateStatus new_migrate_status = OB_MIGRATE_STATUS_NONE; if (OB_ISNULL(partition = task_list_[i].ctx_.get_partition())) { tmp_ret = OB_ERR_SYS; LOG_ERROR("partition/partition storage must not null", K(tmp_ret)); } else if (OB_ISNULL(pg_storage = &(partition->get_pg_storage()))) { tmp_ret = OB_ERR_SYS; LOG_ERROR("pg_storage must not null", K(tmp_ret), KP(pg_storage)); } else if (OB_SUCCESS != (tmp_ret = pg_storage->get_pg_migrate_status(cur_migrate_status))) { LOG_WARN("failed to get migrate status", K(tmp_ret), "pkey", task_list_[i].arg_.key_); } else if (OB_SUCCESS == task_list_[i].result_) { new_migrate_status = OB_MIGRATE_STATUS_NONE; LOG_INFO( "migrate success, set migrate status none", "pkey", task_list_[i].arg_.key_, K(cur_migrate_status)); } else if (!task_list_[i].during_migrating_) { new_migrate_status = OB_MIGRATE_STATUS_NONE; LOG_INFO("migrate fail not during migrting, set migrate status none", "pkey", task_list_[i].arg_.key_, K(cur_migrate_status)); } else if (FAST_MIGRATE_REPLICA_OP == task_list_[i].arg_.type_ && task_list_[i].ctx_.is_takeover_finished_) { new_migrate_status = OB_MIGRATE_STATUS_NONE; LOG_INFO("fast migrate: set status normal when takeover finished", K(ret), K(cur_migrate_status), K(task_list_[i].arg_.key_)); } else { if (OB_SUCCESS != (tmp_ret = ObMigrateStatusHelper::trans_fail_status(cur_migrate_status, new_migrate_status))) { LOG_WARN("failed to trans_migrate_fail_status", K(tmp_ret), "pkey", task_list_[i].arg_.key_, K(cur_migrate_status)); } } if (OB_SUCCESS == tmp_ret) { if (OB_SUCCESS != (tmp_ret = pg_storage->set_pg_migrate_status(new_migrate_status, timestamp))) { if (OB_PARTITION_IS_REMOVED == tmp_ret) { LOG_WARN("partition is removed, no need to set migrate status", "pkey", task_list_[i].arg_.key_, K(tmp_ret)); } else { LOG_ERROR("failed to set migrate status, will retry next loop", "pkey", task_list_[i].arg_.key_, K(tmp_ret)); need_retry = true; } } else { LOG_INFO( "succ to reset migrate status", "pkey", task_list_[i].arg_.key_, K(new_migrate_status), K(tmp_ret)); } } if (!need_retry) { task_list_[i].need_reset_migrate_status_ = false; } else { // If there are tasks that need to be retried, not all tasks are considered to has been completed is_all_finish = false; } } // end of if (task_list_[i].need_reset_migrate_status_) } // end of for loop } } // end of while loop is_inited_ = false; } } int ObPartGroupMigrationTask::init(const ObIArray& task_list, const bool is_batch_mode, storage::ObPartitionService* partition_service, const share::ObTaskId& task_id) { int ret = OB_SUCCESS; SMART_VAR(ObPartMigrationTask, tmp_task) { ObChangeMemberOption change_member_option = NORMAL_CHANGE_MEMBER_LIST; common::SpinWLockGuard guard(lock_); if (is_inited_) { ret = OB_INIT_TWICE; STORAGE_LOG(WARN, "cannot init twice", K(ret)); } else if (NULL == partition_service || task_list.empty() || task_id.is_invalid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(is_batch_mode), KP(partition_service), K(task_id), K(task_list)); } else if (OB_FAIL(cond_.init(ObWaitEventIds::GROUP_MIGRATE_TASK_IDLE_WAIT))) { STORAGE_LOG(WARN, "failed to init cond", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < task_list.count(); ++i) { tmp_task.arg_ = task_list.at(i); tmp_task.arg_.cluster_id_ = (tmp_task.arg_.cluster_id_ == OB_INVALID_CLUSTER_ID) ? GCONF.cluster_id : tmp_task.arg_.cluster_id_; tmp_task.status_ = ObPartMigrationTask::INIT; tmp_task.result_ = OB_SUCCESS; if (OB_FAIL(task_list_.push_back(tmp_task))) { STORAGE_LOG(WARN, "failed to add task list", K(ret)); } if (OB_SUCC(ret)) { if (0 == i) { type_ = task_list.at(i).type_; tenant_id_ = task_list.at(i).key_.get_tenant_id(); change_member_option = task_list.at(i).change_member_option_; restore_version_ = task_list.at(i).restore_version_; if (!is_replica_op_valid(type_)) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "invalid type", K(ret), K(i), K(type_), K(task_list)); } else if (SKIP_CHANGE_MEMBER_LIST == change_member_option && ADD_REPLICA_OP != type_ && REBUILD_REPLICA_OP != type_ && MIGRATE_REPLICA_OP != type_ && FAST_MIGRATE_REPLICA_OP != type_ && CHANGE_REPLICA_OP != type_ && LINK_SHARE_MAJOR_OP != type_ && RESTORE_STANDBY_OP != type_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "only add or migrate or change replica can skip change member list", K(ret), K(type_), K(task_list)); } } else if (type_ != task_list.at(i).type_ || tenant_id_ != task_list.at(i).key_.get_tenant_id() || change_member_option != task_list.at(i).change_member_option_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "migration type or tenant_id or skip_change_member_list not same in task list", K(ret), K(i), K(task_list)); } } } if (OB_SUCC(ret)) { is_inited_ = true; is_batch_mode_ = is_batch_mode; schedule_ts_ = ObTimeUtility::current_time(); start_change_member_ts_ = 0; partition_service_ = partition_service; first_error_code_ = OB_SUCCESS; task_id_ = task_id; is_finished_ = false; need_idle_ = true; change_member_option_ = change_member_option; STORAGE_LOG(INFO, "succeed to init group migration task", K(*this)); } } } return ret; } int ObPartGroupMigrationTask::do_task() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; const int64_t start_ts = ObTimeUtility::current_time(); ObTaskController::get().switch_task(share::ObTaskType::DATA_MAINTAIN); ObCurTraceId::set(task_id_); ObTaskController::get().allow_next_syslog(); LOG_INFO("start group migration task", K(*this)); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { if (OB_FAIL(check_before_do_task())) { LOG_WARN("Failed to check_before_do_task", K(ret)); } else if (OB_FAIL(remove_member_list_if_need())) { LOG_WARN("failed to remove_member_list_if_need", K(ret)); } else if (RESTORE_REPLICA_OP == type_) { const bool need_check_compeleted = true; if (OB_UNLIKELY(0 == task_list_.count())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("restore task_list_ is empty", K(ret), K(type_), K(task_list_.count())); } else if (task_list_[0].arg_.is_physical_restore_leader()) { const share::ObPhysicalRestoreArg& restore_arg = task_list_[0].arg_.phy_restore_arg_; if (OB_FAIL(init_restore_meta_index_(restore_arg))) { LOG_WARN("failed to init meta index", K(ret), K(restore_arg)); } } } else if (VALIDATE_BACKUP_OP == type_) { const share::ObPhysicalValidateArg& validate_arg = task_list_[0].arg_.validate_arg_; ObBackupBaseDataPathInfo path_info; if (!validate_arg.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("validate arg is invalid, please check", K(ret), K(validate_arg)); } else if (OB_FAIL(validate_arg.get_backup_base_data_info(path_info))) { LOG_WARN("failed to get backup base data info", K(ret), K(validate_arg)); } else if (OB_FAIL(meta_index_store_.init(path_info))) { LOG_WARN("init validate meta_index_store failed", K(ret), K(path_info)); } } else { // do nothing } if (OB_FAIL(ret)) { common::SpinWLockGuard guard(lock_); if (OB_SUCCESS == first_error_code_) { first_error_code_ = ret; STORAGE_LOG(WARN, "set first_error_code_", K(ret)); } } while (!is_finished_) { if (OB_SUCCESS != (tmp_ret = check_is_task_cancel())) { STORAGE_LOG(WARN, "failed to check is task cancel", K(tmp_ret)); } if (OB_SUCCESS != (tmp_ret = try_schedule_new_partition_migration())) { STORAGE_LOG(WARN, "failed to try_schedule_new_partition_migration", K(tmp_ret)); } if (OB_SUCCESS != (tmp_ret = try_finish_group_migration())) { STORAGE_LOG(WARN, "failed to try_finish_group_migration", K(tmp_ret)); } if (!is_finished_) { share::dag_yield(); ObThreadCondGuard guard(cond_); if (need_idle_) { if (OB_SUCCESS != (tmp_ret = cond_.wait(PART_GROUP_TASK_IDLE_TIME_MS))) { if (OB_TIMEOUT != tmp_ret) { STORAGE_LOG(WARN, "failed to idle", K(tmp_ret)); } } } need_idle_ = true; } } } ObTaskController::get().allow_next_syslog(); const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; LOG_INFO("group migrate task finish", K(cost_ts), K(*this)); return ret; } int ObPartGroupMigrationTask::remove_member_list_if_need() { int ret = OB_SUCCESS; ObAddr leader_addr; bool need_remove = false; bool need_batch = true; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (CHANGE_REPLICA_OP != type_) { LOG_INFO("no need remove member list"); } else if (OB_FAIL(check_need_batch_remove_member(leader_addr, need_remove, need_batch))) { LOG_WARN("failed to check need batch remove member", K(ret)); } else if (!need_remove) { LOG_INFO("no need remove member"); } else if (need_batch) { hash::ObHashSet removed_pkeys; ObArray report_list; SMART_VAR(ObReportPartMigrationTask, tmp_task) { for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { ObPartMigrationTask& sub_task = task_list_[i]; tmp_task.arg_ = sub_task.arg_; tmp_task.status_ = sub_task.status_; tmp_task.result_ = sub_task.result_; tmp_task.need_report_checksum_ = sub_task.ctx_.need_report_checksum_; tmp_task.ctx_ = &sub_task.ctx_; if (OB_FAIL(report_list.push_back(tmp_task))) { // report_list is reserved before, should not fail here STORAGE_LOG(ERROR, "failed to add report list", K(ret)); } } if (OB_FAIL(try_batch_remove_member(report_list, leader_addr, removed_pkeys))) { LOG_WARN("failed to batch remove member list", K(ret)); } else if ((removed_pkeys.size() != report_list.count()) || (report_list.count() != task_list_.count())) { ret = OB_PARTIAL_FAILED; LOG_WARN("failed to remove member list", K(ret), K(leader_addr), "removed_pkeys", removed_pkeys.size(), "task_count", report_list.count(), K(task_list_.count())); } else { for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { ObPartMigrationTask& sub_task = task_list_[i]; sub_task.result_ = report_list.at(i).result_; } } } } else { if (OB_FAIL(try_single_remove_member())) { LOG_WARN("failed to single remove member list", K(ret)); } } if (OB_FAIL(ret)) { first_error_code_ = ret; } return ret; } int ObPartGroupMigrationTask::check_need_batch_remove_member(ObAddr& leader_addr, bool& need_remove, bool& need_batch) { int ret = OB_SUCCESS; ObMigrateSrcInfo tmp_src_info; const int64_t start_ts = ObTimeUtility::current_time(); leader_addr.reset(); need_remove = false; need_batch = true; const bool is_primary_cluster = GCTX.is_primary_cluster(); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (task_list_.count() <= 0) { ret = OB_ERR_SYS; LOG_ERROR("no task need migrate", K(ret), K(task_list_)); } else if (NORMAL_CHANGE_MEMBER_LIST != task_list_.at(0).arg_.change_member_option_) { LOG_INFO("task skip change member list, skip it", K(task_list_.at(0))); } else { ObReplicaType src_type = task_list_.at(0).arg_.src_.get_replica_type(); ObReplicaType dest_type = task_list_.at(0).arg_.dst_.get_replica_type(); const ObAddr& self_addr = partition_service_->get_self_addr(); // For batch operations of this type, rs guarantees src_type and dest_type must be consistent with the local, // or an error will be report directly for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { ObIPartitionGroupGuard guard; ObPartMigrationTask& task = task_list_.at(i); if (OB_FAIL(ObPartitionService::get_instance().get_partition(task.arg_.key_, guard))) { LOG_WARN("failed to get partition guard", K(ret), "arg", task.arg_); } else if (OB_ISNULL(guard.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret), "arg", task.arg_); } else if (guard.get_partition_group()->get_replica_type() != src_type || src_type != task.arg_.src_.get_replica_type() || dest_type != task.arg_.dst_.get_replica_type() || self_addr != task.arg_.src_.get_server() || self_addr != task.arg_.dst_.get_server()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("replica type not match", K(ret), K(self_addr), K(src_type), K(dest_type), "local_replica_type", guard.get_partition_group()->get_replica_type(), "arg", task.arg_); } } if (OB_SUCC(ret)) { if (src_type == dest_type) { ret = OB_ALREADY_DONE; STORAGE_LOG(WARN, "no change of replica type", K(ret), K(src_type), K(dest_type)); } else if (!ObReplicaTypeCheck::change_replica_op_allow(src_type, dest_type)) { ret = OB_OP_NOT_ALLOW; STORAGE_LOG(WARN, "change replica op not allow", K(src_type), K(dest_type), K(ret)); } else if (ObReplicaTypeCheck::is_paxos_replica(src_type) && !ObReplicaTypeCheck::is_paxos_replica(dest_type)) { LOG_INFO("need remove member list first"); need_remove = true; for (int64_t i = 0; OB_SUCC(ret) && need_batch && i < task_list_.count(); ++i) { const ObReplicaOpArg& arg = task_list_.at(i).arg_; if (OB_FAIL(ObMigrateGetLeaderUtil::get_leader(arg.key_, tmp_src_info, true /*force_update*/))) { STORAGE_LOG(WARN, "failed to get leader address", K(ret), K(arg.key_)); } else if (!tmp_src_info.is_valid()) { need_batch = false; STORAGE_LOG(WARN, "some partition has no leader, cannot batch change member", K(arg), K(tmp_src_info)); } else if (0 == i) { leader_addr = tmp_src_info.src_addr_; } else if (leader_addr != tmp_src_info.src_addr_) { need_batch = false; STORAGE_LOG(WARN, "partitions have diff leader, cannot batch change member", K(tmp_src_info), K(leader_addr), K(arg)); } } } } } #ifdef ERRSIM if (OB_SUCC(ret) && need_batch) { need_batch = GCONF.allow_batch_remove_member_during_change_replica; if (!need_batch) { LOG_ERROR("fake not need batch remove member list"); } } #endif ObTaskController::get().allow_next_syslog(); const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; STORAGE_LOG(INFO, "check_need_batch_remove_member", K(ret), "count", task_list_.count(), K(need_remove), K(need_batch), K(cost_ts)); return ret; } int ObPartGroupMigrationTask::try_single_remove_member() { int ret = OB_SUCCESS; share::ObTaskId dummy_id; const int64_t start_ts = ObTimeUtility::current_time(); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { dummy_id.init(partition_service_->get_self_addr()); for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { const ObReplicaOpArg& arg = task_list_.at(i).arg_; int64_t dummy_orig_quorum = OB_INVALID_COUNT; // only used in ObPartitionService::batch_remove_replica_mc obrpc::ObMemberChangeArg remove_arg = { arg.key_, arg.src_, false, arg.quorum_, dummy_orig_quorum, WITHOUT_MODIFY_QUORUM, dummy_id}; // mock task id if (OB_FAIL(partition_service_->try_remove_from_member_list(remove_arg))) { STORAGE_LOG(WARN, "remove replica from member list failed", K(remove_arg), K(ret)); } else { STORAGE_LOG(INFO, "remove replica from member list successfully", K(i), K(arg)); } } } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; ObTaskController::get().allow_next_syslog(); LOG_INFO("finish try_single_remove_member", K(ret), "count", task_list_.count(), K(cost_ts)); return ret; } int ObPartGroupMigrationTask::set_create_new_pg(const ObPartitionKey& pg_key) { int ret = OB_SUCCESS; ObPartMigrationTask* task = NULL; common::SpinWLockGuard guard(lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { for (int64_t i = 0; NULL == task && i < task_list_.count(); ++i) { if (task_list_[i].arg_.key_ == pg_key) { task = &task_list_[i]; } } if (OB_ISNULL(task)) { ret = OB_ENTRY_NOT_EXIST; STORAGE_LOG(ERROR, "cannot found pg in task", K(ret), K(pg_key), KP(this), K(*this)); } else { task->need_reset_migrate_status_ = true; } } return ret; } int ObPartGroupMigrationTask::set_migrate_in(ObPGStorage& pg_storage) { int ret = OB_SUCCESS; ObPartMigrationTask* task = NULL; const ObPartitionKey& pkey = pg_storage.get_partition_key(); // TODO () change it to pg key int64_t current_ts = ObTimeUtility::current_time(); ObMigrateStatus old_status = OB_MIGRATE_STATUS_MAX; ObMigrateStatus new_status = OB_MIGRATE_STATUS_MAX; common::SpinWLockGuard guard(lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { for (int64_t i = 0; NULL == task && i < task_list_.count(); ++i) { if (task_list_[i].arg_.key_ == pkey) { task = &task_list_[i]; } } if (NULL == task) { ret = OB_ENTRY_NOT_EXIST; STORAGE_LOG(ERROR, "cannot found pkey in task", K(ret), K(pkey), KP(this), K(*this)); } else if (OB_FAIL(pg_storage.get_pg_migrate_status(old_status))) { LOG_WARN("failed to get old status", K(ret), K(pkey)); } else if (BACKUP_REPLICA_OP == task->arg_.type_ || VALIDATE_BACKUP_OP == task->arg_.type_ || BACKUP_BACKUPSET_OP == task->arg_.type_ || BACKUP_ARCHIVELOG_OP == task->arg_.type_) { LOG_INFO("no need set migrate_status", K(task->arg_.type_), K(pkey)); } else if (OB_FAIL(ObMigrateStatusHelper::trans_replica_op(task->arg_.type_, new_status))) { LOG_WARN("failed to trans_replica_op_to_migrate_status", K(ret), "arg", task->arg_); } else if (OB_MIGRATE_STATUS_NONE != old_status) { ret = OB_STATE_NOT_MATCH; LOG_WARN("old migrate is not finish, cannot set new migrate status", K(ret), K(pkey), K(old_status), "arg", task->arg_); } else { if (OB_FAIL(pg_storage.set_pg_migrate_status(new_status, current_ts))) { STORAGE_LOG(WARN, "failed to set migrate status", K(ret), K(pkey), K(new_status)); } else { task->need_reset_migrate_status_ = true; if (old_status == OB_MIGRATE_STATUS_ADD_FAIL || old_status == OB_MIGRATE_STATUS_MIGRATE_FAIL) { task->during_migrating_ = true; } if (ADD_REPLICA_OP == task->arg_.type_ && pg_storage.is_restore()) { change_member_option_ = SKIP_CHANGE_MEMBER_LIST; STORAGE_LOG(INFO, "add replica op when restore skip set member list", K(pkey)); } } STORAGE_LOG(INFO, "finish set migrate status", K(ret), K(pkey), K(old_status), K(new_status), "during_migrating_flag", task->during_migrating_); } } return ret; } int ObPartGroupMigrationTask::check_before_do_task() { // Scheduling is only necessary before starting execution, without concurrent consideration int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (REMOVE_REPLICA_OP == type_) { STORAGE_LOG(INFO, "no need to check remove replica task", K(*this)); } else if (BACKUP_REPLICA_OP == type_) { if (OB_FAIL(check_before_backup())) { STORAGE_LOG(WARN, "failed to check before backup", K(ret), K(*this)); } } if (OB_FAIL(ret)) { } else if (OB_FAIL(check_partition_validation())) { STORAGE_LOG(WARN, "failed to check_partition_validation", K(ret)); } else if (OB_FAIL(check_disk_space())) { STORAGE_LOG(WARN, "failed to check_disk_space", K(ret)); } if (OB_FAIL(ret)) { ObTaskController::get().allow_next_syslog(); first_error_code_ = ret; STORAGE_LOG(WARN, "failed to check task, mark fail", K(ret), K(*this)); } return ret; } int ObPartGroupMigrationTask::check_partition_validation() { // Scheduling is only necessary before starting execution, without concurrent consideration int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { ObPartMigrationTask& sub_task = task_list_[i]; ObPartitionKey& pkey = sub_task.arg_.key_; bool in_member_list = false; bool is_working_partition = partition_service_->is_working_partition(pkey); if (ObPartGroupMigrator::get_instance().is_stop()) { ret = OB_SERVER_IS_STOPPING; STORAGE_LOG(WARN, "server is stopping", K(ret)); } else if (REMOVE_REPLICA_OP == sub_task.arg_.type_) { ret = OB_ERR_SYS; LOG_ERROR("remove replica cannot schedule migrate", K(ret), "arg", sub_task.arg_); } else if (ADD_REPLICA_OP == type_ || MIGRATE_REPLICA_OP == type_ || FAST_MIGRATE_REPLICA_OP == type_) { if (partition_service_->is_partition_exist(pkey)) { ret = OB_ENTRY_EXIST; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(WARN, "can not add replica which is already exsit", K(ret), K(pkey), K(sub_task)); } } else if (REBUILD_REPLICA_OP == type_) { if (!is_working_partition) { if (OB_FAIL(partition_service_->check_self_in_member_list(pkey, in_member_list))) { STORAGE_LOG(WARN, "failed to check_self_in_member_list", K(ret), K(pkey)); } else if (!in_member_list) { ret = OB_WORKING_PARTITION_NOT_EXIST; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(WARN, "can not rebuild, it is not normal replica", K(ret), K(pkey), K(sub_task)); } else { // Allow partition reconstruction in the member list except for non normal partition ObTaskController::get().allow_next_syslog(); STORAGE_LOG(WARN, "it is not normal partition", K(ret), K(pkey), K(sub_task)); } } } else if (COPY_GLOBAL_INDEX_OP == type_ || COPY_LOCAL_INDEX_OP == type_ || LINK_SHARE_MAJOR_OP == type_) { if (!is_working_partition) { ret = OB_STATE_NOT_MATCH; ObTaskController::get().allow_next_syslog(); LOG_WARN("cannot copy index to not working partition", K(ret)); } } else { // do nothing } } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_CHECK_SUB_MIGRATION_TASK) OB_SUCCESS; if (OB_FAIL(ret)) { STORAGE_LOG(ERROR, "fake EN_CHECK_SUB_MIGRATION_TASK", K(ret)); } } #endif } return ret; } int ObPartGroupMigrationTask::check_disk_space() { // Scheduling is only necessary before starting execution, without concurrent consideration int ret = OB_SUCCESS; int64_t required_size = 0; int64_t partition_required_size = 0; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (MIGRATE_REPLICA_OP != type_ && ADD_REPLICA_OP != type_) { STORAGE_LOG(INFO, "only migrate or add replica task need check disk space, others skip it", K(type_)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { ObPartitionKey& pkey = task_list_[i].arg_.key_; if (ObPartGroupMigrator::get_instance().is_stop()) { ret = OB_SERVER_IS_STOPPING; STORAGE_LOG(WARN, "server is stopping", K(ret)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(task_list_[i].arg_.dst_.get_replica_type())) { STORAGE_LOG(INFO, "dst has no ssstore, no need check disk space", "arg", task_list_[i].arg_); } else if (OB_FAIL(get_partition_required_size(pkey, partition_required_size))) { STORAGE_LOG(WARN, "failed to partition_required_size", K(ret), K(pkey)); } else { required_size += partition_required_size; } } if (OB_SUCC(ret) && required_size > 0) { if (OB_FAIL(OB_STORE_FILE.check_disk_full(required_size))) { if (OB_CS_OUTOF_DISK_SPACE == ret) { ret = OB_SERVER_MIGRATE_IN_DENIED; } ObTaskController::get().allow_next_syslog(); STORAGE_LOG(WARN, "failed to check_is_disk_full, cannot migrate in", K(ret), K(required_size)); } } } return ret; } int ObPartGroupMigrationTask::check_before_backup() { int ret = OB_SUCCESS; ObExternBackupInfoMgr extern_backup_info_mgr; ObClusterBackupDest cluster_backup_dest; ObBackupDest backup_dest; ObFakeBackupLeaseService fake_backup_lease_service; ObExternBackupInfo last_backup_info; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (BACKUP_REPLICA_OP != type_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "replica op is not backup, no need check", K(ret)); } else { const ObPhysicalBackupArg& backup_arg = task_list_[0].arg_.backup_arg_; const uint64_t tenant_id = task_list_[0].arg_.key_.get_tenant_id(); if (OB_FAIL(backup_dest.set(backup_arg.uri_header_, backup_arg.storage_info_))) { STORAGE_LOG(WARN, "failed to set backup dest", K(ret), K(backup_arg)); } else if (OB_FAIL(cluster_backup_dest.set(backup_dest, backup_arg.incarnation_))) { STORAGE_LOG(WARN, "failed to set cluster backup dest", K(ret), K(backup_dest)); } else if (OB_FAIL(extern_backup_info_mgr.init(tenant_id, cluster_backup_dest, fake_backup_lease_service))) { STORAGE_LOG(WARN, "failed to init extern backup info", K(ret), K(cluster_backup_dest)); } else if (OB_FAIL(extern_backup_info_mgr.get_last_info(last_backup_info))) { STORAGE_LOG(WARN, "failed to get last info", K(ret), K(cluster_backup_dest)); } else if (!last_backup_info.is_valid() || ObExternBackupInfo::DOING != last_backup_info.status_ || last_backup_info.backup_data_version_ != backup_arg.backup_data_version_ || last_backup_info.backup_schema_version_ != backup_arg.backup_schema_version_ || last_backup_info.backup_type_ != backup_arg.backup_type_ || last_backup_info.inc_backup_set_id_ != backup_arg.backup_set_id_) { ret = OB_ERR_UNEXPECTED; LOG_WARN("last backup info is unexpected", K(ret), K(last_backup_info), K(backup_arg)); } } return ret; } int ObPartGroupMigrationTask::get_partition_required_size(const common::ObPartitionKey& pkey, int64_t& required_size) { // statuc func int ret = OB_SUCCESS; share::ObPartitionTableOperator* pt_operator = NULL; ModulePageAllocator allocator(ObNewModIds::OB_PARTITION_MIGRATE); share::ObPartitionInfo partition_info; partition_info.set_allocator(&allocator); required_size = 0; if (NULL == (pt_operator = GCTX.pt_operator_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "pt_operator must not NULL.", K(ret)); } else if (OB_SUCCESS != (ret = pt_operator->get(pkey.get_table_id(), pkey.get_partition_id(), partition_info))) { STORAGE_LOG(WARN, "fail to get partition info.", K(ret), K(pkey)); } else { const common::ObIArray& replicas = partition_info.get_replicas_v2(); for (int64_t i = 0; OB_SUCC(ret) && i < replicas.count(); ++i) { if (replicas.at(i).required_size_ > required_size) { required_size = replicas.at(i).required_size_; } } } return ret; } int ObPartGroupMigrationTask::try_schedule_new_partition_migration() { int ret = OB_SUCCESS; { common::SpinRLockGuard guard(lock_); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (OB_SUCCESS != first_error_code_) { ret = first_error_code_; STORAGE_LOG(WARN, "first_error_code_ is set, skip schedule new partition migration", K(ret)); } } if (OB_SUCC(ret)) { if (VALIDATE_BACKUP_OP == type_) { if (OB_FAIL(try_schedule_partition_validate())) { STORAGE_LOG(WARN, "failed to try schedule partition validate", K(ret)); } } else if (BACKUP_BACKUPSET_OP == type_) { if (OB_FAIL(try_schedule_partition_backup_backupset())) { STORAGE_LOG(WARN, "failed to try schedule partition backup backupset", K(ret)); } } else if (BACKUP_ARCHIVELOG_OP == type_) { if (OB_FAIL(try_schedule_partition_backup_archivelog())) { STORAGE_LOG(WARN, "failed to try schedule partition backup archivelog", K(ret)); } } else { if (OB_FAIL(try_schedule_partition_migration())) { STORAGE_LOG(WARN, "failed to try scheudule partition migration", K(ret)); } } } return ret; } // caller must not hold wlock int ObPartGroupMigrationTask::schedule_migrate_dag(ObMigrateCtx& migrate_ctx) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(check_partition_checksum(migrate_ctx.replica_op_arg_.key_))) { STORAGE_LOG(WARN, "failed to check partition checksum, cannot schedule migration", K(ret), K(migrate_ctx.replica_op_arg_.key_)); } else if (OB_FAIL(migrate_ctx.generate_and_schedule_migrate_dag())) { STORAGE_LOG(WARN, "failed to generate migrate dag", K(ret)); } if (OB_FAIL(ret)) { share::ObTaskId fake_task_id; const bool during_migrating = false; int64_t backup_set_id = -1; if (VALIDATE_BACKUP_OP == migrate_ctx.replica_op_arg_.type_) { backup_set_id = migrate_ctx.replica_op_arg_.validate_arg_.backup_set_id_; } else if (BACKUP_BACKUPSET_OP == migrate_ctx.replica_op_arg_.type_) { backup_set_id = migrate_ctx.replica_op_arg_.backup_backupset_arg_.backup_set_id_; } if (OB_SUCCESS != (tmp_ret = set_part_task_finish( migrate_ctx.replica_op_arg_.key_, ret, fake_task_id, during_migrating, backup_set_id))) { LOG_WARN("failed to finish part task", K(ret)); } } return ret; } int ObPartGroupMigrationTask::try_finish_group_migration() { int ret = OB_SUCCESS; bool is_sub_task_finish = true; report_list_.reset(); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else { common::SpinWLockGuard guard(lock_); for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { ObPartMigrationTask& task = task_list_[i]; if (ObPartMigrationTask::FINISH != task.status_) { if (OB_SUCCESS != first_error_code_ && ObPartMigrationTask::INIT == task.status_) { task.status_ = ObPartMigrationTask::FINISH; task.result_ = first_error_code_; } else { is_sub_task_finish = false; } } else if (OB_SUCCESS != task.result_) { if (BACKUP_REPLICA_OP == get_type() && OB_PG_IS_REMOVED == task.result_) { // do nothing } else if (OB_SUCCESS == first_error_code_) { first_error_code_ = OB_ERR_SYS; STORAGE_LOG(ERROR, "sub task has error , but first_error_code_ is succ, use OB_ERR_SYS", K(*this)); } } } if (OB_SUCC(ret) && is_sub_task_finish) { SMART_VAR(ObReportPartMigrationTask, tmp_task) { if (OB_FAIL(report_list_.reserve(task_list_.count()))) { STORAGE_LOG(WARN, "failed to reserve report list", K(ret)); } for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { ObPartMigrationTask& sub_task = task_list_[i]; tmp_task.arg_ = sub_task.arg_; tmp_task.status_ = ObIPartMigrationTask::FINISH; tmp_task.result_ = sub_task.result_; tmp_task.need_report_checksum_ = sub_task.ctx_.need_report_checksum_; tmp_task.ctx_ = &sub_task.ctx_; tmp_task.data_statics_ = sub_task.ctx_.data_statics_; if (OB_FAIL(tmp_task.partitions_.assign(sub_task.ctx_.pg_meta_.partitions_))) { STORAGE_LOG(WARN, "failed to assign partitions", K(ret), K(sub_task.ctx_)); } else if (OB_FAIL(report_list_.push_back(tmp_task))) { // report_list is reserved before, should not fail here STORAGE_LOG(ERROR, "failed to add report list", K(ret)); } } } } } if (OB_SUCC(ret) && is_sub_task_finish) { if (OB_FAIL(finish_group_migration(report_list_))) { STORAGE_LOG(WARN, "failed to do finish_group_migration", K(ret)); } } return ret; } int ObPartGroupMigrationTask::finish_group_migration(ObIArray& report_list) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (REMOVE_REPLICA_OP == type_) { // The deleting op of the replica does not change the operation of the member list and reporting operation common::SpinWLockGuard guard(lock_); is_finished_ = true; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(INFO, "group migration all finish", K(ret), K(first_error_code_), K(report_list), K(*this)); } else if (BACKUP_REPLICA_OP == type_ || VALIDATE_BACKUP_OP == type_ || BACKUP_BACKUPSET_OP == type_ || BACKUP_ARCHIVELOG_OP == type_) { // backup base sstable only need change task status if (BACKUP_REPLICA_OP == type_) { ObArray report_res_list; if (OB_SUCCESS != (tmp_ret = ObMigrateUtil::get_report_result(report_list, report_res_list))) { LOG_WARN("failed to get report result", K(tmp_ret), K(report_list)); } else if (OB_SUCCESS != (tmp_ret = partition_service_->report_pg_backup_task(report_res_list))) { LOG_WARN("failed to report pg backup task", K(tmp_ret), K(report_res_list)); } } common::SpinWLockGuard guard(lock_); is_finished_ = true; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(INFO, "group migration-backup all finish", K(ret), K(first_error_code_), K(report_list), K(*this)); } else if (OB_UNLIKELY(report_list.count() <= 0)) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid arguments", K(ret), K(report_list.count())); } else { bool is_all_finish = false; bool is_report_succ = false; hash::ObHashSet removed_pkeys; if (0 == retry_job_start_ts_) { retry_job_start_ts_ = ObTimeUtility::current_time(); } if (OB_SUCCESS != (tmp_ret = report_meta_table(report_list, removed_pkeys))) { STORAGE_LOG(WARN, "fail to report meta table", K(ret)); } else { is_report_succ = true; } if (has_error()) { is_all_finish = true; if (OB_FAIL(build_failed_report_list(first_error_code_, report_list))) { STORAGE_LOG(WARN, "failed to build_failed_report_list", K(ret)); } } else if (is_report_succ) { // TODO () consider more detail about it if (RESTORE_REPLICA_OP == type_ || COPY_LOCAL_INDEX_OP == type_ || RESTORE_FOLLOWER_REPLICA_OP == type_ || RESTORE_STANDBY_OP == type_ || LINK_SHARE_MAJOR_OP == type_) { is_all_finish = true; STORAGE_LOG(INFO, "no need change member list", K(type_)); } else { if (OB_SUCCESS != (tmp_ret = try_change_member_list(report_list, is_all_finish))) { STORAGE_LOG(WARN, "failed to try_change_member_list", K(tmp_ret), K(first_error_code_)); } else { is_all_finish = true; } } if (first_error_code_ != OB_SUCCESS) { is_all_finish = true; if (OB_FAIL(build_failed_report_list(first_error_code_, report_list))) { STORAGE_LOG(WARN, "failed to build_failed_report_list", K(ret)); } } } if (is_all_finish) { { common::SpinWLockGuard guard(lock_); is_finished_ = true; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(INFO, "group migration all finish", K(ret), K(first_error_code_), K(report_list), K(*this)); } if (OB_SUCC(ret)) { for (int64_t i = 0; i < report_list.count(); ++i) { for (int64_t j = 0; j < report_list.at(i).partitions_.count(); ++j) { const ObPartitionKey& pkey = report_list.at(i).partitions_.at(j); if (OB_SUCCESS != (tmp_ret = partition_service_->report_migrate_in_indexes(pkey))) { STORAGE_LOG(WARN, "failed to handle_report_meta_table_callback", K(tmp_ret), K(i), K(report_list.at(i))); } } } } if (RESTORE_REPLICA_OP == type_ && OB_SUCCESS != first_error_code_) { if (OB_SUCCESS != (tmp_ret = report_restore_fatal_error_())) { LOG_WARN("failed to report restore failed", K(tmp_ret)); } } ObRebuildReplicaTaskScheduler& rebuild_scheduler = partition_service_->get_rebuild_replica_service().get_scheduler(); if (OB_SUCC(ret) && REBUILD_REPLICA_OP == type_) { ObRebuildReplicaResult results; for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { if (OB_FAIL(results.set_result(report_list.at(i).arg_.key_, report_list.at(i).result_))) { STORAGE_LOG(WARN, "failed to set rebuild result", K(ret), K(report_list.at(i))); } } } rebuild_scheduler.notify(); } } return ret; } int ObPartGroupMigrationTask::report_restore_fatal_error_() { int ret = OB_SUCCESS; int64_t job_id = 0; if (OB_FAIL(ObBackupInfoMgr::get_instance().get_restore_job_id(tenant_id_, job_id))) { if (OB_ENTRY_NOT_EXIST != ret) { LOG_WARN("failed to get restore info", K(ret), K(tenant_id_)); } else { ret = OB_SUCCESS; LOG_INFO("physical restore info not exist", K(tenant_id_), K(first_error_code_)); } } else if (OB_INVALID_ARGUMENT == first_error_code_ || OB_ERR_SYS == first_error_code_ || OB_INIT_TWICE == first_error_code_ || OB_ERR_UNEXPECTED == first_error_code_ || OB_INVALID_BACKUP_DEST == first_error_code_ || OB_INVALID_DATA == first_error_code_ || OB_CHECKSUM_ERROR == first_error_code_) { if (OB_FAIL(ObRestoreFatalErrorReporter::get_instance().add_restore_error_task( tenant_id_, PHYSICAL_RESTORE_MOD_STORAGE, first_error_code_, job_id, MYADDR))) { LOG_WARN("failed to report restore error", K(ret), K(tenant_id_), K(first_error_code_)); } } else { LOG_INFO("skip not fatal report restore error", K(tenant_id_), K(first_error_code_)); } return ret; } int ObPartGroupMigrationTask::try_change_member_list( ObIArray& report_list, bool& is_all_finish) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; const int64_t change_member_list_timeout = GCONF.sys_bkgd_migration_change_member_list_timeout; const int64_t start_ts = ObTimeUtility::current_time(); ObMigrateSrcInfo leader_info; bool need_batch = false; DEBUG_SYNC(BEFORE_CHANGE_MEMBER_LIST); if (0 == start_change_member_ts_) { start_change_member_ts_ = ObTimeUtility::current_time(); STORAGE_LOG(INFO, "start change member list", K(report_list)); } if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (report_list.count() <= 0) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "task list is empty", K(ret)); } else if (NORMAL_CHANGE_MEMBER_LIST != change_member_option_) { if (MIGRATE_REPLICA_OP == type_) { if (OB_SUCCESS != (tmp_ret = batch_remove_src_replica(report_list))) { STORAGE_LOG(WARN, "failed to batch remove src replica", K(ret)); } } STORAGE_LOG(INFO, "skip change member list is set, no need try_change_memer_list", K(ret), K(*this)); } else if (start_ts - start_change_member_ts_ > change_member_list_timeout) { is_all_finish = true; ret = OB_TIMEOUT; first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG(WARN, "max retry change member time exceeds", K(ret), K(start_change_member_ts_), K(change_member_list_timeout)); } else if (OB_FAIL( ObMigrateGetLeaderUtil::get_leader(report_list.at(0).arg_.key_, leader_info, true /*force_update*/))) { bool is_partition_exist = true; if (OB_SUCCESS != (tmp_ret = check_is_partition_exist(report_list.at(0).arg_.key_, is_partition_exist))) { LOG_WARN("failed to check is partition exist", K(tmp_ret)); } else if (!is_partition_exist) { first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; LOG_WARN("partition not exist, set migration failed", K(ret), "pkey", report_list.at(0).arg_.key_); } else { STORAGE_LOG(WARN, "fail to get leader", K(ret)); } } else if (OB_FAIL(check_need_batch_change_member_list(leader_info.src_addr_, report_list, need_batch))) { STORAGE_LOG(WARN, "failed to check_need_batch_change_member_list", K(ret)); } else if (need_batch) { if (MIGRATE_REPLICA_OP == type_ || FAST_MIGRATE_REPLICA_OP == type_) { if (OB_FAIL(try_batch_change_member_list(report_list, leader_info.src_addr_, is_all_finish))) { STORAGE_LOG(WARN, "failed to try_batch_change_member_list", K(ret), K(leader_info), K(report_list)); } } else if (CHANGE_REPLICA_OP == type_) { if (OB_FAIL(try_batch_add_member(report_list, leader_info.src_addr_, is_all_finish))) { LOG_WARN("failed to try_batch_add_member", K(ret), K(leader_info), K(report_list)); } } else { ret = OB_ERR_SYS; LOG_ERROR("not support batch type", K(ret), K_(type)); } } else { if (OB_FAIL(try_not_batch_change_member_list(report_list, is_all_finish))) { STORAGE_LOG(WARN, "failed to try_not_batch_change_member_list", K(ret)); } } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; STORAGE_LOG(INFO, "try_change_member_list", K(cost_ts), K(ret), K(is_all_finish), "count", report_list.count()); return ret; } int ObPartGroupMigrationTask::check_need_batch_change_member_list( const ObAddr& leader_addr, ObIArray& report_list, bool& need_batch) { int ret = OB_SUCCESS; ObMigrateSrcInfo tmp_leader_info; need_batch = false; const int64_t start_ts = ObTimeUtility::current_time(); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (report_list.count() <= 0 || !leader_addr.is_valid()) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "task list is empty", K(ret), K(leader_addr)); } else { need_batch = true; for (int64_t i = 0; OB_SUCC(ret) && need_batch && i < report_list.count(); ++i) { const ObReplicaOpArg& arg = report_list.at(i).arg_; if (MIGRATE_REPLICA_OP == arg.type_ || FAST_MIGRATE_REPLICA_OP == arg.type_) { if (!ObReplicaTypeCheck::is_paxos_replica(arg.dst_.get_replica_type())) { need_batch = false; } } else if (CHANGE_REPLICA_OP == arg.type_) { if (ObReplicaTypeCheck::is_paxos_replica(arg.src_.get_replica_type()) || !ObReplicaTypeCheck::is_paxos_replica(arg.dst_.get_replica_type())) { need_batch = false; } } else { need_batch = false; } if (OB_SUCC(ret) && need_batch) { if (OB_FAIL(ObMigrateGetLeaderUtil::get_leader(arg.key_, tmp_leader_info, true /*force_update*/))) { STORAGE_LOG(WARN, "failed to get leader address", K(ret), K(arg.key_)); int tmp_ret = OB_SUCCESS; bool is_schema_exist = true; if (OB_SUCCESS != (tmp_ret = check_is_partition_exist(arg.key_, is_schema_exist))) { LOG_WARN("check is partition exist failed", K(tmp_ret), K(arg.key_)); } else if (!is_schema_exist) { first_error_code_ = ret; LOG_WARN("partition schema not exist, set first_error_code_", K(first_error_code_)); } } else if (!tmp_leader_info.is_valid()) { need_batch = false; STORAGE_LOG(WARN, "some partition has no leader, cannot batch change member", K(arg), K(tmp_leader_info)); } else if (leader_addr != tmp_leader_info.src_addr_) { need_batch = false; STORAGE_LOG(WARN, "partitions have diff leader, cannot batch change member", K(tmp_leader_info), K(leader_addr), K(arg)); } } } } #ifdef ERRSIM if (OB_SUCC(ret) && need_batch) { need_batch = GCONF.allow_batch_change_member; if (!need_batch) { LOG_ERROR("fake not need batch change member list"); } } #endif ObTaskController::get().allow_next_syslog(); const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; STORAGE_LOG( INFO, "check_need_batch_change_member_list", K(cost_ts), K(ret), K(need_batch), "count", report_list.count()); return ret; } int ObPartGroupMigrationTask::fast_migrate_add_member_list_one_by_one( common::ObIArray& report_list, bool& is_all_finish) { int ret = OB_SUCCESS; int64_t start_ts = ObTimeUtility::current_time(); int64_t succ_count = 0; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (report_list.count() <= 0) { ret = OB_ERR_SYS; LOG_WARN("task list is empty", K(ret)); } else if (FAST_MIGRATE_REPLICA_OP != type_) { ret = OB_ERR_SYS; LOG_WARN("wrong task type", K(ret), K(type_)); } int64_t switch_epoch = GCTX.get_switch_epoch2(); for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { ObReportPartMigrationTask& cur_task = report_list.at(i); const ObPGKey& pg_key = cur_task.arg_.key_; ObMigrateCtx* cur_ctx = cur_task.ctx_; ObIPartitionGroup* pg = nullptr; int64_t quorum = 0; ObMigrateSrcInfo leader_info; ObMember new_member(GCTX.self_addr_, ObTimeUtility::current_time()); int tmp_ret = OB_SUCCESS; if (OB_ISNULL(cur_ctx) || OB_ISNULL(pg = cur_ctx->partition_guard_.get_partition_group())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("migrate ctx is null", K(ret), K(pg_key), KP(cur_ctx), KP(pg)); } else if (OB_SUCCESS != cur_task.result_) { LOG_INFO("fast migrate: skip add self to member list", K(cur_task.result_), K(pg_key)); } else if (OB_SUCCESS != (tmp_ret = pg->get_log_service()->get_replica_num(quorum))) { LOG_WARN("fail to get quorum", K(tmp_ret), K(pg_key)); } else if (OB_SUCCESS != (tmp_ret = ObMigrateGetLeaderUtil::get_leader(pg_key, leader_info, true /*force_update*/))) { LOG_WARN("fail to get pg leader", K(tmp_ret), K(pg_key)); } else { cur_ctx->is_member_change_finished_ = true; LOG_INFO("fast_migrate: succ to change member list", K(pg_key)); ++succ_count; } } if (OB_FAIL(ret)) { first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG(WARN, "set first_error_code_", K(ret), K(first_error_code_)); } int64_t cost_ts = ObTimeUtility::current_time() - start_ts; is_all_finish = true; LOG_INFO("fast migrate: finish change member list in turn", K(ret), K(cost_ts), "task_count", report_list.count(), K(succ_count)); return ret; } int ObPartGroupMigrationTask::try_batch_change_member_list( ObIArray& report_list, const ObAddr& leader_addr, bool& is_all_finish) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; hash::ObHashSet removed_pkeys; hash::ObHashSet added_pkeys; const int64_t start_ts = ObTimeUtility::current_time(); is_all_finish = false; STORAGE_LOG(INFO, "start try_batch_change_member_list"); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (report_list.count() <= 0) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "task list is empty", K(ret)); } else { if (MIGRATE_REPLICA_OP == type_) { if (OB_FAIL(try_batch_remove_member(report_list, leader_addr, removed_pkeys))) { STORAGE_LOG(WARN, "failed to try_batch_remove_member", K(ret), K(leader_addr), K(report_list)); } } else { ret = OB_ERR_SYS; LOG_WARN("only migrate or fast migrate allow batch change member list", K(ret), K(type_)); } } if (OB_SUCC(ret)) { // For migartion, if no need to retry, then add the deleted partition, otherwise no need to care is_all_finish = true; if (removed_pkeys.size() > 0) { if (OB_SUCCESS != (tmp_ret = try_batch_add_member(report_list, removed_pkeys, leader_addr, added_pkeys, is_all_finish))) { STORAGE_LOG(WARN, "failed to try batch add member", K(ret), K(tmp_ret), K(leader_addr), K(removed_pkeys), K(report_list)); } else if (FAST_MIGRATE_REPLICA_OP == type_) { for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { ObReportPartMigrationTask& report_task = report_list.at(i); if (OB_NOT_NULL(report_task.ctx_) && OB_HASH_EXIST == added_pkeys.exist_refactored(report_task.arg_.key_)) { report_task.ctx_->is_member_change_finished_ = true; LOG_INFO("fast_migrate: succ to change member list", K(report_task.arg_.key_)); } } } } // only migrate need remove src if (added_pkeys.size() > 0 && FAST_MIGRATE_REPLICA_OP != type_) { if (OB_SUCCESS != (tmp_ret = batch_remove_src_replica(report_list))) { STORAGE_LOG(WARN, "failed to batch remove src replica", K(ret), K(added_pkeys)); } } } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; STORAGE_LOG(INFO, "finish try_batch_change_member_list", K(cost_ts), K(ret), K(is_all_finish), "count", report_list.count(), "removed_count", removed_pkeys.size(), "added_count", added_pkeys.size()); return ret; } int ObPartGroupMigrationTask::try_batch_remove_member(ObIArray& report_list, const ObAddr& leader_addr, hash::ObHashSet& removed_pkeys) { int ret = OB_SUCCESS; ObChangeMemberArgs args; ObChangeMemberArg single_arg; ObChangeMemberCtxs change_member_info; const int64_t start_ts = ObTimeUtility::current_time(); STORAGE_LOG(INFO, "start try_batch_remove_member"); DEBUG_SYNC(BEFORE_BATCH_REMOVE_MEMBER); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(removed_pkeys.create(report_list.count()))) { STORAGE_LOG(WARN, "failed to create removed_pkeys", K(ret)); } else if (OB_FAIL(args.reserve(report_list.count()))) { STORAGE_LOG(WARN, "failed to reserve change member ctx", K(ret), K(report_list.count())); } else { ObIPartitionServiceRpc& rpc = partition_service_->get_pts_rpc(); for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { single_arg.partition_key_ = report_list.at(i).arg_.key_; single_arg.member_ = report_list.at(i).arg_.src_; single_arg.quorum_ = report_list.at(i).arg_.quorum_; single_arg.switch_epoch_ = report_list.at(0).arg_.switch_epoch_; if (OB_FAIL(args.push_back(single_arg))) { STORAGE_LOG(WARN, "failed to add single ctx", K(ret), K(single_arg)); } } if (OB_SUCC(ret)) { if (OB_FAIL(rpc.batch_post_remove_replica_mc_msg(leader_addr, args, change_member_info))) { STORAGE_LOG( WARN, "failed to batch_post_remove_replica_mc_msg", K(ret), K(leader_addr), K(args), K(change_member_info)); } } if (OB_SUCC(ret) || OB_PARTIAL_FAILED == ret) { ret = OB_SUCCESS; // ovewrwrite retry if (OB_FAIL(wait_batch_member_change_done(leader_addr, change_member_info))) { STORAGE_LOG( WARN, "failed to wait_change_member_done for remove", K(ret), K(leader_addr), K(change_member_info)); } if (OB_SUCC(ret) || OB_PARTIAL_FAILED == ret) { ret = OB_SUCCESS; // overwrite ret if (change_member_info.count() != report_list.count()) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "change_member_info count not match report_list count", K(ret), K(change_member_info.count()), K(report_list.count()), K(change_member_info), K(report_list)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < change_member_info.count(); ++i) { const ObChangeMemberCtx& ret_ctx = change_member_info.at(i); if (OB_SUCCESS != ret_ctx.ret_value_) { STORAGE_LOG(WARN, "skip add removed pkey", K(ret_ctx)); if (OB_FAIL(set_report_list_result(report_list, ret_ctx.partition_key_, ret_ctx.ret_value_))) { STORAGE_LOG(WARN, "failed to set report list result", K(ret)); } } else { if (OB_FAIL(removed_pkeys.set_refactored(ret_ctx.partition_key_))) { removed_pkeys.reuse(); STORAGE_LOG(WARN, "failed to set removed_pkeys", K(ret)); } } } } } if (OB_FAIL(ret)) { first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG(WARN, "set first_error_code_", K(ret), K(first_error_code_)); ret = OB_SUCCESS; // overwrite ret } } } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; ObTaskController::get().allow_next_syslog(); STORAGE_LOG( INFO, "finish try_batch_remove_member", K(cost_ts), K(ret), K(first_error_code_), "count", report_list.count()); return ret; } int ObPartGroupMigrationTask::try_batch_add_member(ObIArray& report_list, const hash::ObHashSet& removed_pkeys, const ObAddr& leader_addr, hash::ObHashSet& added_keys, bool& is_all_finish) { int ret = OB_SUCCESS; #ifdef ERRSIM UNUSEDx(report_list, removed_pkeys, leader_addr, added_keys, is_all_finish); ret = E(EventTable::EN_FAST_MIGRATE_ADD_MEMBER_FAIL) OB_SUCCESS; if (OB_FAIL(ret)) { LOG_INFO("errsim add member failure", K(ret)); return ret; } #endif ObChangeMemberArgs args; ObChangeMemberCtxs add_member_info; const int64_t start_ts = ObTimeUtility::current_time(); STORAGE_LOG(INFO, "start try_batch_add_member"); DEBUG_SYNC(BEFORE_BATCH_ADD_MEMBER); // is_all_finish is not initialized here. It may have been assigned true outside if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(added_keys.create(removed_pkeys.size()))) { STORAGE_LOG(WARN, "failed to create added_keys", K(ret)); } else if (OB_FAIL(build_add_member_ctx(report_list, removed_pkeys, args))) { STORAGE_LOG(WARN, "failed to build_add_member_ctx", K(ret), K(removed_pkeys), K(report_list)); } else { ObIPartitionServiceRpc& rpc = partition_service_->get_pts_rpc(); if (OB_SUCC(ret)) { if (OB_FAIL(rpc.batch_post_add_replica_mc_msg(leader_addr, args, add_member_info))) { STORAGE_LOG( WARN, "failed to batch_post_add_replica_mc_msg", K(ret), K(leader_addr), K(args), K(add_member_info)); } } if (OB_SUCC(ret) || OB_PARTIAL_FAILED == ret) { is_all_finish = true; ret = OB_SUCCESS; // overwrite ret if (OB_FAIL(wait_batch_member_change_done(leader_addr, add_member_info))) { STORAGE_LOG(WARN, "failed to wait_change_member_done for add", K(ret), K(leader_addr), K(add_member_info)); } if (OB_SUCC(ret) || OB_PARTIAL_FAILED == ret) { ret = OB_SUCCESS; // overwrite ret if (add_member_info.count() != removed_pkeys.size()) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "removed_pkeys count not match add_member_info count", K(ret), K(removed_pkeys.size()), K(add_member_info.count()), K(removed_pkeys), K(add_member_info)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < add_member_info.count(); ++i) { const ObChangeMemberCtx& ret_ctx = add_member_info.at(i); if (OB_SUCCESS != ret_ctx.ret_value_) { STORAGE_LOG(WARN, "skip add added pkey", K(ret_ctx)); if (OB_FAIL(set_report_list_result(report_list, ret_ctx.partition_key_, ret_ctx.ret_value_))) { STORAGE_LOG(WARN, "failed to set report list result", K(ret)); } } else if (OB_FAIL(added_keys.set_refactored(add_member_info.at(i).partition_key_))) { STORAGE_LOG(WARN, "failed to set added keys", K(ret)); added_keys.reuse(); } } } } } } if (OB_FAIL(ret) && is_all_finish) { first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG(WARN, "set first_error_code_", K(ret), K(first_error_code_)); } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(INFO, "finish try_batch_add_member", K(cost_ts), K(ret), K(first_error_code_), K(is_all_finish), "count", removed_pkeys.size()); return ret; } int ObPartGroupMigrationTask::try_batch_add_member( ObIArray& report_list, const ObAddr& leader_addr, bool& is_all_finish) { int ret = OB_SUCCESS; hash::ObHashSet removed_pkeys; hash::ObHashSet added_keys; const int64_t start_ts = ObTimeUtility::current_time(); is_all_finish = false; STORAGE_LOG(INFO, "start try_batch_add_member"); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (report_list.count() <= 0) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "task list is empty", K(ret)); } else if (CHANGE_REPLICA_OP != type_) { ret = OB_ERR_SYS; LOG_ERROR("only migrate or change replica allow batch change member list", K(ret), K(type_)); } else if (OB_FAIL(removed_pkeys.create(report_list.count()))) { STORAGE_LOG(WARN, "failed to create removed_pkeys", K(ret)); } else { // chang replica also needs to set removed_pkeys. The logic behind it is than only pkeys than have been successfully // removed will be added for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { if (OB_FAIL(removed_pkeys.set_refactored(report_list.at(i).arg_.key_))) { LOG_WARN("failed to add change replica pkey", K(ret)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(try_batch_add_member(report_list, removed_pkeys, leader_addr, added_keys, is_all_finish))) { LOG_WARN("failed to try_batch_add_member", K(ret)); } else if (CHANGE_REPLICA_OP == type_) { for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { ObReportPartMigrationTask& report_task = report_list.at(i); if (OB_NOT_NULL(report_task.ctx_) && OB_HASH_EXIST == added_keys.exist_refactored(report_task.arg_.key_)) { report_task.ctx_->is_member_change_finished_ = true; LOG_INFO("change replica: succ to add member list", K(report_task.arg_.key_)); } } } } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; STORAGE_LOG(INFO, "finish try_batch_add_member", K(cost_ts), K(ret), K(is_all_finish), "count", report_list.count(), "added_count", added_keys.size()); return ret; } int ObPartGroupMigrationTask::build_add_member_ctx(const ObIArray& report_list, const hash::ObHashSet& removed_pkeys, ObChangeMemberArgs& ctx) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObChangeMemberArg single_ctx; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(ctx.reserve(removed_pkeys.size()))) { STORAGE_LOG(WARN, "failed to reserve change member ctx", K(ret), K(removed_pkeys.size())); } for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { const ObReplicaOpArg& arg = report_list.at(i).arg_; if (OB_HASH_EXIST == (tmp_ret = removed_pkeys.exist_refactored(arg.key_))) { single_ctx.partition_key_ = arg.key_; single_ctx.member_ = arg.dst_; single_ctx.quorum_ = arg.quorum_; single_ctx.switch_epoch_ = report_list.at(0).arg_.switch_epoch_; if (OB_FAIL(ctx.push_back(single_ctx))) { STORAGE_LOG(WARN, "failed to add single ctx", K(ret), K(single_ctx)); } } else if (OB_HASH_NOT_EXIST != tmp_ret) { ret = tmp_ret; STORAGE_LOG(WARN, "failed to check removed pkey exist", K(ret)); } } return ret; } int ObPartGroupMigrationTask::remove_src_replica(const ObIArray& report_list) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; share::ObTaskId dummy_id; dummy_id.init(OBSERVER.get_self()); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { const ObReplicaOpArg& arg = report_list.at(i).arg_; if (OB_SUCCESS != report_list.at(i).result_) { STORAGE_LOG(INFO, "skip remove failed partition", K(report_list.at(i))); } else { obrpc::ObRemoveReplicaArg remove_replica_arg = {arg.key_, arg.src_}; if (OB_SUCCESS != (tmp_ret = partition_service_->get_pts_rpc().post_remove_replica( arg.src_.get_server(), remove_replica_arg))) { STORAGE_LOG(WARN, "remove replica from source observer failed", K(tmp_ret), K(remove_replica_arg)); } else { STORAGE_LOG(INFO, "remove source replica successfully", K(arg.src_), K(remove_replica_arg)); } } } return ret; } int ObPartGroupMigrationTask::batch_remove_src_replica(const ObIArray& report_list) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; share::ObTaskId dummy_id; dummy_id.init(OBSERVER.get_self()); ObArenaAllocator allocator; typedef hash::ObHashMap JoinMap; JoinMap join_map; // Map aggregated by serve if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(join_map.create(report_list.count(), ObModIds::OB_PARTITION_MIGRATOR))) { STORAGE_LOG(WARN, "failed to create hash table for join", K(ret)); } else { obrpc::ObRemoveReplicaArgs* remove_replica_args = NULL; void* buf = NULL; for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { const ObReplicaOpArg& arg = report_list.at(i).arg_; buf = NULL; remove_replica_args = NULL; if (OB_SUCCESS != report_list.at(i).result_) { STORAGE_LOG(INFO, "skip remove failed partition", "pkey", arg.key_); } else { if (OB_FAIL(join_map.get_refactored(arg.src_.get_server(), remove_replica_args))) { if (OB_HASH_NOT_EXIST != ret) { STORAGE_LOG(WARN, "failed to get remove replica args from map", K(ret)); } else if (NULL == (buf = allocator.alloc(sizeof(obrpc::ObRemoveReplicaArgs)))) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to alloc buf", K(ret)); } else if (NULL == (remove_replica_args = new (buf) obrpc::ObRemoveReplicaArgs())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to new remove replica args"); } else if (OB_FAIL(join_map.set_refactored(arg.src_.get_server(), remove_replica_args))) { STORAGE_LOG(WARN, "failed to set join map", K(ret)); } } if (OB_SUCC(ret)) { obrpc::ObRemoveReplicaArg remove_replica_arg = {arg.key_, arg.src_}; if (OB_FAIL(remove_replica_args->arg_array_.push_back(remove_replica_arg))) { STORAGE_LOG(WARN, "failed to push remove replica arg into array", K(ret)); } } } } if (OB_SUCC(ret)) { for (JoinMap::iterator iter = join_map.begin(); iter != join_map.end(); ++iter) { const ObAddr server = iter->first; remove_replica_args = iter->second; if (OB_SUCCESS != (tmp_ret = partition_service_->get_pts_rpc().batch_post_remove_replica(server, *remove_replica_args))) { STORAGE_LOG(WARN, "remove replica from source observer failed", K(tmp_ret), K(*remove_replica_args)); } else { STORAGE_LOG(INFO, "remove source replica successfully", K(*remove_replica_args)); } } } // demonstrate destruct remove_replica_args for (JoinMap::iterator iter = join_map.begin(); iter != join_map.end(); ++iter) { remove_replica_args = iter->second; if (NULL != remove_replica_args) { remove_replica_args->~ObRemoveReplicaArgs(); } } join_map.clear(); } return ret; } int ObPartGroupMigrationTask::check_member_major_sstable_enough(ObIArray& report_list) { int ret = OB_SUCCESS; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "ObPartGroupMigrationTask has not been inited", K(ret)); } else if (OB_UNLIKELY(report_list.count() <= 0)) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid arguments", K(ret), K(report_list.count())); } else { for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { if (ObReplicaTypeCheck::is_replica_with_ssstore(report_list.at(i).arg_.dst_.get_replica_type()) && (ADD_REPLICA_OP == report_list.at(i).arg_.type_ || MIGRATE_REPLICA_OP == report_list.at(i).arg_.type_ // || CHANGE_REPLICA_OP == report_list.at(i).arg_.type_ // TODO(): open it after add batch check || REBUILD_REPLICA_OP == report_list.at(i).arg_.type_)) { if (OB_FAIL(check_member_pg_major_sstable_enough(report_list.at(i)))) { STORAGE_LOG(WARN, "fail to check member pg major sstable enough", K(ret)); } } } } if (OB_FAIL(ret)) { if (OB_MAJOR_SSTABLE_NOT_EXIST == ret || OB_NOT_SUPPORTED == ret) { first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG(WARN, "set first_error_code_", K(ret), K(first_error_code_), K(report_list)); } else { const int64_t now = ObTimeUtility::current_time(); if (now > retry_job_start_ts_ + RETRY_JOB_MAX_WAIT_TIMEOUT) { ret = OB_TIMEOUT; first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG(WARN, "set first_error_code_", K(ret), K(first_error_code_), K(report_list)); } else { STORAGE_LOG(WARN, "fail to check member major sstable enough, retry it", K(ret)); } } } return ret; } int ObPartGroupMigrationTask::check_member_pg_major_sstable_enough(ObReportPartMigrationTask& task) { int ret = OB_SUCCESS; ObMigrateSrcInfo leader_info; ObIPartitionGroupGuard guard; ObIPartitionGroup* partition = NULL; ObPGStorage* pg_storage = NULL; ObTablesHandle tables_handle; ObArray table_ids; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "ObPartGroupMigrationTask has not been inited", K(ret)); } else if (ObReplicaTypeCheck::is_replica_with_ssstore(task.arg_.dst_.get_replica_type()) && (ADD_REPLICA_OP != task.arg_.type_ && MIGRATE_REPLICA_OP != task.arg_.type_ // || CHANGE_REPLICA_OP != task.arg_.arg_.type_ // TODO(): open it after add batch check && REBUILD_REPLICA_OP != task.arg_.type_)) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "failed to check member pg major sstable enough", K(ret), K(task)); } else if (OB_FAIL(ObMigrateGetLeaderUtil::get_leader(task.arg_.key_, leader_info, true /*force_update*/))) { STORAGE_LOG(WARN, "failed to get leader address", K(ret), K(task.arg_.key_)); } else if (OB_FAIL(partition_service_->get_partition(task.arg_.key_, guard))) { STORAGE_LOG(WARN, "failed to get partition group", K(ret), K(task.arg_)); } else if (OB_ISNULL(partition = guard.get_partition_group())) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "partition group should not be NULL", K(ret), KP(partition)); } else if (OB_ISNULL(pg_storage = &(partition->get_pg_storage()))) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "pg_storage should not be NULL", K(ret), KP(pg_storage)); } else if (OB_FAIL(pg_storage->get_last_all_major_sstable(tables_handle))) { STORAGE_LOG(WARN, "failed to get last all major sstable", K(ret), K(task)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < tables_handle.get_count(); ++i) { ObITable* table = tables_handle.get_tables().at(i); if (OB_ISNULL(table)) { ret = OB_ERR_SYS; STORAGE_LOG(WARN, "error sys, table must not be NULL", K(ret)); } else if (OB_FAIL(table_ids.push_back(table->get_key().table_id_))) { STORAGE_LOG(WARN, "failed to push table id into array", K(ret)); } } if (OB_SUCC(ret)) { obrpc::ObMemberMajorSSTableCheckArg arg; arg.pkey_ = task.arg_.key_; if (OB_FAIL(arg.table_ids_.assign(table_ids))) { STORAGE_LOG(WARN, "fail to assign table ids", K(ret)); } else if (OB_FAIL(partition_service_->get_pts_rpc().check_member_pg_major_sstable_enough( leader_info.src_addr_, arg))) { STORAGE_LOG(WARN, "fail to check member pg major sstable enough", K(ret), K(arg)); if (OB_NOT_SUPPORTED == ret) { if (REBUILD_REPLICA_OP == task.arg_.type_) { // not supported may rpc send to old server, for compat STORAGE_LOG(WARN, "result is not supported, may src server not support the rpc", K(ret)); ret = OB_SUCCESS; } else if (OB_FAIL(partition_service_->get_pts_rpc().check_member_major_sstable_enough( leader_info.src_addr_, arg))) { STORAGE_LOG(WARN, "fail to check member major sstable enough", K(ret), K(leader_info)); } } } } } return ret; } int ObPartGroupMigrationTask::report_meta_table( const ObIArray& report_list, const hash::ObHashSet& member_removed_pkeys) { int ret = OB_SUCCESS; const int64_t ONE_ROUND_WAIT_TIMEOUT = 60 * 1000 * 1000; // 1m const int64_t RETRY_INTERVAL = 100 * 1000; // 100ms const int64_t start_ts = ObTimeUtility::current_time(); ObArray check_result; FLOG_INFO("start report meta table", K(report_list)); if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "ObPartGroupMigrationTask has not been inited", K(ret)); } else if (OB_UNLIKELY(report_list.count() <= 0)) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid arguments", K(ret), K(report_list.count())); } else if (OB_FAIL(check_result.reserve(report_list.count()))) { STORAGE_LOG(WARN, "fail to reserve check result", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { const ObReportPartMigrationTask& report_task = report_list.at(i); if (OB_FAIL(partition_service_->handle_report_meta_table_callback( report_task.arg_.key_, report_task.result_, report_task.need_report_checksum_))) { STORAGE_LOG(WARN, "fail to report meta table", K(ret)); } } if (OB_SUCC(ret) && !has_error()) { for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { const ObReportPartMigrationTask& report_task = report_list.at(i); bool need_check = ObReplicaTypeCheck::is_replica_with_ssstore(report_task.arg_.dst_.get_replica_type()); if (FAST_MIGRATE_REPLICA_OP == type_) { int exist_ret = member_removed_pkeys.exist_refactored(report_list.at(i).arg_.key_); if (OB_HASH_EXIST != exist_ret && OB_HASH_NOT_EXIST != exist_ret) { ret = exist_ret; LOG_WARN("check is partition member list removed fail", K(ret), K(i), "report_task", report_list.at(i)); } else if (OB_HASH_NOT_EXIST == exist_ret) { need_check = false; } } if (OB_FAIL(ret)) { } else if (OB_FAIL(check_result.push_back(!need_check))) { STORAGE_LOG(WARN, "fail to push back check result", K(ret)); } } } // If the previous migration fails, there is no need to wait while (OB_SUCC(ret) && !has_error()) { if (OB_FAIL(check_report_done(report_list, check_result))) { STORAGE_LOG(WARN, "fail to check report done", K(ret)); } else { bool is_all_done = true; for (int64_t i = 0; OB_SUCC(ret) && i < check_result.count() && is_all_done; ++i) { is_all_done = check_result.at(i); } if (is_all_done) { break; } else { const int64_t now = ObTimeUtility::current_time(); if (now > retry_job_start_ts_ + RETRY_JOB_MAX_WAIT_TIMEOUT) { ret = OB_TIMEOUT; first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG(WARN, "wait report meta table done timeout, migration failed", K(ret)); } else if (now > start_ts + ONE_ROUND_WAIT_TIMEOUT) { ret = OB_TIMEOUT; STORAGE_LOG(WARN, "wait report meta table done timeout, retry later", K(ret)); } else { usleep(RETRY_INTERVAL); } } } } } const int64_t end_ts = ObTimeUtility::current_time(); FLOG_INFO("wait report meta table done", "cost_ts", end_ts - start_ts, K(ret)); return ret; } int ObPartGroupMigrationTask::check_report_done( const ObIArray& report_list, ObIArray& result) { int ret = OB_SUCCESS; share::ObPartitionTableOperator* pt_operator = NULL; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "ObPartGroupMigrationTask has not been inited", K(ret)); } else if (report_list.count() <= 0 || (report_list.count() != result.count())) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid arguments", K(ret), K(report_list.count()), K(result.count())); } else if (OB_ISNULL(pt_operator = GCTX.pt_operator_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "error unexpected, pt operator must not be NULL", K(ret)); } else { share::ObPartitionInfo partition_info; share::ObReplicaStatus replica_status; ObArenaAllocator allocator(ObModIds::OB_PARTITION_MIGRATOR); partition_info.set_allocator(&allocator); for (int64_t i = 0; OB_SUCC(ret) && i < result.count(); ++i) { const ObPartitionReplica* replica = NULL; partition_info.reuse(); if (OB_FAIL(partition_service_->get_replica_status(report_list.at(i).arg_.key_, replica_status))) { STORAGE_LOG(WARN, "fail to get replica status", K(ret)); } else if (REPLICA_STATUS_NORMAL != replica_status) { ret = OB_STATE_NOT_MATCH; first_error_code_ = OB_SUCCESS == first_error_code_ ? ret : first_error_code_; STORAGE_LOG( WARN, "partition status not match, migration failed", K(report_list.at(i).arg_.key_), K(replica_status)); } else if (!result.at(i)) { if (OB_FAIL(pt_operator->get(report_list.at(i).arg_.key_.get_table_id(), report_list.at(i).arg_.key_.get_partition_id(), partition_info))) { STORAGE_LOG(WARN, "fail to get partition info.", K(ret), "pkey", report_list.at(i).arg_.key_); } else if (OB_FAIL(partition_info.find(GCTX.self_addr_, replica))) { if (OB_ENTRY_NOT_EXIST != ret) { STORAGE_LOG(WARN, "fail to find replica", K(ret), K(GCTX.self_addr_)); } else { ret = OB_SUCCESS; } } else if (OB_ISNULL(replica)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "replica must not be NULL", K(ret)); } else { result.at(i) = REPLICA_STATUS_NORMAL == replica->status_; } } } } return ret; } int ObPartGroupMigrationTask::wait_batch_member_change_done( const ObAddr& leader_addr, ObChangeMemberCtxs& change_member_info) { int ret = OB_SUCCESS; int64_t max_wait_batch_member_change_done_us = 300 * 1000 * 1000; // 5m const int64_t start_ts = ObTimeUtility::current_time(); #ifdef ERRSIM int64_t tmp = GCONF.max_wait_batch_member_change_done_us; if (tmp != max_wait_batch_member_change_done_us) { max_wait_batch_member_change_done_us = tmp; STORAGE_LOG(INFO, "fake max_wait_batch_member_change_done_us", K(max_wait_batch_member_change_done_us)); } #endif if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { while (OB_SUCC(ret)) { const int64_t now = ObTimeUtility::current_time(); if (now > start_ts + max_wait_batch_member_change_done_us) { ret = OB_TIMEOUT; STORAGE_LOG(WARN, "wait batch member change done retry timeout", K(ret)); break; } else if (OB_FAIL( partition_service_->get_pts_rpc().is_batch_member_change_done(leader_addr, change_member_info))) { if (OB_PARTIAL_FAILED != ret) { STORAGE_LOG(WARN, "failed to is_batch_member_change_done", K(ret), K(leader_addr), K(change_member_info)); } else { STORAGE_LOG(INFO, "failed to is_batch_member_change_done", K(ret), K(leader_addr), K(change_member_info)); } } else { for (int64_t i = 0; OB_SUCC(ret) && i < change_member_info.count(); ++i) { if (OB_SUCCESS != change_member_info.at(i).ret_value_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "is_batch_member_change_done return OB_SUCCESS, butn has failed sub part", K(ret), K(i), K(change_member_info.at(i))); } } break; } if (OB_PARTIAL_FAILED == ret) { for (int64_t i = 0; i < change_member_info.count(); ++i) { if (OB_EAGAIN == change_member_info.at(i).ret_value_) { STORAGE_LOG( INFO, "some partition change member is not finish, need retry", K(i), K(change_member_info.at(i))); ret = OB_SUCCESS; break; } } } else if (OB_TIMEOUT == ret) { STORAGE_LOG(INFO, "send is_batch_member_change_done fail, need retry", K(ret)); ret = OB_SUCCESS; } if (OB_SUCC(ret)) { usleep(100 * 1000); // 100ms } } } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; LOG_INFO("finish wait_batch_member_change_done", K(ret), K(cost_ts)); return ret; } int ObPartGroupMigrationTask::try_not_batch_change_member_list( ObIArray& report_list, bool& is_all_finish) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; bool could_retry = false; SMART_VAR(ObPartMigrationTask, tmp_task) { const int64_t start_ts = ObTimeUtility::current_time(); is_all_finish = false; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (report_list.count() <= 0) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "task list is empty", K(ret)); } else { ObReportPartMigrationTask& first_task = report_list.at(0); if (OB_SUCCESS != (tmp_ret = partition_service_->handle_member_change_callback( first_task.arg_, first_task.result_, could_retry))) { if (MIGRATE_REPLICA_OP == type_ && could_retry && need_retry_change_member_list(tmp_ret)) { STORAGE_LOG(WARN, "failed to handle_replica_callback, retry later", K(tmp_ret), K(first_task)); } else { is_all_finish = true; STORAGE_LOG(WARN, "failed to handle_replica_callback, failed all sub tasks", K(tmp_ret), K(could_retry), K(first_task)); if (OB_FAIL(build_failed_report_list(tmp_ret, report_list))) { STORAGE_LOG(WARN, "failed to build failed report list", K(ret)); } } } else { first_task.ctx_->is_member_change_finished_ = true; is_all_finish = true; for (int64_t i = 1; OB_SUCC(ret) && i < report_list.count(); ++i) { ObReportPartMigrationTask& sub_task = report_list.at(i); if (OB_SUCCESS != (tmp_ret = partition_service_->handle_member_change_callback( sub_task.arg_, sub_task.result_, could_retry))) { sub_task.result_ = tmp_ret; STORAGE_LOG(WARN, "failed to handle_replica_callback,", K(tmp_ret), K(sub_task)); } else { sub_task.ctx_->is_member_change_finished_ = true; } } } const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; STORAGE_LOG(INFO, "finish try_not_batch_change_member_list", "count", report_list.count(), K(start_ts), K(cost_ts), "avg_cost", cost_ts / report_list.count()); } } return ret; } bool ObPartGroupMigrationTask::need_retry_change_member_list(const int32_t result_code) { bool need_retry = false; const int64_t change_member_list_timeout = GCONF.sys_bkgd_migration_change_member_list_timeout; const int64_t now = ObTimeUtility::current_time(); int64_t large_migration_need_retry_cost_time = LARGE_MIGRATION_NEED_RETRY_COST_TIME; #ifdef ERRSIM large_migration_need_retry_cost_time = GCONF.sys_bkgd_migration_large_task_threshold; STORAGE_LOG(INFO, "use GCONF.sys_bkgd_migration_large_task_threshold", K(large_migration_need_retry_cost_time)); #endif if (!is_inited_) { STORAGE_LOG(ERROR, "not inited"); } else { // Retrial condition // 1. It tasks more than 2 hours to copy data // 2. The default retry times has not been exceeded // 3. Match specific error code need_retry = true; if (change_member_list_timeout <= 0) { need_retry = false; STORAGE_LOG(INFO, "change member list time out is 0, no need retry", K(change_member_list_timeout)); } else if (start_change_member_ts_ - schedule_ts_ < large_migration_need_retry_cost_time) { need_retry = false; STORAGE_LOG(INFO, "no large migration, no need retry", K(schedule_ts_), K(start_change_member_ts_), K(large_migration_need_retry_cost_time)); } else if (now > start_change_member_ts_ + change_member_list_timeout) { need_retry = false; STORAGE_LOG(INFO, "retry timeout expired, no need retry", K(now), K(start_change_member_ts_), K(change_member_list_timeout)); } else if (result_code != OB_STATE_NOT_MATCH && result_code != OB_LOG_NOT_SYNC) { need_retry = false; STORAGE_LOG(INFO, "no need retry for these error", K(result_code)); } } return need_retry; } int ObPartGroupMigrationTask::inner_schedule_partition(ObPartMigrationTask*& task, bool& need_schedule) { int ret = OB_SUCCESS; need_schedule = false; task = NULL; common::SpinRLockGuard guard(lock_); for (int64_t i = 0; OB_SUCC(ret) && i < task_list_.count(); ++i) { task = &task_list_[i]; bool as_data_source = false; if (type_ != task->arg_.type_ || !is_replica_op_valid(task->arg_.type_)) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "invalid type, cannot handle this task", K(ret), K(i), K(*task)); } else if (REMOVE_REPLICA_OP == type_) { // do nothing } else if (OB_FAIL(check_can_as_data_source(type_, task->arg_.data_src_.get_replica_type(), task->arg_.dst_.get_replica_type(), as_data_source))) { STORAGE_LOG(WARN, "fail to check can as data source", K(ret), K(task->arg_)); } else if (!as_data_source) { ret = OB_NOT_SUPPORTED; first_error_code_ = ret; LOG_ERROR("invalid data src", K(ret), K(task->arg_)); } else if (ObPartMigrationTask::INIT == task->status_) { if (OB_FAIL(build_migrate_ctx(task->arg_, task->ctx_))) { STORAGE_LOG(WARN, "fail to build migrate ctx", K(ret)); } else { need_schedule = true; break; } } } return ret; } int ObPartGroupMigrationTask::try_schedule_partition_migration() { int ret = OB_SUCCESS; bool need_schedule = true; ObPartMigrationTask* task = NULL; const int64_t data_copy_in_concurrency = GCONF.server_data_copy_in_concurrency; while (OB_SUCC(ret) && need_schedule && ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_MIGRATE) < data_copy_in_concurrency) { if (OB_FAIL(inner_schedule_partition(task, need_schedule))) { STORAGE_LOG(WARN, "failed to inner schedule partition", K(ret)); } else if (!need_schedule) { // do nothing } else if (OB_ISNULL(task)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "task need schedule but is NULL", K(ret), KP(task)); } else if (OB_FAIL(schedule_migrate_dag(task->ctx_))) { STORAGE_LOG(WARN, "failed to schedule migrate dag", K(ret), K(task->arg_)); } } return ret; } int ObPartGroupMigrationTask::try_schedule_partition_validate() { int ret = OB_SUCCESS; bool need_schedule = true; ObPartMigrationTask* task = NULL; int64_t data_validate_concurrency = GCONF.backup_concurrency; int32_t up_limit = 0; if (OB_FAIL(ObDagScheduler::get_instance().get_up_limit(ObIDag::DAG_ULT_BACKUP, up_limit))) { STORAGE_LOG(WARN, "failed to get up limit", K(ret)); } else { data_validate_concurrency = data_validate_concurrency == 0 ? up_limit : data_validate_concurrency; FLOG_INFO("validate concurrency", K(data_validate_concurrency), K(up_limit)); } while (OB_SUCC(ret) && need_schedule && ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_VALIDATE) < data_validate_concurrency) { if (OB_FAIL(inner_schedule_partition(task, need_schedule))) { STORAGE_LOG(WARN, "failed to inner schedule partiion", K(ret)); } else if (!need_schedule) { // do nothing } else if (OB_ISNULL(task)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "task need schedule but is NULL", K(ret), KP(task)); } else if (OB_FAIL(schedule_migrate_dag(task->ctx_))) { STORAGE_LOG(WARN, "failed to schedule migrate dag", K(ret), K(task->arg_)); } } return ret; } int ObPartGroupMigrationTask::try_schedule_partition_backup_backupset() { int ret = OB_SUCCESS; bool need_schedule = true; ObPartMigrationTask* task = NULL; int64_t data_backup_concurrency = GCONF.backup_concurrency; int32_t up_limit = 0; if (OB_FAIL(ObDagScheduler::get_instance().get_up_limit(ObIDag::DAG_ULT_BACKUP, up_limit))) { STORAGE_LOG(WARN, "failed to get up limit", K(ret)); } else { data_backup_concurrency = data_backup_concurrency == 0 ? up_limit : data_backup_concurrency; FLOG_INFO("backup concurrency", K(data_backup_concurrency), K(up_limit)); } while (OB_SUCC(ret) && need_schedule && ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_BACKUP_BACKUPSET) < data_backup_concurrency) { if (OB_FAIL(inner_schedule_partition(task, need_schedule))) { STORAGE_LOG(WARN, "failed to inner schedule partition", K(ret)); } else if (!need_schedule) { STORAGE_LOG(INFO, "no need schedule"); } else if (OB_ISNULL(task)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "task need schedule but is NULL", K(ret), KP(task)); } else if (OB_FAIL(schedule_migrate_dag(task->ctx_))) { STORAGE_LOG(WARN, "failed to schedule migrate dag", K(ret), K(task->arg_)); } } return ret; } int ObPartGroupMigrationTask::try_schedule_partition_backup_archivelog() { int ret = OB_SUCCESS; bool need_schedule = true; ObPartMigrationTask* task = NULL; int64_t data_backup_concurrency = GCONF.backup_concurrency; int32_t up_limit = 0; if (OB_FAIL(ObDagScheduler::get_instance().get_up_limit(ObIDag::DAG_ULT_BACKUP, up_limit))) { STORAGE_LOG(WARN, "failed to get up limit", K(ret)); } else { data_backup_concurrency = data_backup_concurrency == 0 ? up_limit : data_backup_concurrency; FLOG_INFO("backup concurrency", K(data_backup_concurrency), K(up_limit)); } while (OB_SUCC(ret) && need_schedule && ObDagScheduler::get_instance().get_dag_count(ObIDag::DAG_TYPE_BACKUP_ARCHIVELOG) < data_backup_concurrency) { if (OB_FAIL(inner_schedule_partition(task, need_schedule))) { STORAGE_LOG(WARN, "failed to inner schedule partition", K(ret)); } else if (!need_schedule) { STORAGE_LOG(INFO, "no need schedule"); } else if (OB_ISNULL(task)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "task need schedule but is NULL", K(ret), KP(task)); } else if (OB_FAIL(schedule_migrate_dag(task->ctx_))) { STORAGE_LOG(WARN, "failed to schedule migrate dag", K(ret), K(task->arg_)); } } return ret; } // caller hold task_list_ lock int ObPartGroupMigrationTask::get_migrate_task(const ObPartitionKey& pg_key, ObPartMigrationTask*& task) { int ret = OB_SUCCESS; task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!pg_key.is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "get migrate task get invalid argument", K(ret), K(pg_key)); } else { for (int64_t i = 0; NULL == task && i < task_list_.count(); ++i) { if (task_list_[i].arg_.key_ == pg_key) { task = &task_list_[i]; } } if (OB_ISNULL(task)) { ret = OB_ENTRY_NOT_EXIST; STORAGE_LOG(ERROR, "cannot found pkey in task", K(ret), K(pg_key), KP(this), K(*this)); } } return ret; } // TODO(muwei): fix it // int ObPartGroupMigrationTask::set_migrate_task_flags( // const ObPartitionKey &pg_key, // const ObMigrateStatus &status, // const bool is_restore) //{ // int ret = OB_SUCCESS; // ObPartMigrationTask *task = NULL; // if (!is_inited_) { // ret = OB_NOT_INIT; // STORAGE_LOG(WARN, "not inited", K(ret)); // } else if (!pg_key.is_valid() || status >= ObMigrateStatus::OB_MIGRATE_STATUS_MAX) { // ret = OB_INVALID_ARGUMENT; // STORAGE_LOG(WARN, "get migrate task get invalid argument", K(ret), // K(pg_key), K(status), K(is_restore)); // } else { // common::SpinWLockGuard guard(lock_); // if (OB_FAIL(get_migrate_task(pg_key, task))) { // STORAGE_LOG(WARN, "failed to get migrate task", K(ret), K(pg_key)); // } else if (OB_FAIL(set_migrate_task_flags_(status, is_restore, *task))) { // STORAGE_LOG(WARN, "failed to set migrate task flags", K(ret), K(pg_key), K(status)); // } // } // return ret; //} int ObPartGroupMigrationTask::init_restore_meta_index_(const share::ObPhysicalRestoreArg& restore_arg) { int ret = OB_SUCCESS; const bool is_compat_backup_path = !restore_arg.restore_info_.is_compat_backup_path(); const int64_t compatible = restore_arg.restore_info_.compatible_; bool need_check_compeleted = true; const bool need_check_all_meta_files = false; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!restore_arg.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("restore arg is invlaid, pelase check", K(ret), K(restore_arg)); } else if (FALSE_IT(need_check_compeleted = (restore_arg.restore_info_.cluster_version_ > CLUSTER_VERSION_2271))) { } else { if (is_compat_backup_path) { share::ObSimpleBackupSetPath simple_path; if (OB_FAIL(restore_arg.get_largest_backup_set_path(simple_path))) { LOG_WARN("failed to get largest backup set path", K(restore_arg)); ; } else if (OB_FAIL( meta_indexs_.init(simple_path, compatible, need_check_all_meta_files, need_check_compeleted))) { LOG_WARN("init restore meta index map fail", K(ret), K(restore_arg)); } } else { ObBackupBaseDataPathInfo path_info; if (OB_FAIL(restore_arg.get_backup_base_data_info(path_info))) { LOG_WARN("failed to get backup base data info", K(ret), K(restore_arg)); } else if (OB_FAIL(meta_indexs_.init(path_info, compatible, need_check_all_meta_files, need_check_compeleted))) { LOG_WARN("init restore meta index map fail", K(ret), K(restore_arg)); } } } return ret; } // caller hold task_list_ lock int ObPartGroupMigrationTask::set_migrate_task_flags_( const ObMigrateStatus& status, const bool is_restore, ObPartMigrationTask& task) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { task.need_reset_migrate_status_ = true; if (status == OB_MIGRATE_STATUS_ADD_FAIL || status == OB_MIGRATE_STATUS_MIGRATE_FAIL) { task.during_migrating_ = true; } if (ADD_REPLICA_OP == task.arg_.type_ && is_restore) { skip_change_member_list_ = true; STORAGE_LOG(INFO, "add replica op when restore skip set member list", "pkey", task.ctx_.replica_op_arg_.key_); } } return ret; } int ObGroupMigrateDag::report_result(const bool is_batch_mode, const ObIArray& report_list) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!is_batch_mode) { if (OB_FAIL(single_report_result(report_list))) { STORAGE_LOG(WARN, "failed to single_report_result", K(ret)); } } else { if (OB_FAIL(batch_report_result(report_list))) { STORAGE_LOG(WARN, "failed to batch_report_result", K(ret)); } } return ret; } int ObGroupMigrateDag::single_report_result(const ObIArray& report_list) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (report_list.count() != 1) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "cannot single report result more than 1", K(ret), K(report_list)); } else { const ObReportPartMigrationTask& task = report_list.at(0); if (OB_FAIL(partition_service_->retry_post_operate_replica_res(task.arg_, task.result_))) { STORAGE_LOG(WARN, "failed to retry_post_operate_replica_res", K(ret), K(task)); } } return ret; } int ObGroupMigrateDag::batch_report_result(const ObIArray& report_list) { int ret = OB_SUCCESS; ObArray report_res_list; ObPartMigrationRes tmp_res; ObReplicaOpType type; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (report_list.count() <= 0) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "report list must not empty", K(ret)); } else if (FALSE_IT(type = report_list.at(0).arg_.type_)) { } else if (OB_FAIL(ObMigrateUtil::get_report_result(report_list, report_res_list))) { LOG_WARN("failed to get report result", K(ret), K(report_list)); } else if (OB_FAIL(partition_service_->retry_post_batch_migrate_replica_res(type, report_res_list))) { STORAGE_LOG(WARN, "failed to retry_post_operate_replica_res", K(ret), K(type), K(report_res_list)); } return ret; } int ObPartGroupMigrationTask::get_tables_with_major_sstable(const ObPartitionKey& pkey, ObIArray& table_ids) { int ret = OB_SUCCESS; ObIPartitionGroupGuard part_guard; ObIPartitionGroup* partition = NULL; ObPartitionStorage* storage = NULL; ObPGPartitionGuard pg_partition_guard; ObTablesHandle tables_handle; table_ids.reset(); if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "ObPartGroupMigrationTask has not been inited", K(ret)); } else if (!pkey.is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid arguments", K(ret), K(pkey)); } else if (OB_FAIL(ObPartitionService::get_instance().get_partition(pkey, part_guard))) { STORAGE_LOG(WARN, "fail to get partition", K(ret), K(pkey)); } else if (OB_ISNULL(partition = part_guard.get_partition_group())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "partition must not be NULL", K(ret), K(pkey)); } else if (OB_FAIL(partition->get_pg_partition(pkey, pg_partition_guard))) { STORAGE_LOG(WARN, "get pg partition error", K(ret), K(pkey)); } else if (OB_ISNULL(pg_partition_guard.get_pg_partition())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "get pg partition error", K(ret), KP(pg_partition_guard.get_pg_partition())); } else if (OB_ISNULL( storage = static_cast(pg_partition_guard.get_pg_partition()->get_storage()))) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "partition storage must not be NULL", K(ret)); } else if (OB_FAIL(storage->get_partition_store().get_last_all_major_sstable(tables_handle))) { STORAGE_LOG(WARN, "fail to get last all major sstables", K(ret), K(pkey)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < tables_handle.get_count(); ++i) { ObITable* table = tables_handle.get_tables().at(i); if (OB_ISNULL(table)) { ret = OB_ERR_SYS; STORAGE_LOG(WARN, "error sys, table must not be NULL", K(ret)); } else if (OB_FAIL(table_ids.push_back(table->get_key().table_id_))) { STORAGE_LOG(WARN, "fail to push back table id", K(ret)); } } } return ret; } ObFastMigrateDag::ObFastMigrateDag() : ObIDag(ObIDagType::DAG_TYPE_MIGRATE, ObIDagPriority::DAG_PRIO_MIGRATE_MID), group_task_(nullptr), sub_type_(TaskType::INVALID) {} ObFastMigrateDag::~ObFastMigrateDag() { group_task_ = nullptr; } bool ObFastMigrateDag::operator==(const ObIDag& other) const { bool is_same = true; if (this == &other) { // same } else if (get_type() != other.get_type()) { is_same = false; } else { const ObFastMigrateDag& other_dag = static_cast(other); is_same = (group_task_ == other_dag.group_task_ && sub_type_ == other_dag.sub_type_); } return is_same; } int64_t ObFastMigrateDag::hash() const { int64_t hash_value = 0; if (nullptr != group_task_) { hash_value = common::murmurhash(&group_task_, sizeof(group_task_), hash_value); } if (is_valid_task_type(sub_type_)) { hash_value = common::murmurhash(&sub_type_, sizeof(sub_type_), hash_value); } return hash_value; } int64_t ObFastMigrateDag::get_tenant_id() const { int64_t tenant_id = OB_INVALID_TENANT_ID; if (nullptr != group_task_) { tenant_id = group_task_->get_tenant_id(); } return tenant_id; } int ObFastMigrateDag::fill_comment(char* buf, const int64_t buf_len) const { int ret = OB_SUCCESS; if (NULL == buf || buf_len <= 0) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(buf), K(buf_len)); } else { const char* type_str = "UNKOWN"; static const char* TYPE_STRING[] = { "INVALID", "SUSPEND_SRC", "HANDOVER_PG", "CLEAN_UP", "REPORT_META_TABLE", }; STATIC_ASSERT(ARRAYSIZEOF(TYPE_STRING) == TaskType::MAX_TYPE, "TYPE_STRING count mismatch with task type count"); if (is_valid_task_type(sub_type_)) { type_str = TYPE_STRING[sub_type_]; } int64_t pos = 0; if (OB_FAIL(databuff_printf(buf, buf_len, pos, "%s", "fast migrate batch exec: "))) { LOG_WARN("generate comment failed", K(ret)); } else if (OB_FAIL(databuff_print_json_kv(buf, buf_len, pos, "type", type_str))) { LOG_WARN("generate sub_type failed", K(ret)); } else if (OB_FAIL(databuff_print_json_kv_comma(buf, buf_len, pos, K(group_task_)))) { LOG_WARN("generate group task addr failed", K(ret)); } else if (nullptr != group_task_ && OB_FAIL(databuff_print_json_kv_comma(buf, buf_len, pos, "trace_id", group_task_->get_task_id()))) { LOG_WARN("generate trace id failed", K(ret)); } else if (OB_FAIL( databuff_print_json_kv_comma(buf, buf_len, pos, "running_task_count", get_running_task_count()))) { LOG_WARN("generate group task addr failed", K(ret)); } } return ret; } int64_t ObFastMigrateDag::get_compat_mode() const { return static_cast(share::ObWorker::CompatMode::MYSQL); } bool ObFastMigrateDag::is_valid_task_type(TaskType type) { return type > TaskType::INVALID && type < TaskType::MAX_TYPE; } int ObFastMigrateDag::init(ObPartGroupMigrationTask* group_task, TaskType sub_type) { int ret = OB_SUCCESS; if (OB_NOT_NULL(group_task_)) { ret = OB_INIT_TWICE; LOG_WARN("init twice", K(ret), K(group_task_)); } else if (OB_ISNULL(group_task) || !is_valid_task_type(sub_type)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid argument", K(ret), KP(group_task), K(sub_type)); } else { group_task_ = group_task; sub_type_ = sub_type; } return ret; } ObBaseMigrateDag::ObBaseMigrateDag(const ObIDagType type, const ObIDagPriority priority) : ObIDag(type, priority), is_inited_(false), ctx_(NULL), tenant_id_(0), compat_mode_() {} ObBaseMigrateDag::~ObBaseMigrateDag() {} bool ObBaseMigrateDag::operator==(const ObIDag& other) const { bool is_same = true; if (this == &other) { // same } else if (get_type() != other.get_type()) { is_same = false; } else { const ObMigrateDag& other_dag = static_cast(other); if (NULL != ctx_ && NULL != other_dag.ctx_) { if (ctx_->replica_op_arg_.key_ != other_dag.ctx_->replica_op_arg_.key_) { is_same = false; } } } return is_same; } int64_t ObBaseMigrateDag::hash() const { int64_t hash_value = 0; if (NULL != ctx_) { hash_value = common::murmurhash(&ctx_->replica_op_arg_.key_, sizeof(ctx_->replica_op_arg_.key_), hash_value); } return hash_value; } void ObBaseMigrateDag::clear() { ctx_ = NULL; is_inited_ = false; tenant_id_ = 0; } int ObBaseMigrateDag::fill_comment(char* buf, const int64_t buf_len) const { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(ctx_->fill_comment(buf, buf_len))) { LOG_WARN("failed to fill comment", K(ret), K(*ctx_)); } return ret; } ObMigrateDag::ObMigrateDag() : ObBaseMigrateDag(ObIDag::DAG_TYPE_MIGRATE, ObIDag::DAG_PRIO_MIGRATE_HIGH) {} ObMigrateDag::~ObMigrateDag() { int tmp_ret = OB_SUCCESS; ObPartitionService* partition_service = MIGRATOR.get_partition_service(); ObMigrateStatus migrate_status = OB_MIGRATE_STATUS_NONE; bool need_offline_partition = false; if (NULL != ctx_) { ctx_->calc_need_retry(); ObReplicaOpArg tmp_relica_op_arg = ctx_->replica_op_arg_; ObIPartitionGroup* partition = ctx_->get_partition(); if (OB_SUCCESS == ctx_->result_ && OB_SUCCESS != this->get_dag_ret()) { ctx_->result_ = this->get_dag_ret(); LOG_WARN("migrate dag is failed", "result", this->get_dag_ret(), K(*ctx_)); } if (OB_SUCCESS == ctx_->result_ && ctx_->need_rebuild_) { // A state that should not exist tmp_ret = OB_ERR_UNEXPECTED; LOG_ERROR("result success no need rebuild", K(tmp_ret), K(*ctx_)); ctx_->result_ = tmp_ret; ctx_->need_rebuild_ = false; } if (OB_SUCCESS != ctx_->result_ && NULL != partition_service && NULL != partition) { if (OB_SUCCESS != (tmp_ret = ctx_->get_partition()->get_pg_storage().get_pg_migrate_status(migrate_status))) { LOG_WARN("failed to get_migrate_status", K(tmp_ret), K(*ctx_)); } else if (OB_MIGRATE_STATUS_ADD == migrate_status || OB_MIGRATE_STATUS_MIGRATE == migrate_status) { if (OB_NOT_EXIST_REPLICA == ctx_->replica_state_ && !partition_service->is_partition_exist(ctx_->replica_op_arg_.key_)) { if (OB_SUCCESS != (tmp_ret = partition_service->get_election_mgr()->stop_partition(ctx_->replica_op_arg_.key_))) { STORAGE_LOG(WARN, "failed to stop partition in election mgr", K(tmp_ret), K(*ctx_)); } } else if (OB_NORMAL_REPLICA != ctx_->replica_state_) { need_offline_partition = true; if (ctx_->need_rebuild_) { STORAGE_LOG(INFO, "rebuild offline partition without state switch", K(*ctx_)); if (OB_SUCCESS != (tmp_ret = partition->pause())) { STORAGE_LOG(WARN, "fail to pause partition", K(tmp_ret), K(*ctx_)); } } else if (FAST_MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_ && ctx_->is_takeover_finished_) { // do nothing, fast_migrate_mgr will retry LOG_INFO("fast migrate: keep pg when takeover finished", K(ctx_->replica_op_arg_.key_)); } else { // If the migration and replication of partition fail, // offline opertion is performed on the partition and the status is rolled backup to offline STORAGE_LOG(INFO, "rollback partition when migrate fail", K(*ctx_)); ObPartitionState pstate = partition->get_partition_state(); if (ObPartitionState::INIT == pstate) { if (OB_SUCCESS != (tmp_ret = partition->set_valid())) { STORAGE_LOG(WARN, "failed to set partition to valid", K(tmp_ret), KPC(ctx_)); } } else if (OB_SUCCESS != (tmp_ret = partition->offline())) { STORAGE_LOG(WARN, "fail to offline partition", K(tmp_ret), K(*ctx_)); } } } else { // For the rebuild replica opertion, if it fails, it will not be handled temporarily. // The reasion is that there may be trasactions in the rebuild replica state that have partitipated in the // voting. If it is handled, the majority will not be established. } } else if (OB_MIGRATE_STATUS_REBUILD == migrate_status) { if (OB_SUCCESS != (tmp_ret = online_for_rebuild())) { LOG_WARN("failed to online_for_rebuild", K(tmp_ret), K(*ctx_)); } } else if (OB_MIGRATE_STATUS_RESTORE == migrate_status) { if (OB_RESTORE_PARTITION_IS_COMPELETE == ctx_->result_) { LOG_INFO("restore partition compelete", K(*ctx_)); ctx_->need_rebuild_ = false; ctx_->result_ = OB_SUCCESS; } } } ctx_->finish_ts_ = ObTimeUtility::current_time(); if (!ctx_->need_rebuild_) { ctx_->action_ = ObMigrateCtx::END; if (OB_SUCCESS == ctx_->result_ && VALIDATE_BACKUP_OP != ctx_->replica_op_arg_.type_) { ctx_->result_ = update_partition_meta_for_restore(); } else if (RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_) { // this is for restore standby, cause standby need receive log to retry if (OB_SUCCESS != (tmp_ret = online_for_restore())) { STORAGE_LOG(WARN, "failed to online for restore", K(tmp_ret), K(ctx_->replica_op_arg_)); } } if (OB_SUCCESS == ctx_->result_) { ctx_->during_migrating_ = false; } if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } if (OB_SUCCESS != (tmp_ret = ctx_->notice_finish_part_task())) { STORAGE_LOG(WARN, "failed to notice_finish_part_task", K(tmp_ret), K(tmp_relica_op_arg)); } } else { if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } LOG_INFO("migrate retry", "replica_op_arg", ctx_->replica_op_arg_, "rebuild_count", ctx_->rebuild_count_); int tmp_result = ctx_->result_; ctx_->rebuild_migrate_ctx(); // gereate_and_schedule_migrate_dag is successful, you should not access ctx content, // and there will be concurrency problems if (OB_SUCCESS != (tmp_ret = ctx_->generate_and_schedule_migrate_dag())) { LOG_WARN("failed to schedule migrate dag", K(tmp_ret), K(*ctx_)); ctx_->result_ = tmp_result; // set error state before rebuild ctx_->need_rebuild_ = false; ctx_->action_ = ObMigrateCtx::END; if (need_offline_partition) { STORAGE_LOG(INFO, "rollback partition when migrate fail", K(*ctx_)); if (OB_SUCCESS != (tmp_ret = partition->offline())) { STORAGE_LOG(WARN, "fail to offline partition", K(tmp_ret), K(*ctx_)); } } if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } if (OB_SUCCESS != (tmp_ret = ctx_->notice_finish_part_task())) { STORAGE_LOG(WARN, "failed to notice_finish_part_task", K(tmp_ret), K(tmp_relica_op_arg)); } } } // need rebuild } // ctx != NULL } int ObMigrateDag::init(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (!ctx.replica_op_arg_.is_valid() || OB_ISNULL(ctx.group_task_)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(ctx.group_task_), K(ctx.replica_op_arg_)); } else { ObReplicaRestoreStatus restore_status = ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; tenant_id_ = ctx.replica_op_arg_.key_.get_tenant_id(); if (ctx.replica_op_arg_.is_physical_restore_leader()) { if (OB_FAIL(init_for_restore_(ctx))) { LOG_WARN("failed to init for restore", K(ret)); } } else if (RESTORE_REPLICA_OP == ctx.replica_op_arg_.type_) { if (OB_FAIL(ctx.restore_info_.init(ctx.replica_op_arg_.restore_arg_))) { STORAGE_LOG(WARN, "failed to init restore info", K(ret)); } } else if (VALIDATE_BACKUP_OP == ctx.replica_op_arg_.type_) { ObBackupBaseDataPathInfo path_info; if (OB_FAIL(ctx.replica_op_arg_.validate_arg_.get_backup_base_data_info(path_info))) { STORAGE_LOG(WARN, "failed to get backup base data info", K(ret)); } else if (!ctx.macro_index_store_.is_inited() && OB_FAIL(ctx.macro_index_store_.init(path_info))) { STORAGE_LOG(WARN, "failed to init macro index store", K(ret), K(path_info)); } } if (OB_SUCC(ret)) { if (ObReplicaOpPriority::PRIO_LOW == ctx.replica_op_arg_.priority_) { set_priority(ObIDag::DAG_PRIO_MIGRATE_LOW); } else if (ObReplicaOpPriority::PRIO_MID == ctx.replica_op_arg_.priority_) { set_priority(ObIDag::DAG_PRIO_MIGRATE_MID); } else { set_priority(ObIDag::DAG_PRIO_MIGRATE_HIGH); } if (OB_FAIL(ctx.notice_start_part_task())) { LOG_WARN("failed to notice_start_part_task", K(ret)); ctx.result_ = ret; } else if (OB_FAIL(get_compat_mode_with_table_id(ctx.replica_op_arg_.key_.table_id_, compat_mode_))) { LOG_WARN("failed to get table compat mode", K(ret), K(ctx)); } else { ctx_ = &ctx; is_inited_ = true; STORAGE_LOG(INFO, "succeed init migrate dag"); } } } return ret; } int ObMigrateDag::init_for_restore_(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; ObReplicaRestoreStatus restore_status = ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; ObIPartitionGroupGuard guard; ObIPartitionGroup* partition = NULL; ObPartGroupMigrationTask* group_task = NULL; ObPGKey backup_pg_key; ObBackupMetaIndex meta_index; const int64_t compatible = ctx.replica_op_arg_.phy_restore_arg_.restore_info_.compatible_; if (OB_FAIL(ObPartitionService::get_instance().get_partition(ctx.replica_op_arg_.key_, guard))) { LOG_WARN("failed to get physical restore partition", K(ret), "pkey", ctx.replica_op_arg_.key_); } else if (OB_ISNULL(partition = guard.get_partition_group())) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "partition must not null", K(ret), K(ctx.replica_op_arg_)); } else if (FALSE_IT(restore_status = static_cast(partition->get_pg_storage().get_restore_state()))) { } else if (ObReplicaRestoreStatus::REPLICA_RESTORE_ARCHIVE_DATA == restore_status) { STORAGE_LOG(INFO, "restore status is replica restore archive data, no need init for restore", "pkey", ctx.replica_op_arg_.key_); } else if (OB_FAIL(ctx.replica_op_arg_.phy_restore_arg_.get_backup_pgkey(backup_pg_key))) { STORAGE_LOG(WARN, "failed to get backup pgkey", K(ret), K(ctx.replica_op_arg_.phy_restore_arg_)); } else if (FALSE_IT(group_task = reinterpret_cast(ctx.group_task_))) { } else if (OB_ISNULL(ctx.macro_indexs_)) { if (OB_BACKUP_COMPATIBLE_VERSION_V3 == compatible) { const ObBackupMetaType meta_type = ObBackupMetaType::PARTITION_GROUP_META_INFO; ObPhyRestoreMacroIndexStoreV2* phy_restore_macro_index_v2 = NULL; if (OB_ISNULL(phy_restore_macro_index_v2 = MIGRATOR.get_cp_fty()->get_phy_restore_macro_index_v2())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("phy restore macro index should not be NULL", K(ret), KP(phy_restore_macro_index_v2)); } else if (OB_FAIL(group_task->get_meta_indexs().get_meta_index(backup_pg_key, meta_type, meta_index))) { STORAGE_LOG(WARN, "failed to get meta index", K(ret), K(backup_pg_key), K(meta_index)); } else if (OB_FAIL(phy_restore_macro_index_v2->init( meta_index.task_id_, ctx.replica_op_arg_.phy_restore_arg_, restore_status))) { LOG_WARN("failed to int physcial restore macro index", K(ret)); } else { ctx.macro_indexs_ = phy_restore_macro_index_v2; phy_restore_macro_index_v2 = NULL; } if (OB_NOT_NULL(phy_restore_macro_index_v2)) { MIGRATOR.get_cp_fty()->free(phy_restore_macro_index_v2); ctx.macro_indexs_ = NULL; } } else if (OB_BACKUP_COMPATIBLE_VERSION_V1 == compatible || OB_BACKUP_COMPATIBLE_VERSION_V2 == compatible) { ObPhyRestoreMacroIndexStore* phy_restore_macro_index = NULL; if (OB_ISNULL(phy_restore_macro_index = MIGRATOR.get_cp_fty()->get_phy_restore_macro_index())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("phy restore macro index should not be NULL", K(ret), KP(phy_restore_macro_index)); } else if (OB_FAIL(phy_restore_macro_index->init(ctx.replica_op_arg_.phy_restore_arg_, restore_status))) { LOG_WARN("failed to int physcial restore macro index", K(ret)); } else { ctx.macro_indexs_ = phy_restore_macro_index; phy_restore_macro_index = NULL; } if (OB_NOT_NULL(phy_restore_macro_index)) { MIGRATOR.get_cp_fty()->free(phy_restore_macro_index); ctx.macro_indexs_ = NULL; } } else { ret = OB_ERR_UNEXPECTED; LOG_WARN("restore get invalid compatible version", K(ret), K(compatible)); } } else if (!ctx.macro_indexs_->is_inited()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("macro index is not NULL but not init", K(ret), KP(ctx.macro_indexs_)); } return ret; } int ObMigrateDag::update_partition_meta_for_restore() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; bool need_set_restore = true; // If it is a phsical restore, is_restore will set REPLICA_RESTORE_LOG, if not, directly chagne to // REPLICA_NOT_RESTORE. int16_t is_restore = ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; int16_t restore_state = ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "migrate dag do not init", K(ret)); } else if (OB_ISNULL(partition = ctx_->get_partition())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "partition must not null", K(ret), K(*ctx_)); } else if ((RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_ || RESTORE_FOLLOWER_REPLICA_OP == ctx_->replica_op_arg_.type_ || RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_) && OB_SUCC(ctx_->result_)) { const ObPartitionKey& pgkey = ctx_->replica_op_arg_.key_; if (ctx_->replica_op_arg_.is_physical_restore() || ctx_->replica_op_arg_.is_standby_restore()) { is_restore = ctx_->replica_op_arg_.is_physical_restore() ? ObReplicaRestoreStatus::REPLICA_RESTORE_LOG : ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; ObSavedStorageInfoV2 info; ObBaseStorageInfo clog_info; bool has_memtable = partition->get_pg_storage().has_memstore(); bool is_offline = true; restore_state = partition->get_pg_storage().get_restore_state(); if (REPLICA_RESTORE_DATA != restore_state && REPLICA_RESTORE_ARCHIVE_DATA != restore_state && REPLICA_RESTORE_STANDBY != restore_state && REPLICA_RESTORE_CUT_DATA != restore_state && REPLICA_RESTORE_STANDBY_CUT != restore_state) { need_set_restore = false; STORAGE_LOG(ERROR, "physical restore state is unexpected", K(pgkey), K(restore_state)); } if (OB_SUCC(ret) && OB_FAIL(partition->get_log_service()->is_offline(is_offline))) { STORAGE_LOG(WARN, "get log service offline stat fail", K(pgkey)); } // check status, restore finish, no memtable or offline if (OB_SUCC(ret) && !need_set_restore && (!has_memtable || is_offline)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "pg has restore finish, but memtable or log service is not ready", K(ret), K(need_set_restore), K(restore_state), K(has_memtable), K(is_offline)); } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_RESTORE_UPDATE_PARTITION_META_FAILED) OB_SUCCESS; if (-9001 == ret) { STORAGE_LOG(INFO, "inject restore fail before create_memtable", K(ret), K(pgkey)); } else { ret = OB_SUCCESS; } } #endif if (OB_SUCC(ret) && ctx_->pg_meta_.storage_info_.get_data_info().get_schema_version() == 0) { ObDataStorageInfo cur_data_info; if (OB_FAIL(partition->get_saved_data_info(cur_data_info))) { LOG_WARN("failed to get saved data info", K(ret)); } else { ctx_->pg_meta_.storage_info_.get_data_info().set_schema_version(cur_data_info.get_schema_version()); } } if (OB_SUCC(ret) && !has_memtable) { if (REPLICA_RESTORE_CUT_DATA == restore_state || REPLICA_RESTORE_STANDBY_CUT == restore_state) { // do nothing } else if (OB_FAIL(partition->set_storage_info(ctx_->pg_meta_.storage_info_))) { STORAGE_LOG(WARN, "failed to set storage info", K(ret), K(pgkey)); } if (OB_FAIL(ret)) { } else if (OB_FAIL(partition->create_memtable())) { STORAGE_LOG(WARN, "failed to create memtable", K(ret), K(pgkey)); } } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_RESTORE_UPDATE_PARTITION_META_FAILED) OB_SUCCESS; if (-9002 == ret) { STORAGE_LOG(INFO, "inject restore fail before log online", K(ret), K(pgkey)); } else { ret = OB_SUCCESS; } } #endif if (OB_SUCC(ret) && is_offline) { if (OB_FAIL(online_for_restore())) { LOG_WARN("failed to online for restore", K(ret)); } } STORAGE_LOG(INFO, "physical restore update partition meta finish", K(pgkey), K(restore_state), K(has_memtable), K(is_offline), K(info)); } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_RESTORE_UPDATE_PARTITION_META_FAILED) OB_SUCCESS; if (-9003 == ret) { STORAGE_LOG(INFO, "inject restore fail before set_restore_flag", K(ret), K(pgkey)); } else { ret = OB_SUCCESS; } } #endif if (OB_FAIL(ret) || !need_set_restore) { } else if (!partition->get_pg_storage().is_restore()) { ret = OB_RESTORE_PARTITION_TWICE; STORAGE_LOG(WARN, "partition restore twice", K(ret), K(pgkey)); } else if (REPLICA_RESTORE_STANDBY_CUT == restore_state) { const int16_t new_restore_state = ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; if (OB_FAIL(partition->get_pg_storage().set_restore_flag(new_restore_state, OB_INVALID_TIMESTAMP /*not update restore_snapshot_version */, OB_INVALID_TIMESTAMP /*not update restore_schema_version*/))) { LOG_WARN("failed to set restore flag", K(ret), K(*ctx_)); } } else if (OB_FAIL(ObPartitionService::get_instance().set_restore_flag(pgkey, is_restore))) { STORAGE_LOG(WARN, "fail to set partition store meta", K(ret)); } // If it is physical restore, set_archive_restoring will be setted in set_restore_flag function if (OB_FAIL(ret)) { int tmp_ret = OB_SUCCESS; if (OB_SUCCESS != (tmp_ret = partition->get_pg_storage().clear_all_memtables())) { LOG_WARN("failed to clear all memtables", K(tmp_ret)); } } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_RESTORE_UPDATE_PARTITION_META_FAILED) OB_SUCCESS; if (-9004 == ret) { STORAGE_LOG(INFO, "inject restore fail before report to rs", K(ret), K(pgkey)); } else { ret = OB_SUCCESS; } } #endif } return ret; } int ObMigrateDag::online_for_rebuild() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObSavedStorageInfoV2 info; ObBaseStorageInfo& clog_info = info.get_clog_info(); ObDataStorageInfo& data_info = info.get_data_info(); if (NULL != ctx_ && ctx_->need_online_for_rebuild_) { ObIPartitionGroup* partition = ctx_->get_partition(); int64_t restore_snapshot_version = OB_INVALID_TIMESTAMP; uint64_t last_restore_log_id = OB_INVALID_ID; int64_t last_restore_log_ts = OB_INVALID_TIMESTAMP; if (NULL == partition) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret), K(*ctx_)); } else if (partition->is_replica_using_remote_memstore()) { // do nothing, cause data replica no need offline } else if (OB_FAIL(partition->get_all_saved_info(info))) { STORAGE_LOG(WARN, "failed to get saved info", K(ret)); } else if (OB_FAIL(partition->create_memtable())) { LOG_WARN("failed to create memtable", K(ret), "pkey", partition->get_partition_key()); } else if (OB_FAIL(partition->get_log_service()->migrate_set_base_storage_info(clog_info))) { STORAGE_LOG(WARN, "reset clog start point fail", K(ret), K(info)); } else if (OB_FAIL(partition->get_pg_storage().get_restore_replay_info( last_restore_log_id, last_restore_log_ts, restore_snapshot_version))) { LOG_WARN("failed to get_restore_replay_info", KR(ret), "pkey", partition->get_partition_key()); } else if (OB_FAIL(MIGRATOR.get_partition_service()->online_partition(ctx_->replica_op_arg_.key_, data_info.get_publish_version(), restore_snapshot_version, last_restore_log_id, last_restore_log_ts))) { STORAGE_LOG(WARN, "online partition failed", K(ret), K(info), K(restore_snapshot_version), K(last_restore_log_id), K(last_restore_log_ts), K(*ctx_)); } else if (OB_FAIL(ctx_->get_partition()->get_log_service()->restore_replayed_log(clog_info))) { STORAGE_LOG(WARN, "restore and replay log failed.", K(ret), K_(ctx_->replica_op_arg_.key), K(info)); } else if (OB_FAIL(ctx_->get_partition()->get_log_service()->set_online( clog_info, ObVersion(0, 0) /*freeze version*/))) { STORAGE_LOG(WARN, "reset log temporary status failed.", K(ret), K_(ctx_->replica_op_arg_.key), K(info)); } else if (OB_FAIL(ctx_->get_partition()->report_clog_history_online())) { LOG_WARN("failed to report clog history online", K(ret)); } else { STORAGE_LOG(INFO, "succeed online_for_rebuild", K(ret), "pkey", ctx_->replica_op_arg_.key_, K(info)); if (OB_SUCCESS != (tmp_ret = MIGRATOR.get_partition_service()->turn_off_rebuild_flag(ctx_->replica_op_arg_))) { LOG_WARN("Failed to report_rebuild_replica off", K(tmp_ret), "arg", ctx_->replica_op_arg_); } } } return ret; } int ObMigrateDag::online_for_restore() { int ret = OB_SUCCESS; bool is_offline = true; ObSavedStorageInfoV2 info; ObBaseStorageInfo clog_info; ObIPartitionGroup* partition = NULL; if (OB_ISNULL(ctx_)) { // do nothing } else if (RESTORE_STANDBY_OP != ctx_->replica_op_arg_.type_ && RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && RESTORE_FOLLOWER_REPLICA_OP != ctx_->replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "online for restore replica op type is unexpected", K(ret), K(ctx_->replica_op_arg_)); } else if (OB_ISNULL(partition = ctx_->get_partition())) { // do nothing } else if (OB_FAIL(partition->get_log_service()->is_offline(is_offline))) { STORAGE_LOG(WARN, "get log service offline stat fail"); } else if (!is_offline) { STORAGE_LOG(INFO, "partition standby restore is online", K(partition->get_partition_key())); } else if (OB_FAIL(partition->reset_for_replay())) { LOG_WARN("failed to reset partition for replay", K(ret), K(partition->get_partition_key())); } else if (OB_FAIL(partition->get_all_saved_info(info))) { STORAGE_LOG(WARN, "failed to get saved info", K(ret)); } else if (OB_FAIL(partition->get_saved_clog_info(clog_info))) { LOG_WARN("failed to get saved clog info", K(ret)); } else if (OB_FAIL(partition->get_log_service()->migrate_set_base_storage_info(clog_info))) { STORAGE_LOG(WARN, "migrate_set_base_storage_info fail", K(ret), K(clog_info)); } else if (OB_FAIL(partition->get_log_service()->set_online(clog_info, ObVersion(0, 0) /*freeze version*/))) { STORAGE_LOG(WARN, "set_online fail", K(ret), K(clog_info)); } else if (OB_FAIL(partition->report_clog_history_online())) { STORAGE_LOG(WARN, "failed to report clog history online", K(ret), K(partition->get_partition_key())); } else { STORAGE_LOG( INFO, "partition restore online succ", K(partition->get_partition_key()), K(ctx_->replica_op_arg_.type_)); } return ret; } ObBackupDag::ObBackupDag() : ObBaseMigrateDag(ObIDag::DAG_TYPE_BACKUP, ObIDag::DAG_PRIO_BACKUP), backup_data_type_() {} ObBackupDag::~ObBackupDag() { int tmp_ret = OB_SUCCESS; if (NULL != ctx_) { ctx_->calc_need_retry(); ObReplicaOpArg tmp_relica_op_arg = ctx_->replica_op_arg_; ObIPartitionGroup* partition = ctx_->get_partition(); if (OB_SUCCESS == ctx_->result_ && OB_SUCCESS != this->get_dag_ret()) { ctx_->result_ = this->get_dag_ret(); LOG_WARN("migrate dag is failed", "result", this->get_dag_ret(), K(*ctx_)); } if (OB_SUCCESS == ctx_->result_ && ctx_->need_rebuild_) { // Status should not exist status tmp_ret = OB_ERR_UNEXPECTED; LOG_ERROR("result success no need rebuild", K(tmp_ret), K(*ctx_)); ctx_->result_ = tmp_ret; ctx_->need_rebuild_ = false; } ctx_->finish_ts_ = ObTimeUtility::current_time(); if (!ctx_->need_rebuild_) { ctx_->action_ = ObMigrateCtx::END; if (OB_SUCCESS == ctx_->result_) { ctx_->during_migrating_ = false; } if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } if (OB_SUCCESS != (tmp_ret = ctx_->notice_finish_part_task())) { STORAGE_LOG(WARN, "failed to notice_finish_part_task", K(tmp_ret), K(tmp_relica_op_arg)); } } else { if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } LOG_INFO("backup retry", "replica_op_arg", ctx_->replica_op_arg_, "rebuild_count", ctx_->rebuild_count_); int tmp_result = ctx_->result_; ctx_->rebuild_migrate_ctx(); // It has concurrency problems when you use ctx after generate_and_scheduler_migrate_dag successful if (OB_SUCCESS != (tmp_ret = ctx_->generate_and_schedule_backup_dag(backup_data_type_))) { LOG_WARN("failed to schedule migrate dag", K(tmp_ret), K(*ctx_)); ctx_->result_ = tmp_result; // set error code before rebuild ctx_->need_rebuild_ = false; ctx_->action_ = ObMigrateCtx::END; if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } if (OB_SUCCESS != (tmp_ret = ctx_->notice_finish_part_task())) { STORAGE_LOG(WARN, "failed to notice_finish_part_task", K(tmp_ret), K(tmp_relica_op_arg)); } } } // need rebuild } // ctx != NULL } int ObBackupDag::init(const ObBackupDataType& backup_data_type, ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (!ctx.replica_op_arg_.is_valid() || OB_ISNULL(ctx.group_task_) || !backup_data_type.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(ctx.group_task_), K(ctx.replica_op_arg_), K(backup_data_type)); } else if (BACKUP_REPLICA_OP != ctx.replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "replica op type is not backup", K(ret), K(ctx.replica_op_arg_)); } else { tenant_id_ = ctx.replica_op_arg_.key_.get_tenant_id(); if (OB_FAIL(ctx.physical_backup_ctx_.init(*ObPartitionMigrator::get_instance().get_bandwidth_throttle(), ctx.replica_op_arg_.backup_arg_, ctx.replica_op_arg_.key_))) { LOG_WARN("failed to init physical backup ctx", K(ret), K(ctx.replica_op_arg_.backup_arg_)); } else if (ctx.replica_op_arg_.backup_arg_.is_incremental_backup() && OB_ISNULL(ctx.macro_indexs_)) { ObPhyRestoreMacroIndexStoreV2* phy_restore_macro_index_v2 = NULL; if (OB_ISNULL(phy_restore_macro_index_v2 = MIGRATOR.get_cp_fty()->get_phy_restore_macro_index_v2())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("phy restore macro index should not be NULL", K(ret), KP(phy_restore_macro_index_v2)); } else if (OB_FAIL(phy_restore_macro_index_v2->init(ctx.replica_op_arg_.backup_arg_.task_id_, ctx.replica_op_arg_.key_, ctx.replica_op_arg_.backup_arg_, backup_data_type))) { LOG_WARN("failed to int physcial restore macro index", K(ret)); } else { ctx.macro_indexs_ = phy_restore_macro_index_v2; phy_restore_macro_index_v2 = NULL; } if (OB_NOT_NULL(phy_restore_macro_index_v2)) { MIGRATOR.get_cp_fty()->free(phy_restore_macro_index_v2); ctx.macro_indexs_ = NULL; } } if (OB_SUCC(ret)) { set_priority(ObIDag::DAG_PRIO_BACKUP); if (OB_FAIL(ctx.notice_start_part_task())) { LOG_WARN("failed to notice_start_part_task", K(ret)); ctx.result_ = ret; } else if (OB_FAIL(get_compat_mode_with_table_id(ctx.replica_op_arg_.key_.table_id_, compat_mode_))) { LOG_WARN("failed to get table compat mode", K(ret), K(ctx)); } else { ctx_ = &ctx; backup_data_type_ = backup_data_type; is_inited_ = true; STORAGE_LOG(INFO, "succeed init backup dag"); } } } return ret; } ObValidateDag::ObValidateDag() : ObBaseMigrateDag(ObIDag::DAG_TYPE_VALIDATE, ObIDag::DAG_PRIO_VALIDATE) {} ObValidateDag::~ObValidateDag() { int tmp_ret = OB_SUCCESS; int ret = OB_SUCCESS; if (NULL != ctx_) { ctx_->calc_need_retry(); SMART_VAR(ObReplicaOpArg, tmp_relica_op_arg) { tmp_relica_op_arg = ctx_->replica_op_arg_; if (OB_SUCCESS == ctx_->result_ && OB_SUCCESS != this->get_dag_ret()) { ctx_->result_ = this->get_dag_ret(); LOG_WARN("migrate dag is failed", "result", this->get_dag_ret(), K(*ctx_)); } ctx_->action_ = ObMigrateCtx::END; ctx_->during_migrating_ = false; if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } ctx_->finish_ts_ = ObTimeUtility::current_time(); if (OB_SUCCESS != (tmp_ret = ctx_->notice_finish_part_task())) { STORAGE_LOG(WARN, "failed to notice_finish_part_task", K(tmp_ret), K(tmp_relica_op_arg)); } } } // ctx != NULL } bool ObValidateDag::operator==(const ObIDag& other) const { bool is_same = true; if (this == &other) { // same } else if (get_type() != other.get_type()) { is_same = false; } else { const ObValidateDag& other_dag = static_cast(other); if (NULL != ctx_ && NULL != other_dag.ctx_) { const ObPhysicalValidateArg& other_arg = other_dag.ctx_->replica_op_arg_.validate_arg_; const ObPhysicalValidateArg& arg = ctx_->replica_op_arg_.validate_arg_; is_same = arg.backup_set_id_ == other_arg.backup_set_id_ && arg.pg_key_ == other_arg.pg_key_; } else { is_same = false; } } return is_same; } int64_t ObValidateDag::hash() const { int64_t hash_value = 0; if (NULL != ctx_) { const ObPhysicalValidateArg& arg = ctx_->replica_op_arg_.validate_arg_; hash_value = common::murmurhash(&(arg.backup_set_id_), sizeof(arg.backup_set_id_), hash_value); hash_value = common::murmurhash(&(arg.pg_key_), sizeof(arg.pg_key_), hash_value); } return hash_value; } int ObValidateDag::init(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (!ctx.replica_op_arg_.is_valid() || OB_ISNULL(ctx.group_task_)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(ctx.group_task_), K(ctx.replica_op_arg_)); } else if (VALIDATE_BACKUP_OP != ctx.replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "replica op type is not validate", K(ret), K(ctx.replica_op_arg_)); } else { const ObPartitionKey& pkey = ctx.replica_op_arg_.key_; tenant_id_ = pkey.get_tenant_id(); set_priority(ObIDag::DAG_PRIO_VALIDATE); if (OB_FAIL(ctx.notice_start_part_task())) { LOG_WARN("failed to notice_start_part_task", K(ret)); ctx.result_ = ret; } else { ctx_ = &ctx; is_inited_ = true; STORAGE_LOG(INFO, "succeed init validate dag", K(pkey)); } } return ret; } ObBackupBackupsetDag::ObBackupBackupsetDag() : ObBaseMigrateDag(ObIDag::DAG_TYPE_BACKUP_BACKUPSET, ObIDag::DAG_PRIO_BACKUP) {} ObBackupBackupsetDag::~ObBackupBackupsetDag() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (OB_NOT_NULL(ctx_)) { ctx_->calc_need_retry(); SMART_VAR(ObReplicaOpArg, tmp_relica_op_arg) { tmp_relica_op_arg = ctx_->replica_op_arg_; if (OB_SUCCESS == ctx_->result_ && OB_SUCCESS != this->get_dag_ret()) { ctx_->result_ = this->get_dag_ret(); LOG_WARN("backup backupset dag is failed", "result", this->get_dag_ret(), K(*ctx_)); } ctx_->action_ = ObMigrateCtx::END; ctx_->during_migrating_ = false; if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } ctx_->finish_ts_ = ObTimeUtility::current_time(); if (OB_SUCCESS != (tmp_ret = ctx_->notice_finish_part_task())) { STORAGE_LOG(WARN, "failed to notice_finish_part_task", K(tmp_ret), K(tmp_relica_op_arg)); } } } } bool ObBackupBackupsetDag::operator==(const ObIDag& other) const { bool equal = false; if (OB_UNLIKELY(this == &other)) { equal = true; } else { if (get_type() == other.get_type()) { const ObBackupBackupsetDag& other_dag = static_cast(other); if (OB_NOT_NULL(ctx_) && OB_NOT_NULL(other_dag.ctx_)) { const share::ObBackupBackupsetArg& arg = ctx_->replica_op_arg_.backup_backupset_arg_; const share::ObBackupBackupsetArg& other_arg = other_dag.ctx_->replica_op_arg_.backup_backupset_arg_; if (arg.backup_set_id_ == other_arg.backup_set_id_ && arg.pg_key_ == other_arg.pg_key_) { equal = true; STORAGE_LOG( WARN, "two backup backupset dag is same", K(arg), K(other_arg), KP(this), KP(&other), K(*this), K(other)); } } } } return equal; } int64_t ObBackupBackupsetDag::hash() const { int64_t hash_value = 0; if (OB_NOT_NULL(ctx_)) { const share::ObBackupBackupsetArg& arg = ctx_->replica_op_arg_.backup_backupset_arg_; hash_value = common::murmurhash(&(arg.backup_set_id_), sizeof(arg.backup_set_id_), hash_value); hash_value = common::murmurhash(&(arg.pg_key_), sizeof(arg.pg_key_), hash_value); } return hash_value; } int ObBackupBackupsetDag::init(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (IS_INIT) { ret = OB_INIT_TWICE; STORAGE_LOG(ERROR, "cannot init twice", KR(ret)); } else if (!ctx.replica_op_arg_.is_valid() || OB_ISNULL(ctx.group_task_)) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", KR(ret), KP(ctx.group_task_), K(ctx.replica_op_arg_)); } else if (BACKUP_BACKUPSET_OP != ctx.replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "replica op type is not backup backupset", KR(ret), K(ctx.replica_op_arg_)); } else { const ObPartitionKey& pkey = ctx.replica_op_arg_.key_; tenant_id_ = pkey.get_tenant_id(); const share::ObBackupBackupsetArg& arg = ctx.replica_op_arg_.backup_backupset_arg_; if (OB_SUCC(ret)) { set_priority(ObIDag::DAG_PRIO_BACKUP); if (OB_FAIL(ctx.notice_start_part_task())) { STORAGE_LOG(WARN, "failed to notice_start_part_task", KR(ret)); ctx.result_ = ret; } else { ctx_ = &ctx; is_inited_ = true; STORAGE_LOG(INFO, "succeed init backup backupset dag", K(pkey)); } } } return ret; } ObBackupArchiveLogDag::ObBackupArchiveLogDag() : ObBaseMigrateDag(ObIDag::DAG_TYPE_BACKUP_ARCHIVELOG, ObIDag::DAG_PRIO_BACKUP) {} ObBackupArchiveLogDag::~ObBackupArchiveLogDag() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (OB_NOT_NULL(ctx_)) { ctx_->calc_need_retry(); SMART_VAR(ObReplicaOpArg, tmp_relica_op_arg) { tmp_relica_op_arg = ctx_->replica_op_arg_; if (OB_SUCCESS == ctx_->result_ && OB_SUCCESS != this->get_dag_ret()) { ctx_->result_ = this->get_dag_ret(); LOG_WARN("backup archivelog dag is failed", "result", this->get_dag_ret(), K(*ctx_)); } ctx_->action_ = ObMigrateCtx::END; ctx_->during_migrating_ = false; if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { STORAGE_LOG(WARN, "failed to update_partition_migration_status", K(tmp_ret)); } ctx_->finish_ts_ = ObTimeUtility::current_time(); if (OB_SUCCESS != (tmp_ret = ctx_->notice_finish_part_task())) { STORAGE_LOG(WARN, "failed to notice_finish_part_task", K(tmp_ret), K(tmp_relica_op_arg)); } } } } int ObBackupArchiveLogDag::init(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (IS_INIT) { ret = OB_INIT_TWICE; STORAGE_LOG(ERROR, "cannot init twice", KR(ret)); } else if (!ctx.replica_op_arg_.is_valid() || OB_ISNULL(ctx.group_task_)) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", KR(ret), KP(ctx.group_task_), K(ctx.replica_op_arg_)); } else if (BACKUP_ARCHIVELOG_OP != ctx.replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "replica op type is not backup archivelog", KR(ret), K(ctx.replica_op_arg_)); } else { const ObPartitionKey& pkey = ctx.replica_op_arg_.key_; tenant_id_ = pkey.get_tenant_id(); if (OB_SUCC(ret)) { set_priority(ObIDag::DAG_PRIO_BACKUP); if (OB_FAIL(ctx.notice_start_part_task())) { STORAGE_LOG(WARN, "failed to notice_start_part_task", KR(ret)); ctx.result_ = ret; } else { ctx_ = &ctx; is_inited_ = true; STORAGE_LOG(INFO, "succeed init backup archivelog dag", K(pkey)); } } } return ret; } ObMigratePrepareTask::ObMigratePrepareTask() : ObITask(TASK_TYPE_MIGRATE_PREPARE), is_inited_(false), ctx_(NULL), cp_fty_(NULL), rpc_(NULL), srv_rpc_proxy_(NULL), bandwidth_throttle_(NULL), partition_service_(NULL) {} ObMigratePrepareTask::~ObMigratePrepareTask() {} int ObMigratePrepareTask::init() { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (ObIDag::DAG_TYPE_MIGRATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP != dag_->get_type() && ObIDag::DAG_TYPE_VALIDATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_BACKUPSET != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_ARCHIVELOG != dag_->get_type()) { ret = OB_ERR_SYS; LOG_ERROR("dag type not match", K(ret), K(*dag_)); } else { ctx_ = static_cast(dag_)->get_ctx(); is_inited_ = true; bandwidth_throttle_ = MIGRATOR.get_bandwidth_throttle(); partition_service_ = MIGRATOR.get_partition_service(); srv_rpc_proxy_ = MIGRATOR.get_svr_rpc_proxy(); rpc_ = MIGRATOR.get_pts_rpc(); cp_fty_ = MIGRATOR.get_cp_fty(); } return ret; } int ObMigratePrepareTask::prepare_restore_reader_if_needed() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPartGroupMigrationTask* group_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_ || RESTORE_FOLLOWER_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_ISNULL(partition = ctx_->get_partition())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "Unexpected error, the partition is NULL ", K(ret)); } else if (!partition->get_pg_storage().is_restore()) { ret = OB_RESTORE_PARTITION_TWICE; STORAGE_LOG(WARN, "partition restore twice", K(ret), K(ctx_->replica_op_arg_.key_)); } else if (ctx_->replica_op_arg_.is_physical_restore_leader()) { if (OB_FAIL(prepare_restore_reader())) { LOG_WARN("failed to prepare restore reader", K(ret), "op_arg", ctx_->replica_op_arg_); } } else if (ctx_->replica_op_arg_.is_physical_restore_follower()) { ctx_->use_slave_safe_read_ts_ = false; } else { ret = OB_NOT_SUPPORTED; LOG_WARN("only support physical restore", K(ret), "op", ctx_->replica_op_arg_); } } return ret; } int ObMigratePrepareTask::prepare_restore_reader() { int ret = OB_SUCCESS; ObPartGroupMigrationTask* group_task = NULL; const int64_t compatible = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.compatible_; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_BACKUP_COMPATIBLE_VERSION_V3 == compatible) { ObPartitionGroupMetaRestoreReaderV2* restore_meta_reader_v2 = NULL; ObPhyRestoreMacroIndexStoreV2* macro_index = NULL; if (OB_ISNULL(ctx_->macro_indexs_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "macro indexs should not be NULL", K(ret), KP(ctx_->macro_indexs_)); } else if (ObIPhyRestoreMacroIndexStore::PHY_RESTORE_MACRO_INDEX_STORE_V2 != ctx_->macro_indexs_->get_type()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "macro indexs type is unexpected", K(ret), K(ctx_->macro_indexs_->get_type())); } else if (FALSE_IT(macro_index = reinterpret_cast(ctx_->macro_indexs_))) { } else if (OB_ISNULL(restore_meta_reader_v2 = cp_fty_->get_partition_group_meta_restore_reader_v2())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("meta reader should not be NULL", K(ret), KP(restore_meta_reader_v2)); } else if (FALSE_IT(group_task = reinterpret_cast(ctx_->group_task_))) { } else if (OB_FAIL(restore_meta_reader_v2->init(*bandwidth_throttle_, ctx_->replica_op_arg_.phy_restore_arg_, group_task->get_meta_indexs(), *macro_index))) { LOG_WARN("fail to init meta reader", K(ret), KP(restore_meta_reader_v2)); } else { ctx_->restore_meta_reader_ = restore_meta_reader_v2; restore_meta_reader_v2 = NULL; } if (NULL != cp_fty_) { if (NULL != restore_meta_reader_v2) { cp_fty_->free(restore_meta_reader_v2); } } } else if (OB_BACKUP_COMPATIBLE_VERSION_V1 == compatible || OB_BACKUP_COMPATIBLE_VERSION_V2 == compatible) { ObPartitionGroupMetaRestoreReaderV1* restore_meta_reader_v1 = NULL; ObPhyRestoreMacroIndexStore* macro_index = NULL; if (OB_ISNULL(ctx_->macro_indexs_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "macro indexs should not be NULL", K(ret), KP(ctx_->macro_indexs_)); } else if (ObIPhyRestoreMacroIndexStore::PHY_RESTORE_MACRO_INDEX_STORE_V1 != ctx_->macro_indexs_->get_type()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "macro indexs type is unexpected", K(ret), K(ctx_->macro_indexs_->get_type())); } else if (FALSE_IT(macro_index = reinterpret_cast(ctx_->macro_indexs_))) { } else if (OB_ISNULL(restore_meta_reader_v1 = cp_fty_->get_partition_group_meta_restore_reader_v1())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("meta reader should not be NULL", K(ret), KP(restore_meta_reader_v1)); } else if (FALSE_IT(group_task = reinterpret_cast(ctx_->group_task_))) { } else if (OB_FAIL(restore_meta_reader_v1->init(*bandwidth_throttle_, ctx_->replica_op_arg_.phy_restore_arg_, group_task->get_meta_indexs(), *macro_index))) { LOG_WARN("fail to init meta reader", K(ret), KP(restore_meta_reader_v1)); } else { ctx_->restore_meta_reader_ = restore_meta_reader_v1; restore_meta_reader_v1 = NULL; } if (NULL != cp_fty_) { if (NULL != restore_meta_reader_v1) { cp_fty_->free(restore_meta_reader_v1); } } } else { ret = OB_ERR_UNEXPECTED; LOG_WARN("compatible version is unexpected", K(ret), K(compatible)); } return ret; } // TODO() interface optimization for restore and rebuild int ObMigratePrepareTask::create_partition_if_needed( ObIPartitionGroupGuard& pg_guard, const ObPGPartitionStoreMeta& partition_meta) { int ret = OB_SUCCESS; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; STORAGE_LOG(ERROR, "not inited", K(ret)); } else if (OB_UNLIKELY(!partition_meta.is_valid())) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "partition_meta is not valid", K(ret), K(partition_meta)); } else { const ObPartitionKey& pg_key = ctx_->replica_op_arg_.key_; const ObPartitionKey& pkey = partition_meta.pkey_; ObIPartitionGroupGuard guard; ObIPartitionGroup* pg = NULL; int tmp_ret = OB_SUCCESS; if (OB_SUCCESS == (tmp_ret = ObPartitionService::get_instance().get_partition(pkey, guard))) { STORAGE_LOG(INFO, "partition exist, not need to create", K(ret), K(pkey)); } else if (OB_PARTITION_NOT_EXIST != tmp_ret) { ret = tmp_ret; STORAGE_LOG(WARN, "restore get partition fail", K(ret), K(pkey)); } else { obrpc::ObCreatePartitionArg arg; arg.schema_version_ = partition_meta.create_schema_version_; arg.lease_start_ = partition_meta.create_timestamp_; arg.restore_ = REPLICA_RESTORE_DATA; const bool is_replay = false; const uint64_t unused = 0; const bool in_slog_trans = false; ObTablesHandle sstables_handle; if (OB_ISNULL(pg = pg_guard.get_partition_group())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "partition group should not be null", K(ret), K(pg_key)); } else if (OB_FAIL(pg->create_pg_partition(pkey, partition_meta.multi_version_start_, partition_meta.data_table_id_, arg, in_slog_trans, is_replay, unused, sstables_handle))) { STORAGE_LOG(WARN, "fail to create pg partition", K(ret), K(partition_meta)); } else { sstables_handle.reset(); STORAGE_LOG( INFO, "succeed to create follower pg partition during physical restore", K(pkey), K(partition_meta)); } } } return ret; } int ObMigratePrepareTask::process() { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(prepare_migrate())) { LOG_WARN("failed to prepare migrate", K(ret)); } else if (OB_FAIL(generate_and_schedule_tasks())) { LOG_WARN("failed to generate and schedule tasks", K(ret)); } if (OB_FAIL(ret) && NULL != ctx_) { if (ctx_->is_need_retry(ret)) { ctx_->need_rebuild_ = true; LOG_INFO("migrate prepare task need retry", K(ret), K(*ctx_)); } ctx_->set_result_code(ret); } LOG_INFO("end ObMigratePrepareTask process", K(ret)); return ret; } int ObMigratePrepareTask::generate_and_schedule_tasks() { int ret = OB_SUCCESS; ObMigrateTransTableTaskGeneratorTask* migrate_trans_table_task_generator_task = NULL; ObMigrateTaskGeneratorTask* migrate_task_generator_task = NULL; ObMigrateRecoveryPointTaskGeneratorTask* recovery_point_task_generator_task = NULL; ObFakeTask* wait_trans_table_finish_task = NULL; ObFakeTask* wait_recovery_point_finish_task = NULL; // migrate_trans_table_task_generator_task -> wait_trans_table_finish_task // wait_trans_table_finish_task -> recovery_point_task_generator_task // wait_trans_table_finish_task -> wait_recovery_point_finish_task // recovery_point_task_generator_task -> wait_recovery_point_finish_task // wait_recovery_point_finish_task -> migrate_task_generator_task if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (OB_FAIL(dag_->alloc_task(wait_trans_table_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(dag_->alloc_task(migrate_trans_table_task_generator_task))) { LOG_WARN("failed to alloc migrate trans generator task", K(ret)); } else if (OB_FAIL(migrate_trans_table_task_generator_task->init(*wait_trans_table_finish_task))) { LOG_WARN("failed to init migrate trans table task generator task", K(ret)); } else if (OB_FAIL(add_child(*migrate_trans_table_task_generator_task))) { LOG_WARN("failed to add migrate trans table task generator task", K(ret)); } else if (OB_FAIL(dag_->add_task(*migrate_trans_table_task_generator_task))) { LOG_WARN("failed to add migrate trans table task generator task into dag", K(ret)); } else if (OB_FAIL(migrate_trans_table_task_generator_task->add_child(*wait_trans_table_finish_task))) { LOG_WARN("failed to add wait rans table finish task", K(ret)); } else if (OB_FAIL(dag_->alloc_task(wait_recovery_point_finish_task))) { LOG_WARN("failed to alloc wait recovery point finish task", K(ret)); } else if (OB_FAIL(wait_trans_table_finish_task->add_child(*wait_recovery_point_finish_task))) { LOG_WARN("failed to add wait recovery point finish task", K(ret)); } else if (OB_FAIL(dag_->add_task(*wait_trans_table_finish_task))) { LOG_WARN("failed to add wait trans table finish task into dag", K(ret)); } else if (!ctx_->recovery_point_ctx_.recovery_point_key_array_.empty()) { if (OB_FAIL(dag_->alloc_task(recovery_point_task_generator_task))) { LOG_WARN("failed to alloc migrate task generator task", K(ret)); } else if (OB_FAIL(recovery_point_task_generator_task->init(*wait_recovery_point_finish_task))) { LOG_WARN("failed to init migrate task generator task", K(ret)); } else if (OB_FAIL(wait_trans_table_finish_task->add_child(*recovery_point_task_generator_task))) { LOG_WARN("failed to add migrate task generator task", K(ret)); } else if (OB_FAIL(dag_->add_task(*recovery_point_task_generator_task))) { LOG_WARN("failed to add migrate task generator task into dag", K(ret)); } } if (OB_FAIL(ret)) { } else if (OB_FAIL(dag_->alloc_task(migrate_task_generator_task))) { LOG_WARN("failed to alloc migrate task generator task", K(ret)); } else if (OB_FAIL(migrate_task_generator_task->init())) { LOG_WARN("failed to init migrate task generator task", K(ret)); } else if (OB_FAIL(wait_recovery_point_finish_task->add_child(*migrate_task_generator_task))) { LOG_WARN("failed to add migrate task generator task", K(ret)); } else if (OB_FAIL(dag_->add_task(*wait_recovery_point_finish_task))) { LOG_WARN("failed to add wait recovery point finish task", K(ret)); } else if (OB_FAIL(dag_->add_task(*migrate_task_generator_task))) { LOG_WARN("failed to add wait recovery point finish task", K(ret)); } return ret; } int ObITableTaskGeneratorTask::need_generate_sstable_migrate_tasks(bool& need_schedule) { int ret = OB_SUCCESS; need_schedule = true; bool is_replica_with_data = true; if (NULL != ctx_) { if (OB_FAIL(ctx_->change_replica_with_data(is_replica_with_data))) { LOG_WARN("failed to check replica", K(ret), K(*ctx_)); } else if (is_replica_with_data) { need_schedule = false; LOG_INFO("change replica op with data, no need migrate task", K(*ctx_)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(ctx_->replica_op_arg_.dst_.get_replica_type())) { need_schedule = false; LOG_INFO("dst replica has no ssstore, no need migrate task", K(*ctx_)); } else if (ctx_->is_leader_restore_archive_data()) { need_schedule = false; LOG_INFO("no need restore archive data", "arg", ctx_->replica_op_arg_); } else if (ADD_REPLICA_OP == ctx_->replica_op_arg_.type_ && (REPLICA_LOGICAL_RESTORE_DATA == ctx_->is_restore_ || REPLICA_RESTORE_DATA == ctx_->is_restore_ || REPLICA_RESTORE_STANDBY == ctx_->is_restore_)) { need_schedule = false; LOG_INFO("no need migrate task for add replica when restoring", "arg", ctx_->replica_op_arg_); } } return ret; } int ObMigratePrepareTask::prepare_migrate() { int ret = OB_SUCCESS; const bool is_write_lock = true; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else { ObMigrateCtxGuard guard(is_write_lock, *ctx_); if (ctx_->is_leader_restore_archive_data()) { LOG_INFO("no need restore archive data", "arg", ctx_->replica_op_arg_); } else if (OB_FAIL(prepare_restore_reader_if_needed())) { LOG_WARN("failed to prepare restore reader", K(ret)); } else if (OB_FAIL(choose_migrate_src())) { LOG_WARN("fail to choose migrate src", K(ret)); } else if (OB_FAIL(check_backup_data_continues())) { LOG_WARN("failed to check backup data continues", K(ret)); } else if (OB_FAIL(build_migrate_pg_partition_info())) { LOG_WARN("failed to build migrate pg partition info", K(ret), K(*ctx_)); } else if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need replay for BACKUP_REPLICA_OP", "arg", ctx_->replica_op_arg_); } else if (VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need replay for VALIDATE_BACKUP_OP", "arg", ctx_->replica_op_arg_); } else if (BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need replay for BACKUP_BACKUPSET_OP", "arg", ctx_->replica_op_arg_); } else if (BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need replay for BACKUP_ARCHIVELOG_OP", "arg", ctx_->replica_op_arg_); } else if (OB_ISNULL(ctx_->get_partition())) { if (OB_FAIL(prepare_new_partition())) { LOG_WARN("failed to prepare new partition", K(ret), "arg", ctx_->replica_op_arg_); } else { ctx_->create_new_pg_ = true; } } if (OB_SUCC(ret)) { ATOMIC_AAF(&ctx_->data_statics_.partition_count_, ctx_->pg_meta_.partitions_.count()); ctx_->need_rebuild_ = false; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(INFO, "finish prepare migrate", "pkey", ctx_->replica_op_arg_.key_, K(*ctx_)); } else if (NULL != ctx_) { ctx_->result_ = ret; } } return ret; } int ObMigrateTaskSchedulerTask::try_hold_local_partition() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPGStorage* pg_storage = NULL; common::ObAddr leader; ObReplicaProperty replica_property; int16_t restore_state = 0; int64_t trans_table_end_log_ts = 0; int64_t trans_table_timestamp = 0; ObRole role; ObPartitionGroupMeta local_pg_meta; ObPartGroupMigrationTask* group_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_ || BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_ || BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need to hold local partition", "type", ctx_->replica_op_arg_.type_); } else if (OB_NOT_NULL(partition = ctx_->get_partition())) { // do nothing pg_storage = &(partition->get_pg_storage()); LOG_INFO("partition already hold, do not hold again"); } else if (OB_FAIL(ObPartitionService::get_instance().get_partition( ctx_->replica_op_arg_.key_, ctx_->partition_guard_))) { if (OB_PARTITION_NOT_EXIST != ret) { STORAGE_LOG(WARN, "get partition failed.", K(ret), K(*ctx_)); } else if (ctx_->is_only_copy_sstable()) { LOG_WARN("failed to get partition for copy index", K(ret), K(*ctx_)); } else if (CHANGE_REPLICA_OP == ctx_->replica_op_arg_.type_) { LOG_WARN("partition not exist, cannot change replica", K(ret), K(*ctx_)); } else if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { LOG_WARN("partition not exist, cannot rebuild replica", K(ret), K(*ctx_)); } else if (LINK_SHARE_MAJOR_OP == ctx_->replica_op_arg_.type_) { LOG_WARN("partition not exist, cannot link share major", K(ret), K(*ctx_)); } else { ret = OB_SUCCESS; ctx_->replica_state_ = OB_NOT_EXIST_REPLICA; } } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group()) || OB_ISNULL(pg_storage = &(partition->get_pg_storage()))) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "partition must not null", K(ret)); } else if (FALSE_IT(ctx_->is_restore_ = partition->get_pg_storage().get_restore_state())) { } else if (ctx_->replica_op_arg_.is_physical_restore() && REPLICA_RESTORE_DATA != ctx_->is_restore_ && REPLICA_RESTORE_ARCHIVE_DATA != ctx_->is_restore_) { ObMigrateStatus migrate_status; if (OB_FAIL(partition->get_pg_storage().get_pg_migrate_status(migrate_status))) { STORAGE_LOG(WARN, "get pg migrate status fail", K(ret), K(ctx_->replica_op_arg_)); } else if (OB_MIGRATE_STATUS_NONE != migrate_status) { ret = OB_STATE_NOT_MATCH; STORAGE_LOG( WARN, "physical restore failed", K(ret), K(ctx_->is_restore_), K(migrate_status), K(ctx_->replica_op_arg_)); } else { // restore = 0,1,3,4,5 skip physical restore and ret will be set succ before dag finish ret = OB_RESTORE_PARTITION_IS_COMPELETE; STORAGE_LOG(INFO, "physical restore base data is already finished", K(ret), K(ctx_->is_restore_), K(ctx_->replica_op_arg_)); } } else if (ctx_->replica_op_arg_.is_standby_restore() && REPLICA_RESTORE_STANDBY != ctx_->is_restore_ && REPLICA_RESTORE_STANDBY_CUT != ctx_->is_restore_) { ret = OB_ERR_UNEXPECTED; LOG_WARN("standby replica restore but restore status is wrong", K(*ctx_)); } else if (OB_FAIL(partition->get_leader(leader))) { LOG_WARN("failed to get leader", K(ret), "arg", ctx_->replica_op_arg_); } else if (OB_FAIL(partition->get_role(role))) { LOG_WARN("failed to get partition role", K(ret), "arg", ctx_->replica_op_arg_); } else if (leader.is_valid() && leader == MYADDR // support rebuild in leader reconfirm && partition->get_log_service()->is_leader_active() // TODO(wait yanmu) //&& is_strong_leader(role) && (ADD_REPLICA_OP == ctx_->replica_op_arg_.type_ || MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_ || FAST_MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_ || REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_ || CHANGE_REPLICA_OP == ctx_->replica_op_arg_.type_ || LINK_SHARE_MAJOR_OP == ctx_->replica_op_arg_.type_)) { if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(MIGRATOR.get_partition_service()->turn_off_rebuild_flag(ctx_->replica_op_arg_))) { LOG_WARN("Failed to report_rebuild_replica off", K(ret), "arg", ctx_->replica_op_arg_); } else { ret = OB_NO_NEED_REBUILD; LOG_WARN("leader can not as rebuild dst", K(ret), K(leader), "myaddr", MYADDR, "arg", ctx_->replica_op_arg_); } } else { ret = OB_ERR_SYS; LOG_WARN("leader cannot as add_replica,migrate,change_replica,share_major dst", K(ret), K(leader), "myaddr", MYADDR, "arg", ctx_->replica_op_arg_); } } else if (OB_ISNULL(ctx_->group_task_)) { ret = OB_ERR_UNEXPECTED; LOG_ERROR("group task should not be null here", K(ret)); } else if (FALSE_IT(group_task = reinterpret_cast(ctx_->group_task_))) { } else if (OB_FAIL(group_task->set_migrate_in(partition->get_pg_storage()))) { LOG_WARN("failed to set migrate int", "pkey", ctx_->replica_op_arg_.key_, K(ret)); } if (OB_FAIL(ret) || OB_ISNULL(partition) || OB_ISNULL(pg_storage)) { // do nothing } else if (OB_FAIL(partition->get_replica_state(ctx_->replica_state_))) { LOG_WARN("failed to get replica state", K(ret), K(*ctx_)); } else if (OB_FAIL(pg_storage->get_replica_property(replica_property))) { LOG_WARN("failed to get replica property", K(ret), "arg", ctx_->replica_op_arg_); } else if (OB_FAIL(pg_storage->get_pg_meta(local_pg_meta))) { LOG_WARN("failed to get pg meta", K(ret)); } else { if (OB_FAIL(pg_storage->get_trans_table_end_log_ts_and_timestamp(trans_table_end_log_ts, trans_table_timestamp))) { if (OB_ENTRY_NOT_EXIST != ret && OB_PARTITION_NOT_EXIST != ret) { LOG_WARN("failed to get trans table end_log_ts and timestamp", K(ret), K(ctx_->replica_op_arg_.key_)); } else if (ADD_REPLICA_OP == ctx_->replica_op_arg_.type_ || MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_ || FAST_MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_ || RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_ || RESTORE_FOLLOWER_REPLICA_OP == ctx_->replica_op_arg_.type_ || RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_) { ret = OB_SUCCESS; } else { LOG_WARN("failed to get trans table end_log_ts and timestamp", K(ret), K(ctx_->replica_op_arg_)); } } if (OB_FAIL(ret)) { } else { ctx_->local_publish_version_ = local_pg_meta.storage_info_.get_data_info().get_publish_version(); ctx_->local_last_replay_log_id_ = local_pg_meta.storage_info_.get_clog_info().get_last_replay_log_id(); ctx_->local_last_replay_log_ts_ = std::max(trans_table_end_log_ts, local_pg_meta.storage_info_.get_data_info().get_last_replay_log_ts()); common::ObReplicaType replica_type = ctx_->replica_op_arg_.dst_.get_replica_type(); common::ObReplicaType local_replica_type = partition->get_replica_type(); if (local_replica_type != replica_type && !ObReplicaTypeCheck::change_replica_op_allow(local_replica_type, replica_type)) { ret = OB_OP_NOT_ALLOW; STORAGE_LOG(WARN, "change replica op not allow", K(ret), K(local_replica_type), K(replica_type), K(*ctx_)); } } if (OB_SUCC(ret)) { if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (pg_storage->is_replica_with_remote_memstore()) { ctx_->use_slave_safe_read_ts_ = false; LOG_INFO("rebuild replica op which memstore percent is 0 do not use slave safe read ts", K(replica_property)); } } else if (ctx_->is_leader_restore_archive_data()) { if (OB_FAIL(partition->get_pg_storage().get_pg_meta(ctx_->pg_meta_))) { ctx_->result_ = ret; STORAGE_LOG(WARN, "get local pg meta fail", K(ret), K(ctx_->replica_op_arg_.key_)); } } } } if (OB_FAIL(ret) && NULL != ctx_) { ctx_->can_rebuild_ = false; LOG_WARN("failed during try_hold_local_partition, set cannot rebuild", K(ret), K(*ctx_)); } return ret; } int ObMigratePrepareTask::choose_migrate_src() { int ret = OB_SUCCESS; const int64_t local_cluster_id = GCONF.cluster_id; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else { const ObReplicaOpType op_type = ctx_->replica_op_arg_.type_; switch (op_type) { case VALIDATE_BACKUP_OP: case BACKUP_BACKUPSET_OP: case BACKUP_ARCHIVELOG_OP: { LOG_INFO("no need to choose migrate src", K(ret)); break; } case RESTORE_REPLICA_OP: { if (OB_FAIL(choose_restore_migrate_src(ctx_->replica_op_arg_, ctx_->pg_meta_, ctx_->migrate_src_info_))) { LOG_WARN("failed to choose restore migrate src", K(ret)); } break; } case BACKUP_REPLICA_OP: { if (OB_FAIL(choose_backup_migrate_src(ctx_->replica_op_arg_, ctx_->pg_meta_, ctx_->migrate_src_info_))) { LOG_WARN("failed to choose restore migrate src", K(ret)); } break; } case REBUILD_REPLICA_OP: { if (OB_FAIL(choose_ob_rebuild_src(ctx_->replica_op_arg_, ctx_->pg_meta_, ctx_->migrate_src_info_))) { LOG_WARN("failed to choose rebuild src", K(ret)); } break; } case CHANGE_REPLICA_OP: case ADD_REPLICA_OP: case MIGRATE_REPLICA_OP: { if (ctx_->replica_op_arg_.cluster_id_ != local_cluster_id) { ret = OB_ERR_SYS; LOG_WARN("can not migrate from diff cluster", K(*ctx_), K(local_cluster_id)); } else if (OB_FAIL(choose_ob_migrate_src(ctx_->replica_op_arg_, ctx_->pg_meta_, ctx_->migrate_src_info_))) { LOG_WARN("fail to choose ob master migrate src", K(ret)); } LOG_INFO("migrate choose src", K(*ctx_)); break; } case RESTORE_STANDBY_OP: { if (OB_FAIL(choose_standby_restore_src(ctx_->replica_op_arg_, ctx_->pg_meta_, ctx_->migrate_src_info_))) { LOG_WARN("failed to choose standby restore src", K(ret)); } break; } case COPY_LOCAL_INDEX_OP: case COPY_GLOBAL_INDEX_OP: case RESTORE_FOLLOWER_REPLICA_OP: case FAST_MIGRATE_REPLICA_OP: { // these operation type no need choose src, use rs provided if (OB_FAIL(choose_recommendable_src(ctx_->replica_op_arg_, ctx_->pg_meta_, ctx_->migrate_src_info_))) { LOG_WARN("failed to choose recommendable src", K(ret)); } break; } default: { ret = OB_ERR_UNEXPECTED; LOG_WARN("unexpected replica op type", K(ret), K(*ctx_)); break; } } // end-switch } return ret; } int ObMigratePrepareTask::choose_ob_migrate_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; const int64_t local_cluster_id = GCONF.cluster_id; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (ctx_->replica_op_arg_.cluster_id_ != local_cluster_id && MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; LOG_WARN("can not migrate from diff cluster", K(*ctx_), K(local_cluster_id)); } else if (OB_FAIL(choose_ob_src(is_valid_migrate_src, arg, pg_meta, src_info))) { LOG_WARN("failed to choose ob src", K(ret), K(arg)); } else if (ctx_->is_copy_index()) { ctx_->recovery_point_ctx_.recovery_point_key_array_.reset(); } return ret; } int ObMigratePrepareTask::choose_ob_rebuild_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(choose_rebuild_src(arg, meta, src_info))) { LOG_WARN("failed to choose local cluster rebuild src, try choose remote cluster rebuild src", K(ret), K(arg)); } return ret; } int ObMigratePrepareTask::choose_rebuild_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; ObArray src_info_array; meta.reset(); src_info_array.reset(); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(partition_service_->get_migrate_leader_and_parent(arg.key_, src_info_array))) { LOG_WARN("failed to get migrat src", K(ret), K(arg)); } else if (OB_FAIL(choose_rebuild_candidate(arg, src_info_array, meta, src_info))) { LOG_WARN("fail to choose rebuild candidate", K(ret), K(arg)); } if (OB_FAIL(ret)) { // overwrite ret src_info_array.reset(); if (OB_FAIL(partition_service_->get_rebuild_src(arg.key_, arg.dst_.get_replica_type(), src_info_array))) { LOG_WARN("failed to get migrat src", K(ret), K(arg)); } else if (OB_FAIL(choose_rebuild_candidate(arg, src_info_array, meta, src_info))) { LOG_WARN("fail to choose rebuild candidate", K(ret), K(arg)); } } return ret; } int ObMigratePrepareTask::copy_needed_table_info(const ObReplicaOpArg& arg, const obrpc::ObPGPartitionMetaInfo& data_src_result, ObIArray& table_info_res, ObIArray& table_id_list, bool& found) { int ret = OB_SUCCESS; found = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else { if (COPY_LOCAL_INDEX_OP != arg.type_) { if (OB_FAIL(table_id_list.assign(data_src_result.table_id_list_))) { LOG_WARN("failed to assign table id list", K(ret)); } else if (OB_FAIL(table_info_res.assign(data_src_result.table_info_))) { LOG_WARN("failed to copy needed table info", K(ret), K(data_src_result)); } else { found = true; } } else { for (int64_t i = 0; OB_SUCC(ret) && i < data_src_result.table_id_list_.count() && !found; ++i) { found = arg.index_id_ == data_src_result.table_id_list_.at(i); if (found) { if (OB_FAIL(table_id_list.push_back(data_src_result.table_id_list_.at(i)))) { LOG_WARN("failed to push table id into array", K(ret), K(arg)); } else if (OB_FAIL(table_info_res.push_back(data_src_result.table_info_.at(i)))) { LOG_WARN("failed to push table info into array", K(ret), K(data_src_result), K(i)); } } } } if (OB_SUCC(ret)) { if (found) { if (table_id_list.count() != table_info_res.count()) { ret = OB_ERR_SYS; LOG_WARN("table id list count not equal to table info count", K(ret), "table_id_list count", table_id_list.count(), "table_info count", table_info_res.count()); } } } } return ret; } int ObMigratePrepareTask::fetch_partition_group_info( const ObReplicaOpArg& arg, const ObMigrateSrcInfo& src_info, ObPartitionGroupInfoResult& result) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (!src_info.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("fetch rebuild partition info get invalid argument", K(ret), K(src_info)); } else if (OB_FAIL(rpc_->post_fetch_partition_group_info_request(src_info.src_addr_, arg.key_, arg.dst_.get_replica_type(), src_info.cluster_id_, ctx_->use_slave_safe_read_ts_, result.result_))) { LOG_WARN("fail to post fetch partition info request", K(ret), K(src_info), K(arg)); } else { result.choose_src_info_ = src_info; } return ret; } int ObMigratePrepareTask::choose_rebuild_candidate(const ObReplicaOpArg& arg, const common::ObIArray& src_info_array, ObPartitionGroupMeta& meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObArray same_region_array; ObArray diff_region_array; bool find_src = false; const int64_t local_cluster_id = GCONF.cluster_id; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (src_info_array.empty() || !arg.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("choose fetch candidate get invalid argument", K(ret), K(src_info_array.count()), K(arg)); } else if (OB_FAIL(split_candidate_with_region(src_info_array, same_region_array, diff_region_array))) { LOG_WARN("fail to split candidate with region", K(ret)); } else if (src_info_array.count() != (same_region_array.count() + diff_region_array.count())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("split candidate error ", K(ret), K(src_info_array.count()), K(same_region_array.count()), K(diff_region_array.count())); } else { // first choose same region ObPartitionGroupInfoResult result; ObPartitionGroupInfoResult tmp_result; if (OB_SUCCESS != (tmp_ret = fetch_suitable_rebuild_src(arg, same_region_array, result))) { LOG_WARN("fail to fetch same region suitable src and result", K(tmp_ret), K(arg)); } else { find_src = result.is_valid() && result.result_.is_log_sync_; if (find_src && OB_FAIL(ctx_->recovery_point_ctx_.recovery_point_key_array_.assign( result.result_.recovery_point_key_array_))) { LOG_WARN("failed to assign recovery point key array", K(ret), K(arg)); } } // if not find same region src, then try choose from all region if (OB_SUCC(ret) && !find_src) { if (OB_FAIL(fetch_suitable_rebuild_src(arg, diff_region_array, tmp_result))) { LOG_WARN("fail to fetch diff region suitable src and result", K(ret), K(arg)); } else if (OB_FAIL(copy_rebuild_partition_info_result(tmp_result, result, find_src))) { LOG_WARN("fail to copy rebuild partition info", K(ret), K(tmp_result), K(result)); } else if (!find_src) { } else if (OB_FAIL(ctx_->recovery_point_ctx_.recovery_point_key_array_.assign( result.result_.recovery_point_key_array_))) { LOG_WARN("failed to assgin recovery point key array", K(ret), K(result)); } } if (OB_SUCC(ret) && !result.is_valid()) { ret = OB_DATA_SOURCE_NOT_EXIST; LOG_WARN("cannot choose_ob_rebuild_src", K(ret), K(arg), K(src_info_array), K(result), K(find_src)); } else { LOG_INFO("choose_ob_rebuild_src", "src", result.choose_src_info_); if (OB_FAIL(meta.deep_copy(result.result_.pg_meta_))) { LOG_WARN("failed to deep copy meta", K(ret)); } else { ctx_->mig_src_file_id_ = result.result_.pg_file_id_; ctx_->fetch_pg_info_compat_version_ = result.result_.compat_version_; src_info = result.choose_src_info_; if (src_info.cluster_id_ != local_cluster_id) { // restore point can not create to standby cluster ctx_->recovery_point_ctx_.recovery_point_key_array_.reset(); } } } } return ret; } int ObMigratePrepareTask::copy_rebuild_partition_info_result( ObPartitionGroupInfoResult& tmp_result, ObPartitionGroupInfoResult& target_result, bool& find_src) { int ret = OB_SUCCESS; find_src = false; if (tmp_result.is_valid()) { bool need_assign = false; if (target_result.is_valid()) { if (tmp_result.result_.pg_meta_.storage_info_.get_data_info().get_publish_version() > target_result.result_.pg_meta_.storage_info_.get_data_info().get_publish_version()) { need_assign = true; } } else { need_assign = true; } if (need_assign) { target_result.reset(); if (OB_FAIL(target_result.assign(tmp_result))) { LOG_WARN("fail to assign out result", K(ret), K(tmp_result)); } else { find_src = target_result.is_valid() && target_result.result_.is_log_sync_; } } } return ret; } int ObMigratePrepareTask::split_candidate_with_region(const ObIArray& src_info_array, ObIArray& same_region_array, ObIArray& diff_region_array) { int ret = OB_SUCCESS; ObRegion local_addr_region; const ObAddr& self_addr = OBSERVER.get_self(); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (src_info_array.empty()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("split candidate with region get invaid argument", K(ret), K(src_info_array.count())); } else if (OB_FAIL(partition_service_->get_server_region(self_addr, local_addr_region))) { LOG_WARN("fail to get local src region", K(ret), K(self_addr)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < src_info_array.count(); ++i) { ObRegion tmp_region; const ObMigrateSrcInfo& src = src_info_array.at(i); if (OB_FAIL(partition_service_->get_server_region_across_cluster(src.src_addr_, tmp_region))) { LOG_WARN("fail to get server region", K(ret), K(src)); } else if (local_addr_region == tmp_region) { if (OB_FAIL(same_region_array.push_back(src))) { LOG_WARN("fail to push src into same region array", K(ret), K(src)); } } else { if (OB_FAIL(diff_region_array.push_back(src))) { LOG_WARN("fail to push src into diff region array", K(ret), K(src)); } } } } return ret; } int ObMigratePrepareTask::fetch_suitable_rebuild_src(const ObReplicaOpArg& arg, const common::ObIArray& src_array, ObPartitionGroupInfoResult& out_result) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPGStorage* pg_storage = NULL; ObPartitionGroupMeta local_pg_meta; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (!arg.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("fetch suitable src get invalid argument", K(ret), K(arg)); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group()) || OB_ISNULL(pg_storage = &(partition->get_pg_storage()))) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "partition must not null", K(ret)); } else if (OB_FAIL(pg_storage->get_pg_meta(local_pg_meta))) { LOG_WARN("failed to get pg meta", K(ret)); } else { ObPartitionGroupInfoResult tmp_result; int64_t max_major_snapshot_version = local_pg_meta.report_status_.snapshot_version_; int64_t max_minor_snapshot_version = ctx_->local_publish_version_; // TODO() handle 22x compatible uint64_t max_migrate_replay_log_ts = std::max(local_pg_meta.get_migrate_replay_log_ts(), ctx_->local_last_replay_log_ts_); int64_t max_last_replay_log_id = ctx_->local_last_replay_log_id_; for (int64_t i = 0; OB_SUCC(ret) && i < src_array.count(); ++i) { tmp_result.reset(); const ObMigrateSrcInfo& src_info = src_array.at(i); bool is_valid_src = false; bool is_suitable = false; if (!src_info.is_valid()) { ret = OB_ERR_SYS; LOG_ERROR("invalid src addr", K(src_info)); } else if (OB_SUCCESS != (tmp_ret = fetch_partition_group_info(arg, src_info, tmp_result))) { LOG_WARN("fail to fetch rebuild partition info", K(tmp_ret), K(src_info)); } else if (OB_FAIL(is_valid_rebuild_src(tmp_result.result_, *ctx_, is_valid_src))) { LOG_WARN("failed to check rebuild src valid", K(ret), K(tmp_result)); } else if (!is_valid_src) { // do nothing } else { const ObDataStorageInfo& data_info = tmp_result.result_.pg_meta_.storage_info_.get_data_info(); const int64_t remote_minor_snapshot_version = data_info.get_publish_version(); const int64_t remote_major_snapshot_version = tmp_result.result_.pg_meta_.report_status_.snapshot_version_; const uint64_t remote_last_replay_log_id = tmp_result.result_.pg_meta_.storage_info_.get_clog_info().get_last_replay_log_id(); const uint64_t remote_migrate_replay_log_ts = tmp_result.result_.pg_meta_.get_migrate_replay_log_ts(); bool is_suitable = false; const bool is_empty_pg = tmp_result.result_.pg_meta_.partitions_.empty(); // new freeze might generate empty log_ts_range, still need use version compare // pg upgrade from 22x will always has replay_log_ts with 4096 if (remote_migrate_replay_log_ts > max_migrate_replay_log_ts) { is_suitable = true; } else if (!ctx_->is_migrate_compat_version() && remote_migrate_replay_log_ts < max_migrate_replay_log_ts) { // from 3.1, replay_log_ts is an accurate metric is_suitable = false; } else if (is_empty_pg) { if (remote_last_replay_log_id > max_last_replay_log_id) { is_suitable = true; } } else if (remote_minor_snapshot_version > max_minor_snapshot_version) { is_suitable = true; } else if (remote_minor_snapshot_version < max_minor_snapshot_version) { is_suitable = false; } else if (remote_major_snapshot_version > max_major_snapshot_version) { is_suitable = true; } else if (remote_major_snapshot_version < max_major_snapshot_version) { is_suitable = false; } else if (remote_last_replay_log_id > max_last_replay_log_id) { is_suitable = true; } else { // remote_last_replay_log_id <= max_last_replay_log_id is_suitable = false; } #ifdef ERRSIM if (OB_SUCC(ret) && !is_suitable) { is_suitable = ObServerConfig::get_instance()._ob_enable_rebuild_on_purpose; } #endif if (is_suitable) { out_result.reset(); if (OB_FAIL(out_result.assign(tmp_result))) { LOG_WARN("fail to assign out result", K(ret), K(tmp_result)); } else { max_minor_snapshot_version = std::max(remote_minor_snapshot_version, max_minor_snapshot_version); max_major_snapshot_version = remote_major_snapshot_version; max_last_replay_log_id = remote_last_replay_log_id; max_migrate_replay_log_ts = remote_migrate_replay_log_ts; LOG_INFO("find one candidate src", K(src_info), K(remote_minor_snapshot_version), K(max_minor_snapshot_version), K(remote_major_snapshot_version), K(remote_last_replay_log_id), K(remote_migrate_replay_log_ts)); } } else { LOG_INFO("src snapshot version cannot use as src", K(src_info), K(remote_minor_snapshot_version), K(max_minor_snapshot_version), K(remote_major_snapshot_version), K(max_major_snapshot_version), K(remote_last_replay_log_id), K(max_last_replay_log_id), K(remote_migrate_replay_log_ts), K(max_migrate_replay_log_ts), K_(ctx_->fetch_pg_info_compat_version)); } } } } return ret; } int ObMigratePrepareTask::choose_restore_migrate_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; const int64_t compatible = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.compatible_; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(ctx_->restore_meta_reader_)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("restore meta read is NULL", K(ret), KP(ctx_->restore_meta_reader_)); } else { src_info.src_addr_ = arg.data_src_.get_server(); src_info.cluster_id_ = arg.cluster_id_; if (OB_FAIL(ctx_->restore_meta_reader_->fetch_partition_group_meta(pg_meta))) { LOG_WARN("fail to fetch partition store meta", K(ret), K(arg)); } else { if (OB_BACKUP_COMPATIBLE_VERSION_V1 == compatible || OB_BACKUP_COMPATIBLE_VERSION_V2 == compatible) { ctx_->fetch_pg_info_compat_version_ = ObFetchPGInfoArg::FETCH_PG_INFO_ARG_COMPAT_VERSION_V1; } else if (OB_BACKUP_COMPATIBLE_VERSION_V3 == compatible) { ctx_->fetch_pg_info_compat_version_ = ObFetchPGInfoArg::FETCH_PG_INFO_ARG_COMPAT_VERSION_V2; } else { ret = OB_ERR_UNEXPECTED; LOG_WARN("restore compatible version is unexpected", K(ret), K(compatible), "arg", arg); } } } if (OB_SUCC(ret)) { LOG_INFO("succeed to choose_restore_migrate_src", K(ret), K(arg), K(pg_meta)); } return ret; } int ObMigratePrepareTask::choose_phy_restore_follower_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; ObPartitionGroupInfoResult src_result; ObMigrateSrcInfo tmp_src_info; tmp_src_info.src_addr_ = arg.data_src_.get_server(); tmp_src_info.cluster_id_ = arg.cluster_id_; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(fetch_partition_group_info(arg, tmp_src_info, src_result))) { LOG_WARN("failed to fetch partition info", K(ret), K(arg)); } else if (OB_FAIL(pg_meta.deep_copy(src_result.result_.pg_meta_))) { LOG_WARN("Failed to copy pg meta", K(ret)); } else { src_info = tmp_src_info; LOG_INFO("succeed to choose_phy_restore_follower src", K(ret), K(arg), K(pg_meta)); } return ret; } // TODO(muwei.ym) delete it later int ObMigratePrepareTask::choose_backup_migrate_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!pg_meta.is_valid()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "backup pg meta is invalid", K(ret), K(pg_meta)); } else { src_info.src_addr_ = arg.data_src_.get_server(); src_info.cluster_id_ = arg.cluster_id_; } DEBUG_SYNC(BACKUP_BEFROE_CHOOSE_SRC); if (OB_SUCC(ret)) { STORAGE_LOG(INFO, "succeed to choose_backup_migrate_src", K(ret), K(arg), K(pg_meta)); } return ret; } int ObMigratePrepareTask::choose_standby_restore_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; const int64_t local_cluster_id = GCONF.cluster_id; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (ctx_->replica_op_arg_.cluster_id_ != local_cluster_id && RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; LOG_WARN("can not migrate from diff cluster", K(*ctx_), K(local_cluster_id)); } else if (OB_FAIL(choose_ob_src(is_valid_standby_restore_src, arg, pg_meta, src_info))) { LOG_WARN("failed to choose ob src", K(ret), K(arg)); } else { ctx_->recovery_point_ctx_.recovery_point_key_array_.reset(); } return ret; } int ObMigratePrepareTask::choose_ob_src( IsValidSrcFunc is_valid_src, const ObReplicaOpArg& arg, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; ObPartitionGroupInfoResult data_src_result; ObArray src_info_array; bool find_suitable_src = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(get_minor_src_candidate_with_region(arg, src_info_array))) { LOG_WARN("failed to get minor src candidate with region", K(ret), K(arg)); } else if (!src_info_array.empty()) { if (OB_FAIL(get_migrate_suitable_src(src_info_array, arg, is_valid_src, find_suitable_src, pg_meta, src_info))) { LOG_WARN("failed to get migrate suitable src", K(ret)); } } // rewrite ret if (!find_suitable_src) { src_info_array.reset(); if (OB_FAIL(get_minor_src_candidate_without_region(arg, src_info_array))) { LOG_WARN("failed to get minor src candidate without region", K(ret), K(arg)); } else if (src_info_array.empty()) { ret = OB_DATA_SOURCE_NOT_EXIST; STORAGE_LOG(WARN, "no available src, migrate failed", K(ret), K(arg), K(src_info_array)); } else if (OB_FAIL( get_migrate_suitable_src(src_info_array, arg, is_valid_src, find_suitable_src, pg_meta, src_info))) { LOG_WARN("failed to get migrate suitable src", K(ret)); } } if (OB_SUCC(ret) && !find_suitable_src) { ret = OB_DATA_SOURCE_NOT_EXIST; STORAGE_LOG(WARN, "no available src, migrate failed", K(ret), K(arg), K(src_info_array)); } return ret; } int ObMigratePrepareTask::is_valid_migrate_src( const obrpc::ObFetchPGInfoResult& result, ObMigrateCtx& ctx, bool& is_valid) { int ret = OB_SUCCESS; is_valid = false; const ObDataStorageInfo& data_info = result.pg_meta_.storage_info_.get_data_info(); const ObBaseStorageInfo& clog_info = result.pg_meta_.storage_info_.get_clog_info(); const ObReplicaType replica_type = ctx.replica_op_arg_.dst_.get_replica_type(); const int64_t remote_last_replay_log_id = result.pg_meta_.storage_info_.get_clog_info().get_last_replay_log_id(); const int64_t remote_publish_version = data_info.get_publish_version(); const bool is_empty_pg = result.pg_meta_.partitions_.empty(); if (!ObReplicaTypeCheck::is_replica_with_memstore(replica_type) && !ObReplicaTypeCheck::is_replica_with_ssstore(replica_type)) { is_valid = true; LOG_INFO("data is not needed, skip check src valid"); } else if (data_info.is_created_by_new_minor_freeze() && data_info.get_last_replay_log_ts() < ctx.local_last_replay_log_ts_ && CHANGE_REPLICA_OP != ctx.replica_op_arg_.type_) { LOG_INFO("replica with smaller last_replay_log_id is not a valid migrate src", "remote last_replay_log_ts", data_info.get_last_replay_log_ts(), "local last_replay_log_ts", ctx.local_last_replay_log_ts_); } else if (!can_migrate_src_skip_log_sync(result, ctx) && !result.is_log_sync_) { LOG_INFO("src log not sync", K(result)); } else if (CHANGE_REPLICA_OP == ctx.replica_op_arg_.type_) { // no need check last replay log id is_valid = true; } else if (remote_last_replay_log_id < ctx.local_last_replay_log_id_) { LOG_INFO("src last replay log id must not less than local", "remote_last_replay_log_id", remote_last_replay_log_id, "local_last_replay_log_id", ctx.local_last_replay_log_id_); } else { is_valid = true; } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_VALID_MIGRATE_SRC) OB_SUCCESS; } #endif return ret; } int ObMigratePrepareTask::is_valid_standby_restore_src( const obrpc::ObFetchPGInfoResult& result, ObMigrateCtx& ctx, bool& is_valid) { int ret = OB_SUCCESS; is_valid = false; const ObDataStorageInfo& data_info = result.pg_meta_.storage_info_.get_data_info(); const ObBaseStorageInfo& clog_info = result.pg_meta_.storage_info_.get_clog_info(); const ObReplicaType replica_type = ctx.replica_op_arg_.dst_.get_replica_type(); const int64_t remote_last_replay_log_id = result.pg_meta_.storage_info_.get_clog_info().get_last_replay_log_id(); // TODO() standby cluster need choose data version >= self data version as src if (!ObReplicaTypeCheck::is_replica_with_ssstore(result.pg_meta_.replica_type_)) { LOG_WARN("src not has ssstore, cannot copy sstable", K(result)); } else if (REPLICA_NOT_RESTORE != result.pg_meta_.is_restore_) { LOG_INFO("data src is doing restore, cannot copy sstable"); } else if (!ObReplicaTypeCheck::is_replica_with_memstore(replica_type) && !ObReplicaTypeCheck::is_replica_with_ssstore(replica_type)) { is_valid = true; LOG_INFO("data is not needed, skip check src valid"); } else if (remote_last_replay_log_id < ctx.local_last_replay_log_id_) { LOG_INFO("src last replay log id must not less than local", "remote_last_replay_log_id", remote_last_replay_log_id, "local_last_replay_log_id", ctx.local_last_replay_log_id_); } else { is_valid = true; } return ret; } int ObMigratePrepareTask::is_valid_rebuild_src( const obrpc::ObFetchPGInfoResult& result, ObMigrateCtx& ctx, bool& is_valid) { int ret = OB_SUCCESS; is_valid = false; const ObReplicaType replica_type = ctx.replica_op_arg_.dst_.get_replica_type(); const int64_t remote_last_replay_log_id = result.pg_meta_.storage_info_.get_clog_info().get_last_replay_log_id(); if (REBUILD_REPLICA_OP != ctx.replica_op_arg_.type_) { ret = OB_INVALID_ARGUMENT; LOG_WARN("is valid rebuild src get invalid argument", K(ret), K(ctx)); } else if (ObReplicaTypeCheck::is_replica_with_ssstore(replica_type) && !ObReplicaTypeCheck::is_replica_with_ssstore(result.pg_meta_.replica_type_)) { LOG_WARN("src not has ssstore, cannot copy sstable", K(result)); } else if (remote_last_replay_log_id <= ctx.local_last_replay_log_id_) { LOG_INFO("src last replay log id must not less than local", "remote_last_replay_log_id", remote_last_replay_log_id, "local_last_replay_log_id", ctx.local_last_replay_log_id_); #ifdef ERRSIM is_valid = ObServerConfig::get_instance()._ob_enable_rebuild_on_purpose; #endif } else { is_valid = true; } return ret; } bool ObMigratePrepareTask::can_migrate_src_skip_log_sync(const obrpc::ObFetchPGInfoResult& result, ObMigrateCtx& ctx) { bool b_ret = false; if (REPLICA_NOT_RESTORE == result.pg_meta_.is_restore_) { b_ret = false; } else if (ADD_REPLICA_OP == ctx.replica_op_arg_.type_ || MIGRATE_REPLICA_OP == ctx.replica_op_arg_.type_) { if (result.pg_meta_.is_restore_ > REPLICA_NOT_RESTORE && result.pg_meta_.is_restore_ <= REPLICA_RESTORE_MEMBER_LIST) { b_ret = true; } } return b_ret; } int ObMigratePrepareTask::build_migrate_pg_partition_info() { int ret = OB_SUCCESS; ObIPGPartitionBaseDataMetaObReader* reader = NULL; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_ || BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_ || BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { // do nothing } else if (OB_FAIL(get_partition_table_info_reader(ctx_->migrate_src_info_, reader))) { LOG_WARN("failed to get partition info reader", K(ret), K(ctx_)); } else { if (COPY_LOCAL_INDEX_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(build_index_partition_info(ctx_->replica_op_arg_, reader))) { LOG_WARN("failed to build index partition info", K(ret), K(ctx_->replica_op_arg_)); } } else { if (OB_FAIL(build_table_partition_info(ctx_->replica_op_arg_, reader))) { LOG_WARN("failed to build table partition info", K(ret), K(ctx_->replica_op_arg_)); } } } if (NULL != reader) { cp_fty_->free(reader); } return ret; } int ObMigratePrepareTask::build_migrate_partition_info(const ObPGPartitionMetaInfo& partition_meta_info, const common::ObIArray& table_info_res, const common::ObIArray& table_id_list, ObPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; ObArray local_tables_info; ObMigrateTableInfo table_info; int64_t cost_ts = ObTimeUtility::current_time(); const ObPartitionKey& pkey = partition_meta_info.meta_.pkey_; ObMigratePartitionInfo& info = part_migrate_ctx.copy_info_; part_migrate_ctx.ctx_ = ctx_; DEBUG_SYNC(BEFORE_BUILD_MIGRATE_PARTITION_INFO); if (!is_inner_table(ctx_->replica_op_arg_.key_.get_table_id())) { DEBUG_SYNC(BEFORE_BUILD_MIGRATE_PARTITION_INFO_USER_TABLE); } if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (!partition_meta_info.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("build migrate partition info get invalid argument", K(ret), K(partition_meta_info)); } else if (OB_FAIL(info.meta_.deep_copy(partition_meta_info.meta_))) { LOG_WARN("fail to deep copy partition store meta", K(ret), K(partition_meta_info.meta_)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < table_info_res.count(); ++i) { table_info.reuse(); local_tables_info.reuse(); const obrpc::ObFetchTableInfoResult& table_res = table_info_res.at(i); const uint64_t table_id = table_id_list.at(i); LOG_INFO("build_migrate_partition_info for table", "table_id", table_id); if (OB_SUCC(ret)) { if (OB_FAIL(remove_uncontinue_local_tables(pkey, table_id))) { LOG_WARN("failed to remove uncontinue local tables", K(ret), K(pkey), K(table_id)); } else if (OB_FAIL(get_local_table_info(table_id, pkey, local_tables_info))) { LOG_WARN("failed to get local table info", K(ret)); } else if (OB_FAIL(build_migrate_table_info( table_id, pkey, local_tables_info, table_res, table_info, part_migrate_ctx))) { LOG_WARN("failed to build migrate table info", K(ret)); } else if (OB_FAIL(info.table_id_list_.push_back(table_id))) { LOG_WARN("failed to push table id into array", K(ret)); } else if (OB_FAIL(info.table_infos_.push_back(table_info))) { LOG_WARN("failed to add migrate table info", K(ret)); } else { LOG_INFO("add table info", K(table_info)); if (table_res.multi_version_start_ > info.meta_.multi_version_start_) { info.meta_.multi_version_start_ = table_res.multi_version_start_; LOG_INFO("update migrate multi version start", K(table_res.multi_version_start_)); } } } } } cost_ts = ObTimeUtility::current_time() - cost_ts; LOG_INFO("build_migrate_partition_info", K(cost_ts), "pkey", info.meta_.pkey_, "src", info.src_, "table_count", info.table_id_list_.count()); return ret; } int ObMigratePrepareTask::get_local_table_info( const uint64_t table_id, const ObPartitionKey& pkey, ObIArray& local_tables_info) { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPGPartitionGuard guard; ObPGPartition* pg_partition = NULL; ObIPartitionStorage* storage = NULL; ObTablesHandle handle; local_tables_info.reuse(); bool is_ready_for_read = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (NULL == (partition = ctx_->partition_guard_.get_partition_group())) { LOG_INFO("local partition not exist, no thing reuse", K(table_id)); } else if (OB_FAIL(partition->get_pg_partition(pkey, guard))) { if (OB_ENTRY_NOT_EXIST != ret) { LOG_WARN("failed to get pg partition", K(ret), K(pkey)); } else { ret = OB_SUCCESS; } } else if (OB_ISNULL(pg_partition = guard.get_pg_partition())) { ret = OB_ERR_SYS; LOG_ERROR("pg partition should not be NULL", K(ret), KP(pg_partition), K(pkey)); } else if (OB_ISNULL(storage = pg_partition->get_storage())) { ret = OB_ERR_SYS; LOG_ERROR("storage must not null", K(ret), KP(storage)); } else if (OB_FAIL(static_cast(storage)->get_partition_store().get_migrate_tables( table_id, handle, is_ready_for_read))) { LOG_WARN("failed to get effective tables", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < handle.get_count(); ++i) { const ObITable* table = handle.get_table(i); if (OB_ISNULL(table)) { ret = OB_ERR_SYS; LOG_ERROR("table must not null", K(ret)); } else if (table->is_memtable()) { ret = OB_ERR_SYS; LOG_ERROR("Unexpected local memtable", K(ret), K(i), KPC(table), K(handle)); } else if (OB_FAIL(local_tables_info.push_back(table->get_key()))) { LOG_WARN("failed to add local table info into array", K(ret), K(table->get_key())); } } } return ret; } int ObMigratePrepareTask::build_migrate_table_info(const uint64_t table_id, const ObPartitionKey& pkey, ObIArray& local_tables_info, const obrpc::ObFetchTableInfoResult& result, ObMigrateTableInfo& info, ObPartitionMigrateCtx& part_ctx) { int ret = OB_SUCCESS; ObArray remote_major_sstables; ObArray remote_inc_sstables; ObArray remote_gc_major_sstables; ObArray remote_gc_inc_sstables; ObArray local_major_sstables; ObArray local_inc_sstables; bool& need_reuse_local_minor = part_ctx.need_reuse_local_minor_; info.reuse(); info.table_id_ = table_id; // TODO only use source minor sstable now, consider performance later need_reuse_local_minor = false; FLOG_INFO("start to build_migrate_table_info", K(result), K(info), K(local_major_sstables), K(local_inc_sstables), K(remote_gc_major_sstables), K(pkey), K(table_id), K(need_reuse_local_minor)); if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(ObTableKeyMgrUtil::classify_mgirate_tables( table_id, local_tables_info, local_major_sstables, local_inc_sstables))) { LOG_WARN("failed to classify local migrate tables", K(ret)); } else if (OB_FAIL(ObTableKeyMgrUtil::classify_mgirate_tables( table_id, result.table_keys_, remote_major_sstables, remote_inc_sstables))) { LOG_WARN("failed to classify remote migrate tables", K(ret)); } else if (OB_FAIL(ObTableKeyMgrUtil::classify_mgirate_tables( table_id, result.gc_table_keys_, remote_gc_major_sstables, remote_gc_inc_sstables))) { LOG_WARN("failed to classify remote gc sstables", K(ret)); } else if (OB_FAIL(check_remote_sstables(table_id, remote_major_sstables, remote_inc_sstables))) { LOG_WARN("failed to check remote sstables", K(ret), K(table_id)); } else if (OB_FAIL(build_migrate_major_sstable(need_reuse_local_minor, local_major_sstables, local_inc_sstables, remote_major_sstables, remote_inc_sstables, info.major_sstables_, part_ctx))) { LOG_WARN("failed to build migrate major sstable", K(ret)); } else if (OB_FAIL(build_migrate_minor_sstable(need_reuse_local_minor, local_inc_sstables, remote_inc_sstables, remote_gc_inc_sstables, info.minor_sstables_))) { LOG_WARN("failed to build migrate minor sstable", K(ret)); } else if (OB_FAIL(fill_log_ts_for_compat(info))) { LOG_WARN("Failed to fill new log ts fro compat", K(info)); } else if (OB_FAIL(check_can_reuse_sstable(pkey, info, part_ctx))) { LOG_WARN("failed to check can reuse sstable", K(ret)); } if (OB_SUCC(ret)) { info.multi_version_start_ = result.multi_version_start_; info.ready_for_read_ = result.is_ready_for_read_; FLOG_INFO("succeed build_migrate_table_info", K(result), K(info), K(local_major_sstables), K(local_inc_sstables), K(remote_gc_major_sstables), K(pkey), K(table_id), K(need_reuse_local_minor)); } return ret; } int ObMigratePrepareTask::fill_log_ts_for_compat(ObMigrateTableInfo& info) { int ret = OB_SUCCESS; if (!ctx_->is_migrate_compat_version()) { } else { ObTableCompater table_compater; if (OB_FAIL(table_compater.add_tables(info.major_sstables_))) { STORAGE_LOG(WARN, "Failed to add major sstables for compat", K(ret), K(info.major_sstables_)); } else if (OB_FAIL(table_compater.add_tables(info.minor_sstables_))) { STORAGE_LOG(WARN, "Failed to add minor sstables for compat", K(ret), K(info.minor_sstables_)); } else if (OB_FAIL(table_compater.fill_log_ts())) { STORAGE_LOG(WARN, "Failed to set new log ts for compat", K(ret), K(table_compater)); } else { ctx_->pg_meta_.storage_info_.get_data_info().set_last_replay_log_ts(ObTableCompater::OB_MAX_COMPAT_LOG_TS); } } return ret; } int ObMigratePrepareTask::check_remote_sstables(const uint64_t table_id, common::ObIArray& remote_major_sstables, common::ObIArray& remote_inc_tables) { int ret = OB_SUCCESS; bool need_check_major_sstable = true; const share::schema::ObTableSchema* table_schema = NULL; share::schema::ObSchemaGetterGuard schema_guard; share::schema::ObMultiVersionSchemaService& schema_service = share::schema::ObMultiVersionSchemaService::get_instance(); const uint64_t fetch_tenant_id = is_inner_table(table_id) ? OB_SYS_TENANT_ID : extract_tenant_id(table_id); if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (is_trans_table_id(table_id)) { need_check_major_sstable = false; } else if (OB_FAIL(schema_service.get_tenant_full_schema_guard(fetch_tenant_id, schema_guard))) { LOG_WARN("failed to get schema guard", K(ret)); } else if (OB_FAIL(schema_guard.get_table_schema(table_id, table_schema))) { LOG_WARN("Failed to get table schema", K(ret), K(table_id)); } else if (OB_ISNULL(table_schema)) { need_check_major_sstable = false; LOG_INFO("table schema is null, skip check major", K(table_id)); } else if (table_schema->is_index_table() && INDEX_STATUS_AVAILABLE != table_schema->get_index_status()) { need_check_major_sstable = false; LOG_INFO("index table schema is not available, skip check major", K(table_id)); } else if (REPLICA_NOT_RESTORE != ctx_->pg_meta_.is_restore_) { need_check_major_sstable = false; LOG_INFO("table is in restore status, skip check major", K(table_id)); } if (OB_SUCC(ret)) { if (remote_major_sstables.empty()) { if (need_check_major_sstable) { ret = OB_DATA_SOURCE_NOT_VALID; LOG_WARN("major remote table not found", K(ret), K(table_id)); } else if (remote_inc_tables.count() > 0) { } else { // no logic migrate, allow inc table be empty } } } if (OB_SUCC(ret) && remote_inc_tables.count() > 0) { if (OB_FAIL(check_remote_inc_sstables_continuity(remote_inc_tables))) { LOG_WARN( "failed to check remote inc sstables continuity", K(ret), K(remote_major_sstables), K(remote_inc_tables)); } } return ret; } int ObMigratePrepareTask::check_remote_inc_sstables_continuity(common::ObIArray& remote_inc_tables) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (OB_UNLIKELY(remote_inc_tables.empty())) { // skip empty inc tables } else { int64_t last_end_log_ts = remote_inc_tables.at(0).get_start_log_ts(); for (int64_t i = 0; OB_SUCC(ret) && i < remote_inc_tables.count(); ++i) { const ObITable::TableKey& remote_inc_table = remote_inc_tables.at(i); if (remote_inc_table.is_complement_minor_sstable()) { // skip buffer minor sstable and complement sstable } else if (remote_inc_table.get_start_log_ts() != last_end_log_ts) { ret = OB_DATA_SOURCE_NOT_VALID; LOG_WARN( "minor sstables not continue", K(ret), K(i), K(remote_inc_table), K(last_end_log_ts), K(remote_inc_tables)); } else { last_end_log_ts = remote_inc_table.get_end_log_ts(); } } } return ret; } // TODO only use source minor sstable now, consider performance later int ObMigratePrepareTask::build_remote_minor_sstables(const common::ObIArray& local_minor_sstables, const common::ObIArray& tmp_remote_minor_sstables, const common::ObIArray& tmp_remote_gc_minor_sstables, common::ObIArray& remote_minor_sstables, common::ObIArray& remote_gc_minor_sstables, bool& need_reuse_local_minor) { return OB_NOT_SUPPORTED; int ret = OB_SUCCESS; remote_minor_sstables.reuse(); remote_gc_minor_sstables.reuse(); need_reuse_local_minor = !ctx_->is_migrate_compat_version(); if (local_minor_sstables.empty()) { if (OB_FAIL(remote_minor_sstables.assign(tmp_remote_minor_sstables))) { LOG_WARN("failed to assign tables", K(ret)); } else if (OB_FAIL(remote_gc_minor_sstables.assign(tmp_remote_gc_minor_sstables))) { LOG_WARN("failed to assign tables", K(ret)); } } else if (need_reuse_local_minor) { const ObLogTsRange local_last_log_ts_range = local_minor_sstables.at(local_minor_sstables.count() - 1).log_ts_range_; for (int64_t i = 0; OB_SUCC(ret) && i < tmp_remote_minor_sstables.count(); ++i) { const ObITable::TableKey& remote_sstable = tmp_remote_minor_sstables.at(i); if (remote_minor_sstables.empty() && remote_sstable.log_ts_range_.start_log_ts_ > local_last_log_ts_range.end_log_ts_) { need_reuse_local_minor = false; break; } else if (remote_sstable.log_ts_range_.end_log_ts_ > local_last_log_ts_range.end_log_ts_) { if (OB_FAIL(remote_minor_sstables.push_back(remote_sstable))) { LOG_WARN("failed to push back remote sstable", K(ret), K(remote_sstable)); } } } if (OB_SUCC(ret)) { if (!need_reuse_local_minor) { if (OB_FAIL(remote_minor_sstables.assign(tmp_remote_minor_sstables))) { LOG_WARN("failed to assign remote minor sstable", K(ret), K(tmp_remote_minor_sstables)); } } else { for (int64_t i = 0; OB_SUCC(ret) && i < tmp_remote_gc_minor_sstables.count(); ++i) { const ObITable::TableKey& remote_sstable = tmp_remote_gc_minor_sstables.at(i); if (remote_sstable.log_ts_range_.end_log_ts_ > local_last_log_ts_range.end_log_ts_) { if (OB_FAIL(remote_gc_minor_sstables.push_back(remote_sstable))) { LOG_WARN("failed to push back remote sstable", K(ret), K(remote_sstable)); } } } } } } else { // migrate from compat version, only use remote sstable if (OB_FAIL(remote_minor_sstables.assign(tmp_remote_minor_sstables))) { LOG_WARN("failed to assign remote minor sstable", K(ret), K(tmp_remote_minor_sstables)); } } return ret; } int ObMigratePrepareTask::build_migrate_major_sstable(const bool need_reuse_local_minor, ObIArray& local_major_tables, ObIArray& local_inc_tables, ObIArray& remote_major_tables, ObIArray& remote_inc_tables, ObIArray& copy_sstables, ObPartitionMigrateCtx& part_ctx) { int ret = OB_SUCCESS; if (OB_FAIL(build_migrate_major_sstable_v2_(need_reuse_local_minor, local_major_tables, local_inc_tables, remote_major_tables, remote_inc_tables, copy_sstables, part_ctx))) { LOG_WARN("failed to build_migrate_major_sstable", K(ret)); } return ret; } int ObMigratePrepareTask::build_migrate_major_sstable_(ObIArray& local_major_tables, ObIArray& local_inc_tables, ObIArray& remote_major_tables, ObIArray& remote_inc_tables, ObIArray& copy_sstables) { int ret = OB_SUCCESS; // static func, skip check is_inited_ int64_t local_major_max_snapshot_version = 0; int64_t local_inc_max_snapshot_version = 0; int64_t local_max_snapshot_version = 0; int64_t max_snapshot_version = 0; int64_t remote_min_base_version = 0; int64_t remote_max_snapshot_version = 0; bool local_has_major = false; // D type replica major need to consider compaction_interval. Do local compaction when greater than 0, and meanwhile // need local major const int64_t follower_replica_merge_level = GCONF._follower_replica_merge_level; for (int64_t i = 0; i < local_major_tables.count(); ++i) { const ObITable::TableKey& local_major_table = local_major_tables.at(i); if (ObITable::is_major_sstable(local_major_table.table_type_)) { local_has_major = true; local_major_max_snapshot_version = local_major_table.trans_version_range_.snapshot_version_; } } if (local_inc_tables.count() > 0) { local_inc_max_snapshot_version = local_inc_tables.at(local_inc_tables.count() - 1).trans_version_range_.snapshot_version_; } local_max_snapshot_version = std::max(local_inc_max_snapshot_version, local_major_max_snapshot_version); if (remote_inc_tables.count() > 0) { remote_min_base_version = remote_inc_tables.at(0).trans_version_range_.base_version_; remote_max_snapshot_version = remote_inc_tables.at(remote_inc_tables.count() - 1).trans_version_range_.snapshot_version_; } if (local_max_snapshot_version >= remote_min_base_version) { max_snapshot_version = std::max(local_max_snapshot_version, remote_max_snapshot_version); } else { max_snapshot_version = local_max_snapshot_version; } for (int64_t i = 0; OB_SUCC(ret) && i < remote_major_tables.count(); ++i) { bool use_local = false; const ObITable::TableKey& remote_major_table = remote_major_tables.at(i); for (int64_t j = 0; OB_SUCC(ret) && j < local_major_tables.count() && !use_local; ++j) { if (local_major_tables.at(j) == remote_major_table) { use_local = true; LOG_INFO("use local major sstable", "table_key", remote_major_table); } } if (OB_SUCC(ret) && !use_local) { // check major tables can use minor sstable to merge if (local_has_major && max_snapshot_version >= remote_major_table.trans_version_range_.snapshot_version_ && is_follower_d_major_merge(follower_replica_merge_level)) { use_local = true; LOG_INFO("local table can merge become need major table, no need copy", K(local_inc_tables), K(remote_inc_tables), K(local_major_tables), "need table_key", remote_major_table); } } if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { use_local = false; } if (OB_SUCC(ret) && !use_local) { ObMigrateTableInfo::SSTableInfo info; info.src_table_key_ = remote_major_table; info.dest_base_version_ = remote_major_table.trans_version_range_.base_version_; info.dest_log_ts_range_ = remote_major_table.log_ts_range_; if (OB_FAIL(copy_sstables.push_back(info))) { LOG_WARN("failed to add major sstables", K(ret)); } else { ctx_->need_report_checksum_ = true; LOG_INFO("use remote major sstable", "table_key", remote_major_table); } } } if (OB_SUCC(ret)) { FLOG_INFO("build_migrate_major_sstable", K(local_major_tables), K(remote_major_tables), K(copy_sstables)); } return ret; } int ObMigratePrepareTask::build_migrate_major_sstable_v2_(const bool need_reuse_local_minor, ObIArray& local_major_tables, ObIArray& local_inc_tables, ObIArray& remote_major_tables, ObIArray& remote_inc_tables, ObIArray& copy_sstables, ObPartitionMigrateCtx& part_ctx) { int ret = OB_SUCCESS; int64_t max_snapshot_version = 0; UNUSED(local_inc_tables); UNUSED(remote_inc_tables); UNUSED(need_reuse_local_minor); bool need_add_local_major = false; // D type replica major need to consider compaction_interval. Do local compaction when greater than 0, and meanwhile // need local major const int64_t follower_replica_merge_level = GCONF._follower_replica_merge_level; for (int64_t i = 0; i < local_major_tables.count(); ++i) { const ObITable::TableKey& local_major_table = local_major_tables.at(i); if (ObITable::is_major_sstable(local_major_table.table_type_)) { max_snapshot_version = local_major_table.get_snapshot_version(); } } for (int64_t i = 0; OB_SUCC(ret) && i < remote_major_tables.count(); ++i) { const ObITable::TableKey& remote_major_table = remote_major_tables.at(i); if (remote_major_table.get_snapshot_version() <= max_snapshot_version && !remote_major_table.is_trans_sstable()) { need_add_local_major = true; continue; } else { ObMigrateTableInfo::SSTableInfo info; info.src_table_key_ = remote_major_table; info.dest_base_version_ = remote_major_table.get_base_version(); info.dest_log_ts_range_ = remote_major_table.log_ts_range_; if (OB_FAIL(copy_sstables.push_back(info))) { LOG_WARN("failed to push back sstable info", K(ret), K(info)); } else { ctx_->need_report_checksum_ = true; LOG_INFO("use remote major sstable", "table_key", remote_major_table); } } } if (OB_SUCC(ret) && need_add_local_major) { ObTableHandle table_handle; for (int64_t i = 0; OB_SUCC(ret) && i < local_major_tables.count(); ++i) { const ObITable::TableKey& local_major_table = local_major_tables.at(i); ObITable* table = NULL; table_handle.reset(); if (OB_FAIL(ObPartitionService::get_instance().acquire_sstable(local_major_table, table_handle))) { LOG_WARN("failed to get complete sstable by key", K(ret), K(local_major_table)); } else if (NULL == (table = table_handle.get_table())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("table should not be NULL", K(ret), KP(table)); } else if (OB_FAIL(part_ctx.add_sstable(*reinterpret_cast(table)))) { LOG_WARN("failed to add sstable", K(ret)); } } } if (OB_SUCC(ret)) { FLOG_INFO("build_migrate_major_sstable", K(local_major_tables), K(remote_major_tables), K(copy_sstables), K(need_reuse_local_minor)); } return ret; } int ObMigratePrepareTask::get_migrate_suitable_src(const common::ObIArray& src_info_array, const ObReplicaOpArg& arg, IsValidSrcFunc is_valid_src, bool& find_suitable_src, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; find_suitable_src = false; pg_meta.reset(); src_info.reset(); ObPartitionGroupInfoResult data_src_result; ObPartitionGroupInfoResult tmp_data_src_result; int64_t max_last_replay_log_id = 0; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (src_info_array.empty() || !arg.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("get migrate suitable src get invalid argument", K(ret), K(src_info_array), K(arg)); } else { // skip ret, try all servers for (int64_t i = 0; i < src_info_array.count(); ++i) { const ObMigrateSrcInfo& tmp_src_info = src_info_array.at(i); bool is_src_valid = true; int64_t remote_last_replay_log_id = 0; tmp_data_src_result.reset(); if (MYADDR == tmp_src_info.src_addr_) { // do nothing } else if (OB_FAIL(fetch_partition_group_info(arg, tmp_src_info, tmp_data_src_result))) { LOG_WARN("failed to fetch data src result", K(ret), K(arg), K(tmp_src_info)); ret = OB_SUCCESS; } else if (OB_FAIL(is_valid_src(tmp_data_src_result.result_, *ctx_, is_src_valid))) { LOG_WARN("Failed to check is valid migrate src", K(ret), K(arg.data_src_), K(tmp_data_src_result)); } else if (!is_src_valid) { STORAGE_LOG(INFO, "migrate src is not suitable, skip it", K(arg), K(tmp_src_info)); } else if (FALSE_IT( remote_last_replay_log_id = tmp_data_src_result.result_.pg_meta_.storage_info_.get_clog_info().get_last_replay_log_id())) { } else if (max_last_replay_log_id <= remote_last_replay_log_id) { data_src_result.reset(); if (OB_FAIL(data_src_result.assign(tmp_data_src_result))) { LOG_WARN("failed to assign partition info result", K(ret), K(tmp_data_src_result)); } else { max_last_replay_log_id = remote_last_replay_log_id; find_suitable_src = true; } } } if (OB_SUCC(ret) && find_suitable_src) { if (OB_FAIL(pg_meta.deep_copy(data_src_result.result_.pg_meta_))) { LOG_WARN("Failed to copy pg meta", K(ret)); } else if (OB_FAIL(ctx_->recovery_point_ctx_.recovery_point_key_array_.assign( data_src_result.result_.recovery_point_key_array_))) { LOG_WARN("failed to assgin recovery point key array", K(ret), K(data_src_result)); } else if (OB_FAIL(ctx_->set_is_restore_for_add_replica(pg_meta.is_restore_))) { LOG_WARN("Failed to set is restore", K(ret), K(pg_meta)); } else { ctx_->mig_src_file_id_ = data_src_result.result_.pg_file_id_; ctx_->fetch_pg_info_compat_version_ = data_src_result.result_.compat_version_; src_info = data_src_result.choose_src_info_; LOG_INFO("succeed to choose_ob_migrate_src", K(ret), K(arg), K(pg_meta)); } } } return ret; } int ObMigratePrepareTask::get_minor_src_candidate_with_region( const ObReplicaOpArg& arg, ObIArray& src_info_array) { int ret = OB_SUCCESS; ObArray tmp_info_array; ObArray same_region_array; ObArray diff_region_array; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (OB_FAIL(partition_service_->get_migrate_member_list_src(arg.key_, tmp_info_array))) { LOG_WARN("failed to get migrate src", K(ret), "pkey", arg.key_); } else if (tmp_info_array.empty()) { LOG_INFO("migrate meta table src is empty"); } else if (OB_FAIL(split_candidate_with_region(tmp_info_array, same_region_array, diff_region_array))) { LOG_WARN("failed to split candidate with region", K(ret), K(arg)); } else if (OB_FAIL(src_info_array.assign(same_region_array))) { LOG_WARN("failed to assign src info array", K(ret)); } else { LOG_INFO("get minor src with region", K(tmp_info_array), K(same_region_array)); } return ret; } int ObMigratePrepareTask::get_minor_src_candidate_without_region( const ObReplicaOpArg& arg, common::ObIArray& src_info_array) { int ret = OB_SUCCESS; src_info_array.reset(); bool found_recommendable_src = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (OB_FAIL(partition_service_->get_migrate_leader_and_parent(arg.key_, src_info_array))) { LOG_WARN("failed to get migrate src", K(ret), "pkey", arg.key_); } // skip get migrate leader and parent ret, overwrite it { ret = OB_SUCCESS; for (int64_t i = 0; i < src_info_array.count() && !found_recommendable_src; ++i) { const ObMigrateSrcInfo& src_info = src_info_array.at(i); if (arg.data_src_.get_server() == src_info.src_addr_) { found_recommendable_src = true; } } if (!found_recommendable_src) { ObMigrateSrcInfo recommendable_src_info; recommendable_src_info.src_addr_ = arg.data_src_.get_server(); recommendable_src_info.cluster_id_ = arg.cluster_id_; if (OB_FAIL(src_info_array.push_back(recommendable_src_info))) { LOG_WARN("failed to push migrate src into array", K(ret), K(recommendable_src_info)); } } } LOG_INFO("get minor src candidate without region", K(ret), K(src_info_array)); return ret; } int ObMigratePrepareTask::choose_recommendable_src( const ObReplicaOpArg& arg, ObPartitionGroupMeta& pg_meta, ObMigrateSrcInfo& src_info) { int ret = OB_SUCCESS; ObPartitionGroupInfoResult src_result; ObMigrateSrcInfo tmp_src_info; tmp_src_info.src_addr_ = arg.data_src_.get_server(); tmp_src_info.cluster_id_ = arg.cluster_id_; bool need_recovery_point = true; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(fetch_partition_group_info(arg, tmp_src_info, src_result))) { LOG_WARN("failed to fetch partition info", K(ret), K(arg)); } else if (ctx_->is_only_copy_sstable()) { if (!ObReplicaTypeCheck::is_replica_with_ssstore(src_result.result_.pg_meta_.replica_type_)) { ret = OB_DATA_SOURCE_NOT_EXIST; LOG_WARN("src not has ssstore, cannot copy sstable", K(src_result)); } else { need_recovery_point = false; } } if (OB_FAIL(ret)) { } else if (OB_FAIL(pg_meta.deep_copy(src_result.result_.pg_meta_))) { LOG_WARN("Failed to copy pg meta", K(ret)); } else { ctx_->mig_src_file_id_ = src_result.result_.pg_file_id_; ctx_->fetch_pg_info_compat_version_ = src_result.result_.compat_version_; src_info = tmp_src_info; LOG_INFO("succeed to choose_recommendable src", K(ret), K(arg), K(pg_meta)); } return ret; } int ObMigratePrepareTask::build_migrate_minor_sstable(const bool need_reuse_local_minor, ObIArray& local_inc_tables, ObIArray& remote_inc_tables, ObIArray& remote_gc_inc_sstables, ObIArray& copy_sstables) { int ret = OB_SUCCESS; // ObITable::Tablekey's table_key.version is mark major version flag which is smaller than copy sstables need to // remove remote'version is monotonically increasing continuously, and there is only one minor sstable for each // version there are some minor sstables for on version in local // TODO() errsim case may not pass if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (ctx_->is_copy_index() || LINK_SHARE_MAJOR_OP == ctx_->replica_op_arg_.type_) { // copy index only need major sstable } else if (remote_inc_tables.empty()) { LOG_INFO("remote inc table is empty, skip it", K(copy_sstables), K(local_inc_tables), K(remote_inc_tables), K(remote_gc_inc_sstables)); } else if (!need_reuse_local_minor) { ObMigrateTableInfo::SSTableInfo info; for (int64_t i = 0; OB_SUCC(ret) && i < remote_inc_tables.count(); ++i) { info.reset(); const ObITable::TableKey& remote_table = remote_inc_tables.at(i); info.src_table_key_ = remote_table; info.dest_base_version_ = remote_table.trans_version_range_.base_version_; info.dest_log_ts_range_ = remote_table.log_ts_range_; if (OB_FAIL(copy_sstables.push_back(info))) { LOG_WARN("failed to push back sstable info", K(ret), K(info)); } } } else { // TODO reuse local minor with continuous log ts range ret = OB_NOT_SUPPORTED; LOG_WARN("Reusing local minor sstable within migration is not supported", K(ret)); } if (OB_SUCC(ret)) { LOG_INFO("build_migrate_minor_sstable", K(copy_sstables), K(local_inc_tables), K(remote_inc_tables), K(remote_gc_inc_sstables)); } return ret; } int ObMigratePrepareTask::check_can_reuse_sstable( const ObPartitionKey& pkey, ObMigrateTableInfo& table_info, ObPartitionMigrateCtx& part_ctx) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { } else if (NULL == (ctx_->partition_guard_.get_partition_group())) { LOG_INFO("local partition not exist, no thing reuse", K(table_info)); // skip backup } else { // check major sstable bool is_reuse = false; for (int64_t i = table_info.major_sstables_.count() - 1; OB_SUCC(ret) && i >= 0; --i) { is_reuse = false; if (OB_FAIL(check_and_reuse_sstable(pkey, table_info.major_sstables_.at(i).src_table_key_, is_reuse, part_ctx))) { LOG_WARN("failed to check can reuse sstable", K(ret), "table_key", table_info.major_sstables_.at(i)); } else if (!is_reuse) { // do nothing } else if (OB_FAIL(table_info.major_sstables_.remove(i))) { LOG_WARN("failed to remove reused table", K(ret), "table_key", table_info.major_sstables_.at(i)); } } if (OB_SUCC(ret)) { // check minor sstable for (int64_t i = table_info.minor_sstables_.count() - 1; OB_SUCC(ret) && i >= 0; --i) { is_reuse = false; ObITable::TableKey table_key; if (OB_FAIL(ObDestTableKeyManager::convert_sstable_info_to_table_key( table_info.minor_sstables_.at(i), table_key))) { LOG_WARN("failed to convert sstable info to table key", K(ret), K(table_info)); } else if (OB_FAIL(check_and_reuse_sstable(pkey, table_key, is_reuse, part_ctx))) { LOG_WARN("failed to check can reuse sstable", K(ret), "table_key", table_key); } else if (!is_reuse) { // do nothing } else if (OB_FAIL(table_info.minor_sstables_.remove(i))) { LOG_WARN("failed to remove reused table", K(ret), "table_key", table_key); } } } } return ret; } int ObMigrateUtil::get_report_result(const common::ObIArray& report_list, common::ObIArray& report_res_list) { int ret = OB_SUCCESS; ObPartMigrationRes tmp_res; if (report_list.count() <= 0) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "report list must not empty", K(ret)); } else if (OB_FAIL(report_res_list.reserve(report_list.count()))) { STORAGE_LOG(WARN, "failed to reserve report_res_list", K(ret), K(report_list)); } else { ObReplicaOpType type = report_list.at(0).arg_.type_; for (int64_t i = 0; OB_SUCC(ret) && i < report_list.count(); ++i) { const ObReplicaOpArg& arg = report_list.at(i).arg_; if (type != arg.type_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(ERROR, "op type not same", K(ret), K(type), K(i), K(arg), K(report_list)); } else { tmp_res.key_ = arg.key_; tmp_res.src_ = arg.src_; tmp_res.dst_ = arg.dst_; tmp_res.data_src_ = arg.data_src_; tmp_res.quorum_ = arg.quorum_; tmp_res.backup_arg_ = arg.backup_arg_; tmp_res.validate_arg_ = arg.validate_arg_; tmp_res.backup_backupset_arg_ = arg.backup_backupset_arg_; tmp_res.backup_archivelog_arg_ = arg.backup_archive_log_arg_; tmp_res.data_statics_ = report_list.at(i).data_statics_; tmp_res.result_ = report_list.at(i).result_; if (OB_FAIL(report_res_list.push_back(tmp_res))) { STORAGE_LOG(WARN, "failed to add report res list", K(ret)); } } } } return ret; } // for split reuse, because dest pkey may not eqaul to table key's pkey int ObMigratePrepareTask::check_and_reuse_sstable( const ObPartitionKey& pkey, const ObITable::TableKey& table_key, bool& is_reuse, ObPartitionMigrateCtx& part_ctx) { int ret = OB_SUCCESS; ObTableHandle table_handle; const bool in_slog_trans = false; ObITable* table = NULL; is_reuse = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (!pkey.is_valid() || !table_key.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("table key is invalid", K(ret), K(pkey), K(table_key)); } else if (table_key.is_memtable()) { LOG_INFO("memtable no need check and reuse, skip it", K(table_key)); } else if (table_key.is_trans_sstable()) { LOG_INFO("trans sstable can not reuse, skip it", K(table_key)); } else if (!table_key.is_table_log_ts_comparable()) { // old minor sstable always need to migrate to fill log ts LOG_INFO("old minor sstable can not reuse, skip it", K(table_key)); } else { if (OB_FAIL(ObPartitionService::get_instance().acquire_sstable(table_key, table_handle))) { if (OB_ENTRY_NOT_EXIST != ret && OB_PARTITION_NOT_EXIST != ret) { LOG_WARN("failed to get complete sstable by key", K(ret), K(table_key)); } else { is_reuse = false; ret = OB_SUCCESS; } } else if (NULL == (table = table_handle.get_table())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("table should not be NULL", K(ret), KP(table)); } else if (OB_FAIL(part_ctx.add_sstable(*reinterpret_cast(table)))) { LOG_WARN("failed to add sstable", K(ret)); } else { is_reuse = true; LOG_INFO("table has already been in table mgr", K(table_key)); } } return ret; } int ObMigratePrepareTask::prepare_new_partition() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObStorageFileHandle file_handle; ObPartGroupMigrationTask* group_task = NULL; bool has_mark_creating = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (NULL != ctx_->get_partition()) { ret = OB_ERR_SYS; LOG_ERROR("cannot prepare new partition for exist partition", K(ret), K(*ctx_), K(ctx_->get_partition())); } else if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_PARTITION_NOT_EXIST; LOG_ERROR("cannot create new partition for rebuild", K(ret), K(*ctx_)); } else if (ctx_->is_only_copy_sstable()) { ret = OB_PARTITION_NOT_EXIST; LOG_WARN("partition not exist, cannot copy only sstable", K(ret)); } else if (OB_FAIL(OB_SERVER_FILE_MGR.alloc_file(ctx_->replica_op_arg_.key_.get_tenant_id(), common::is_sys_table(ctx_->replica_op_arg_.key_.get_table_id()), file_handle))) { STORAGE_LOG(WARN, "fail to alloc file", K(ret)); } else if (OB_ISNULL(file_handle.get_storage_file())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("storage file is null", K(ret), K(file_handle)); } else if (OB_UNLIKELY(0 >= (ctx_->mig_dest_file_id_ = file_handle.get_storage_file()->get_file_id()))) { ret = OB_ERR_UNEXPECTED; LOG_WARN("invalid file id", K(ret), K(ctx_->mig_dest_file_id_)); } else if (OB_FAIL(SLOGGER.begin(OB_LOG_CS_MIGRATE_PARTITION))) { STORAGE_LOG(WARN, "fail to begin commit log.", K(ret)); } else if (OB_FAIL(partition_service_->mark_pg_creating(ctx_->replica_op_arg_.key_))) { STORAGE_LOG(WARN, "fail to mark pg creating.", K(ret)); } else if (FALSE_IT(has_mark_creating = true)) { } else if (OB_FAIL(create_new_partition( ctx_->migrate_src_info_.src_addr_, ctx_->replica_op_arg_, ctx_->partition_guard_, file_handle))) { STORAGE_LOG(WARN, "failed to create new partition", K(ret)); } else if (OB_ISNULL(ctx_->get_partition()) || OB_ISNULL(&(ctx_->get_partition()->get_pg_storage()))) { ret = OB_ERR_SYS; LOG_ERROR("partition or storage must not null", K(ctx_->get_partition()), K(ret)); } else if (FALSE_IT(group_task = reinterpret_cast(ctx_->group_task_))) { } else if (OB_FAIL(group_task->set_create_new_pg(ctx_->get_partition()->get_partition_key()))) { LOG_WARN("failed to set create new pg", K(ret)); } else { ctx_->need_offline_ = true; ctx_->during_migrating_ = true; LOG_INFO("create new partition for migrate, record migrate start", K(ctx_->replica_op_arg_)); } if (OB_FAIL(ret)) { // partition_service_->add_new_parititoon must have commit log or abort log if (OB_SUCCESS != (tmp_ret = SLOGGER.abort())) { STORAGE_LOG(WARN, "fail to abort log.", K(tmp_ret)); } } else if (OB_FAIL(partition_service_->add_new_partition(ctx_->partition_guard_))) { STORAGE_LOG(WARN, "add partition to manager failed", K(ctx_->replica_op_arg_), K(ret)); } if (has_mark_creating) { partition_service_->mark_pg_created(ctx_->replica_op_arg_.key_); } if (OB_FAIL(ret) && NULL != ctx_) { ctx_->can_rebuild_ = false; LOG_WARN("failed during prepare_new_partition, set cannot rebuild", K(ret), K(*ctx_)); } return ret; } int ObMigrateUtil::enable_replay_with_new_partition(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; LOG_INFO("enable_replay_with_new_partition"); if (OB_ISNULL(partition = ctx.get_partition())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("partition should not be null", K(ret)); } else if (OB_FAIL(partition->create_memtable())) { LOG_WARN("failed to create memtable", K(ret), "pkey", partition->get_partition_key()); } else if (OB_FAIL(ObPartitionService::get_instance().log_new_partition(partition, ctx.pg_meta_.storage_info_.get_data_info().get_publish_version(), ctx.pg_meta_.restore_snapshot_version_, ctx.pg_meta_.last_restore_log_id_, ctx.pg_meta_.last_restore_log_ts_))) { STORAGE_LOG(WARN, "add partition to manager failed", K(ctx.replica_op_arg_), K(ret)); } if (OB_SUCC(ret)) { if (ADD_REPLICA_OP == ctx.replica_op_arg_.type_ && (REPLICA_RESTORE_DATA == ctx.is_restore_ || REPLICA_RESTORE_ARCHIVE_DATA == ctx.is_restore_ || REPLICA_RESTORE_STANDBY == ctx.is_restore_)) { ctx.need_offline_ = false; LOG_INFO("no need set clog online for add replica during restore", "pkey", ctx.replica_op_arg_.key_, "restore_state", ctx.is_restore_); } else if (OB_FAIL(partition->get_log_service()->set_online( ctx.pg_meta_.storage_info_.get_clog_info(), ObVersion(0, 0) /*freeze version*/))) { STORAGE_LOG(WARN, "failed to set clog online", K(ret), K(ctx)); } else if (OB_FAIL(partition->report_clog_history_online())) { LOG_WARN("failed to report clog history", K(ret)); } else { STORAGE_LOG(INFO, "succeed to enable replay for new partition", K(ctx.pg_meta_.storage_info_)); } } if (OB_FAIL(ret)) { ctx.can_rebuild_ = false; LOG_WARN("failed during enable_replay_with_new_partition, set cannot rebuild", K(ret), K(ctx)); } return ret; } int ObMigrateTaskGeneratorTask::deal_with_old_partition() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; clog::ObIPartitionLogService* pls = NULL; const bool write_slog = true; bool is_log_sync = false; LOG_INFO("start enable replay"); if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_ISNULL(partition = ctx_->get_partition())) { ret = OB_ERR_SYS; LOG_ERROR("cannot use old partition", K(ret)); } else if (NULL == (pls = partition->get_log_service())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("Unexpected error, pls is NULL, ", K(ret), KP(pls)); } else if (OB_FAIL(pls->is_log_sync_with_leader(is_log_sync))) { LOG_WARN("failed to get log is in sync", K(ret)); } else if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; LOG_ERROR("cannot enable_replay_with_old_partition for rebuild", K(ret), K(*ctx_)); } else { bool is_replica_with_data = true; common::ObReplicaType replica_type = ctx_->replica_op_arg_.dst_.get_replica_type(); common::ObReplicaType local_replica_type = partition->get_replica_type(); const ObReplicaProperty& local_replica_property = partition->get_replica_property(); ctx_->need_offline_ = true; const ObPartitionSplitInfo& split_info = ctx_->pg_meta_.split_info_; if (split_info.is_valid() && OB_FAIL(partition->save_split_info(split_info))) { LOG_WARN("failed to save split info", K(ret), K(split_info), K(*ctx_)); } else if (OB_FAIL(update_multi_version_start())) { LOG_WARN("failed to update multi version start", K(ret), K(*ctx_)); } else if (local_replica_type != replica_type && !ObReplicaTypeCheck::change_replica_op_allow(local_replica_type, replica_type)) { ret = OB_OP_NOT_ALLOW; STORAGE_LOG(WARN, "change replica op not allow", K(ret), K(local_replica_type), K(replica_type), K(*ctx_)); } else if (ctx_->replica_op_arg_.is_physical_restore()) { if (RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && RESTORE_FOLLOWER_REPLICA_OP != ctx_->replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; LOG_WARN( "replica op type is not restore or follower restore or standby restore", K(ret), K(ctx_->replica_op_arg_)); } else { ctx_->need_offline_ = false; if (OB_FAIL(partition->set_replica_type(replica_type, write_slog))) { LOG_WARN("Failed to set replica type", K(ret), K(ctx_->replica_op_arg_)); } else { ctx_->during_migrating_ = true; LOG_INFO("record migrate start", K(ctx_->replica_op_arg_)); } } } else if (is_log_sync && (0 != local_replica_property.get_memstore_percent() || (0 == local_replica_property.get_memstore_percent() && ObReplicaTypeCheck::is_replica_with_memstore(local_replica_type))) && (OB_NORMAL_REPLICA == ctx_->replica_state_ || OB_RESTORE_REPLICA == ctx_->replica_state_) && ADD_REPLICA_OP != ctx_->replica_op_arg_.type_ && MIGRATE_REPLICA_OP != ctx_->replica_op_arg_.type_ && FAST_MIGRATE_REPLICA_OP != ctx_->replica_op_arg_.type_) { ctx_->need_offline_ = false; LOG_INFO("reuse memtable", K(ctx_->replica_op_arg_), K(is_log_sync)); if (OB_FAIL(partition->set_replica_type(replica_type, write_slog))) { LOG_WARN("Failed to set replica type", K(ret), K(ctx_->replica_op_arg_)); } else { ctx_->during_migrating_ = true; LOG_INFO("record migrate start", K(ctx_->replica_op_arg_)); } } else if (ctx_->is_only_copy_sstable()) { ret = OB_LOG_NOT_SYNC; LOG_WARN("log is not sync, cannot copy only sstable now", K(ret), K(is_log_sync), K(*ctx_)); } else if (OB_FAIL(ctx_->change_replica_with_data(is_replica_with_data))) { LOG_WARN("failed to check replica", K(ret), K(*ctx_)); } else if (is_replica_with_data) { ret = OB_LOG_NOT_SYNC; LOG_WARN("log is not sync, cannot change replica now", K(is_log_sync), K(*ctx_)); } else if (OB_FAIL(partition->set_replica_type(replica_type, write_slog))) { LOG_WARN("Failed to set replica type", K(ret), K(ctx_->replica_op_arg_)); } else { ctx_->during_migrating_ = true; LOG_INFO("record migrate start", K(ctx_->replica_op_arg_)); if (OB_FAIL(partition->pause())) { LOG_WARN("failed to pause partition", K(ret)); } } } return ret; } int ObMigrateUtil::enable_replay_with_old_partition(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; LOG_INFO("start enable replay"); if (!ctx.is_valid()) { ret = OB_NOT_INIT; LOG_ERROR("ctx not valid", K(ret), K(ctx)); } else if (OB_ISNULL(partition = ctx.get_partition())) { ret = OB_ERR_SYS; LOG_ERROR("cannot use old partition", K(ret)); } else { int64_t retry_times = 0; ObPartitionGroupMeta& meta = ctx.pg_meta_; ObBaseStorageInfo& clog_info = meta.storage_info_.get_clog_info(); ObDataStorageInfo& data_info = meta.storage_info_.get_data_info(); int64_t restore_snapshot_version = meta.restore_snapshot_version_; uint64_t last_restore_log_id = meta.last_restore_log_id_; int64_t last_restore_log_ts = meta.last_restore_log_ts_; const int64_t local_last_replay_log_id = partition->get_log_service()->get_next_index_log_id() - 1; const int64_t src_last_replay_log_id = std::max(clog_info.get_last_replay_log_id(), data_info.get_last_replay_log_id()); if (src_last_replay_log_id <= local_last_replay_log_id && REBUILD_REPLICA_OP == ctx.replica_op_arg_.type_ && partition->get_pg_storage().is_replica_with_remote_memstore()) { if (OB_FAIL(partition->set_storage_info(meta.storage_info_))) { LOG_WARN("failed to set storage info for migration", K(ret), "pkey", partition->get_partition_key()); } LOG_INFO("local last replay log id is new enough, no need offline and online pg", K(local_last_replay_log_id), K(src_last_replay_log_id)); } else { while (OB_SUCC(ret) && retry_times < OB_MIGRATE_ONLINE_RETRY_COUNT) { if (OB_FAIL(partition->pause())) { LOG_WARN("failed to pause partition", K(ret), "pkey", partition->get_partition_key()); } else if (OB_FAIL(partition->set_storage_info(meta.storage_info_))) { LOG_WARN("failed to set storage info", K(ret), "pkey", partition->get_partition_key()); } else if (OB_FAIL(partition->create_memtable())) { LOG_WARN("failed to create memtable", K(ret), "pkey", partition->get_partition_key()); } else if (OB_FAIL(ctx.get_partition()->get_log_service()->migrate_set_base_storage_info(clog_info))) { STORAGE_LOG(WARN, "reset clog start point fail", K(ret), K(meta.storage_info_)); } else if (OB_FAIL(MIGRATOR.get_partition_service()->online_partition(ctx.replica_op_arg_.key_, data_info.get_publish_version(), restore_snapshot_version, last_restore_log_id, last_restore_log_ts))) { STORAGE_LOG(WARN, "online partition failed", K(ctx.replica_op_arg_), K(meta), K(ret)); } else if (OB_PERMANENT_OFFLINE_REPLICA == ctx.replica_state_ && F_WORKING != ctx.get_partition()->get_partition_state() && OB_FAIL(ctx.get_partition()->switch_partition_state(F_WORKING))) { STORAGE_LOG(WARN, "partition set partition state F_WORKING failed", K(ctx.replica_op_arg_.key_), K(ret)); } ++retry_times; if (OB_FAIL(ret) && retry_times < OB_MIGRATE_ONLINE_RETRY_COUNT) { ret = OB_SUCCESS; usleep(RETRY_TASK_SLEEP_INTERVAL_S); } else { STORAGE_LOG(INFO, "online partition finish", K(ret), K(ctx.replica_op_arg_.key_), K(meta.storage_info_)); break; } } if (OB_SUCC(ret)) { if (OB_FAIL(ctx.get_partition()->get_log_service()->restore_replayed_log(clog_info))) { STORAGE_LOG(WARN, "restore and replay log failed.", K(ret), K_(ctx.replica_op_arg_.key), K(meta)); } else if (OB_FAIL(ctx.get_partition()->get_log_service()->set_online( clog_info, ObVersion(0, 0) /*freeze version*/))) { STORAGE_LOG(WARN, "reset log temporary status failed.", K(ret), K_(ctx.replica_op_arg_.key), K(meta)); } else if (OB_FAIL(ctx.get_partition()->report_clog_history_online())) { LOG_WARN("failed to report clog history online", K(ret)); } else { STORAGE_LOG(INFO, "succeed to enable replay for exist partition"); } } } } return ret; } int ObMigrateUtil::push_reference_tables_if_need( ObMigrateCtx& ctx, const ObPartitionSplitInfo& split_info, const int64_t last_replay_log_id) { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; if (!ctx.is_valid()) { ret = OB_NOT_INIT; LOG_ERROR("ctx not valid", K(ret), K(ctx)); } else if (OB_ISNULL(partition = ctx.get_partition())) { ret = OB_ERR_SYS; LOG_ERROR("cannot use old partition", K(ret)); } else { ObPartitionGroupMeta& meta = ctx.pg_meta_; if (split_info.get_src_partition() == meta.pg_key_) { const int64_t source_log_id = split_info.get_source_log_id(); if (last_replay_log_id >= source_log_id && OB_INVALID_ID != source_log_id) { if (OB_FAIL( partition->push_reference_tables(split_info.get_dest_partitions(), split_info.get_split_version()))) { LOG_WARN("failed to push reference tables", K(split_info)); } } } } return ret; } int ObMigrateUtil::merge_trans_table(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; ObIPartitionGroup* pg = nullptr; ObSSTable* trans_sstable = nullptr; ObPartitionKey pg_key; bool is_replica_with_data = false; if (OB_ISNULL(trans_sstable = static_cast(ctx.trans_table_handle_.get_table()))) { FLOG_INFO("trans sstable is empty", K(ret), K(ctx)); if (need_migrate_trans_table(ctx.replica_op_arg_.type_) && ctx.is_migrate_compat_version()) { if (OB_FAIL(ObMigrateUtil::create_empty_trans_sstable_for_compat(ctx))) { LOG_WARN("failed to create empty trans sstable for compat", K(ret), K(ctx)); } } else if (LINK_SHARE_MAJOR_OP == ctx.replica_op_arg_.type_) { ctx.old_trans_table_seq_ = ctx.get_partition()->get_pg_storage().get_trans_table_seq(); } } else if (OB_FAIL(wait_trans_table_merge_finish(ctx))) { LOG_WARN("failed to wait trans table merge finish", K(ret), K(ctx)); } else if (OB_ISNULL(pg = ctx.get_partition())) { ret = OB_ERR_SYS; LOG_WARN("partition group must not null", K(ret)); } else if (FALSE_IT(pg_key = pg->get_partition_key())) { } else if (FALSE_IT(ctx.old_trans_table_seq_ = pg->get_pg_storage().get_trans_table_seq())) { } else if (OB_FAIL( ObMigrateUtil::add_trans_sstable_to_part_ctx(trans_sstable->get_key().pkey_, *trans_sstable, ctx))) { LOG_WARN("failed to add trans sstable to part migrate ctx", K(ret), KPC(trans_sstable)); } // TODO() need reconsider sstable reuse, now skip trans sstable reuse /* else { ObTransTableMergeTask merge_task; ObMacroBlockWriter writer; ObTableHandle new_trans_sstable; ObSSTable* new_sstable = nullptr; ctx.old_trans_table_seq_ = pg->get_pg_storage().get_trans_table_seq(); if (OB_FAIL(merge_task.init(pg_key, trans_sstable))) { LOG_WARN("failed to init merge task", K(ret), K(pg_key)); } else if (OB_FAIL(merge_task.init_schema_and_writer(writer))) { LOG_WARN("failed to init schema and writer", K(ret), K(pg_key)); } else if (OB_FAIL(merge_task.merge_trans_table(writer))) { LOG_WARN("failed to merge trans table", K(ret), K(pg_key)); } else if (OB_FAIL(merge_task.get_merged_trans_sstable(new_trans_sstable, writer))) { LOG_WARN("failed to get merged trans sstable", K(ret), K(pg_key)); } else if (OB_FAIL(new_trans_sstable.get_sstable(new_sstable))) { LOG_WARN("failed to get new trans sstable", K(ret), K(pg_key)); } else if (OB_FAIL(ObMigrateUtil::add_trans_sstable_to_part_ctx(new_sstable->get_key().pkey_, *new_sstable, ctx))) { LOG_WARN("failed to add trans sstable to part migrate ctx", K(ret), KPC(new_sstable)); } } */ return ret; } int ObMigrateUtil::create_empty_trans_sstable_for_compat(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; ObTableSchema table_schema; ObPartitionKey trans_table_pkey; const int64_t SCHEMA_VERSION = 1; if (OB_FAIL(trans_table_pkey.generate_trans_table_pkey(ctx.pg_meta_.pg_key_))) { LOG_WARN("failed to generate trans table pkey", K(ret), K(ctx)); } else if (OB_FAIL( table_schema.generate_kv_schema(trans_table_pkey.get_tenant_id(), trans_table_pkey.get_table_id()))) { LOG_WARN("failed to generate trans table schema", K(ret), K(trans_table_pkey)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(ctx.replica_op_arg_.dst_.get_replica_type())) { LOG_INFO("no need create_empty_trans_sstable_for_compat without ssstore", "pg_key", ctx.pg_meta_.pg_key_, "type", ctx.replica_op_arg_.dst_.get_replica_type()); } else { ObCreateSSTableParamWithTable param; ObITable::TableKey table_key; ObSSTable* sstable = nullptr; ObTableHandle table_handle; table_key.table_type_ = ObITable::TableType::TRANS_SSTABLE; table_key.table_id_ = trans_table_pkey.get_table_id(); table_key.trans_version_range_.multi_version_start_ = 0; table_key.trans_version_range_.base_version_ = 0; table_key.trans_version_range_.snapshot_version_ = ObTimeUtility::current_time(); table_key.log_ts_range_.start_log_ts_ = 0; table_key.log_ts_range_.end_log_ts_ = ctx.pg_meta_.storage_info_.get_data_info().get_last_replay_log_ts(); table_key.log_ts_range_.max_log_ts_ = table_key.log_ts_range_.end_log_ts_; table_key.pkey_ = trans_table_pkey; param.table_key_ = table_key; param.schema_ = &table_schema; param.schema_version_ = SCHEMA_VERSION; param.logical_data_version_ = table_key.version_ + 1; ObPGCreateSSTableParam pg_create_sstable_param; pg_create_sstable_param.with_table_param_ = ¶m; if (OB_FAIL(ctx.get_partition()->create_sstable(pg_create_sstable_param, table_handle))) { LOG_WARN("fail to create sstable", K(ret)); } else if (OB_FAIL(table_handle.get_sstable(sstable))) { LOG_WARN("failed to get sstable", K(ret), K(table_handle)); } else if (OB_ISNULL(sstable)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("sstable should not be null", K(ret)); } else if (OB_FAIL(ObMigrateUtil::add_trans_sstable_to_part_ctx(trans_table_pkey, *sstable, ctx))) { LOG_WARN("failed to add trans sstable to part ctx", K(ret), K(trans_table_pkey)); } } if (OB_SUCC(ret)) { ctx.old_trans_table_seq_ = ctx.get_partition()->get_pg_storage().get_trans_table_seq(); } return ret; } int ObMigrateUtil::add_trans_sstable_to_part_ctx( const ObPartitionKey& trans_table_pkey, ObSSTable& sstable, ObMigrateCtx& ctx) { int ret = OB_SUCCESS; bool found = false; for (int64_t i = 0; OB_SUCC(ret) && i < ctx.part_ctx_array_.count(); ++i) { ObPartitionMigrateCtx& part_ctx = ctx.part_ctx_array_.at(i); if (part_ctx.copy_info_.meta_.pkey_.is_trans_table()) { if (OB_FAIL(part_ctx.add_sstable(sstable))) { LOG_WARN("failed to add new trans sstable to partition migrate ctx", K(ret), K(trans_table_pkey)); } else { FLOG_INFO("succeed to add trans sstable to partition migrate ctx", K(ret), K(sstable)); } found = true; break; } } if (OB_SUCC(ret) && !found) { ObPartitionMigrateCtx part_ctx; part_ctx.copy_info_.meta_.pkey_ = trans_table_pkey; part_ctx.copy_info_.meta_.multi_version_start_ = 1; part_ctx.copy_info_.meta_.create_timestamp_ = ObTimeUtility::current_time(); part_ctx.is_partition_exist_ = false; part_ctx.need_reuse_local_minor_ = false; part_ctx.ctx_ = &ctx; if (OB_FAIL(part_ctx.add_sstable(sstable))) { LOG_WARN("failed to add sstable to partition ctx", K(ret), K(sstable)); } else if (OB_FAIL(ctx.part_ctx_array_.push_back(part_ctx))) { LOG_WARN("failed to push back migrate partition ctx", K(ret), K(trans_table_pkey)); } } return ret; } int ObMigrateTaskGeneratorTask::deal_with_rebuild_partition() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; common::ObAddr leader; ctx_->need_offline_ = true; ObRole role; LOG_INFO("start deal_with_rebuild_partition"); if (NULL != ctx_) { FLOG_INFO("rebuild is start", "pkey", ctx_->replica_op_arg_.key_); } if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (REBUILD_REPLICA_OP != ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; LOG_ERROR("replica op is not rebuild op", K(ret), "arg", ctx_->replica_op_arg_); } else if (OB_ISNULL(partition = ctx_->get_partition())) { ret = OB_ERR_SYS; LOG_ERROR("cannot use old partition", K(ret)); } else if (OB_FAIL(partition->get_leader(leader))) { LOG_WARN("failed to get leader", K(ret), "arg", ctx_->replica_op_arg_); } else if (OB_FAIL(partition->get_role(role))) { LOG_WARN("failed to get real role", K(ret), "arg", ctx_->replica_op_arg_); } else if (leader.is_valid() && leader == MYADDR // support rebuild in leader reconfirm && partition->get_log_service()->is_leader_active()) { // TODO(wait for yanmu) //&& is_strong_leader(role)) { if (OB_FAIL(MIGRATOR.get_partition_service()->turn_off_rebuild_flag(ctx_->replica_op_arg_))) { LOG_WARN("Failed to report_rebuild_replica off", K(ret), "arg", ctx_->replica_op_arg_); } else { ret = OB_NO_NEED_REBUILD; LOG_WARN("leader can not as rebuild dst", K(ret), K(leader), "myaddr", MYADDR, "arg", ctx_->replica_op_arg_); } } else { ObBaseStorageInfo& remote_clog_info = ctx_->pg_meta_.storage_info_.get_clog_info(); const ObPartitionSplitInfo& split_info = ctx_->pg_meta_.split_info_; const int64_t local_max_confirm_log_id = partition->get_log_service()->get_next_index_log_id() - 1; const int64_t src_last_replay_log_id = remote_clog_info.get_last_replay_log_id(); const bool is_replica_with_remote_memstore = partition->get_pg_storage().is_replica_with_remote_memstore(); bool skip_log_id_check = false; #ifdef ERRSIM skip_log_id_check = ObServerConfig::get_instance()._ob_enable_rebuild_on_purpose; #endif if (OB_FAIL(ret)) { } else if (src_last_replay_log_id <= local_max_confirm_log_id && !is_replica_with_remote_memstore && !skip_log_id_check) { ret = OB_NO_NEED_REBUILD; LOG_WARN("local log id is new enough , no need to rebuild, will turn off rebuild flag", K(ret), K(src_last_replay_log_id), K(local_max_confirm_log_id), K(remote_clog_info)); if (OB_SUCCESS != (tmp_ret = partition_service_->turn_off_rebuild_flag(ctx_->replica_op_arg_))) { LOG_WARN("Failed to report_rebuild_replica off", K(tmp_ret), "arg", ctx_->replica_op_arg_); } } else if (OB_FAIL(update_multi_version_start())) { LOG_WARN("failed to update multi version start", K(ret), K(*ctx_)); } else if (split_info.is_valid() && OB_FAIL(partition->save_split_info(split_info))) { LOG_WARN("failed to save split info", K(ret), K(split_info), K(*ctx_)); } else { ctx_->during_migrating_ = true; if (!is_replica_with_remote_memstore) { ctx_->need_online_for_rebuild_ = true; if (OB_FAIL(partition->pause())) { LOG_WARN("failed to pause partition", K(ret), "pkey", partition->get_partition_key()); } else if (is_standby_leader(role)) { uint64_t unused_log_id = OB_INVALID_ID; ObBaseStorageInfo clog_info; if (OB_FAIL(partition->get_log_service()->get_base_storage_info(clog_info, unused_log_id))) { LOG_WARN("failed to get base storage info", K(ret)); } else if (OB_FAIL(remote_clog_info.standby_force_update_member_list(clog_info.get_membership_log_id(), clog_info.get_membership_timestamp(), clog_info.get_replica_num(), clog_info.get_curr_member_list(), clog_info.get_ms_proposal_id()))) { LOG_WARN("failed to update restore standby member list", K(ret), K(clog_info), K(remote_clog_info)); } } if (OB_SUCC(ret)) { LOG_INFO("succeed to offline_rebuild_partition", K(leader), "arg", ctx_->replica_op_arg_, K(src_last_replay_log_id), K(local_max_confirm_log_id)); } } else { LOG_INFO("succeed to deal_with_replica using remote memestore replica", K(leader), "arg", ctx_->replica_op_arg_, K(src_last_replay_log_id), K(local_max_confirm_log_id)); } } } return ret; } int ObMigrateTaskGeneratorTask::deal_with_standby_restore_partition() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; LOG_INFO("start deal_with_standby restore_partition"); if (NULL != ctx_) { FLOG_INFO("standby restore is start", "pkey", ctx_->replica_op_arg_.key_); } if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (RESTORE_STANDBY_OP != ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; LOG_ERROR("replica op is not standby restore op", K(ret), "arg", ctx_->replica_op_arg_); } else if (OB_ISNULL(partition = ctx_->get_partition())) { ret = OB_ERR_SYS; LOG_ERROR("cannot use old partition", K(ret)); } else { ObBaseStorageInfo& remote_clog_info = ctx_->pg_meta_.storage_info_.get_clog_info(); uint64_t unused_log_id = OB_INVALID_ID; ObBaseStorageInfo clog_info; if (OB_FAIL(partition->get_log_service()->get_base_storage_info(clog_info, unused_log_id))) { LOG_WARN("failed to get base storage info", K(ret)); } else if (OB_FAIL(update_multi_version_start())) { LOG_WARN("failed to update multi version start", K(ret), K(*ctx_)); } else if (OB_FAIL(partition->get_log_service()->set_offline())) { LOG_WARN("failed to offline log service", K(ret)); } else if (OB_FAIL(remote_clog_info.standby_force_update_member_list(clog_info.get_membership_log_id(), clog_info.get_membership_timestamp(), clog_info.get_replica_num(), clog_info.get_curr_member_list(), clog_info.get_ms_proposal_id()))) { LOG_WARN("failed to update restore standby member list", K(ret), K(clog_info), K(remote_clog_info)); } } return ret; } int ObMigrateTaskGeneratorTask::generate_pg_validate_tasks(ObIArray& last_task_array) { int ret = OB_SUCCESS; ObITask* last_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (VALIDATE_BACKUP_OP != ctx_->replica_op_arg_.type_) { ret = OB_INVALID_ARGUMENT; LOG_WARN("unexpected op type, ", K(ret), K(ctx_->replica_op_arg_.type_)); } else if (OB_FAIL(generate_validate_tasks(last_task))) { LOG_WARN("failed to generate validate tasks", K(ret)); } else if (OB_FAIL(last_task_array.push_back(last_task))) { LOG_WARN("failed to push last task into last task array", K(ret)); ; } return ret; } int ObMigrateTaskGeneratorTask::generate_validate_tasks(ObITask*& last_task) { int ret = OB_SUCCESS; ObFakeTask* wait_validate_finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(generate_wait_migrate_finish_task(wait_validate_finish_task))) { LOG_WARN("failed to generate_wait_migrate_finish_task", K(ret)); } else if (OB_ISNULL(wait_validate_finish_task)) { ret = OB_ERR_SYS; LOG_ERROR("wait_validate_finsh_task must not be NULL", K(ret)); } else if (OB_FAIL(generate_validate_tasks(*wait_validate_finish_task))) { LOG_WARN("failed to generate_validate_tasks", K(ret)); } if (OB_SUCC(ret)) { last_task = wait_validate_finish_task; } return ret; } int ObMigrateTaskGeneratorTask::generate_validate_tasks(ObFakeTask& wait_validate_finish_task) { int ret = OB_SUCCESS; ObITask* parent_task = this; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(generate_validate_backup_tasks(parent_task))) { LOG_WARN("failed to generate validate backup tasks", K(ret)); } else if (OB_FAIL(parent_task->add_child(wait_validate_finish_task))) { LOG_WARN("failed to add wait_validate_finish_task", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_validate_backup_tasks(share::ObITask*& parent_task) { int ret = OB_SUCCESS; ObFakeTask* wait_finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (OB_ISNULL(parent_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid argument", K(ret), KP(parent_task)); } else if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_validate_backup_task(parent_task, wait_finish_task))) { LOG_WARN("failed to generate validate backup task", K(ret)); } else if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_DEBUG("succeed to generate_validate_backup_tasks"); } return ret; } int ObMigrateTaskGeneratorTask::generate_validate_backup_task(share::ObITask* parent_task, share::ObITask* child_task) { int ret = OB_SUCCESS; const int64_t task_idx = 0; const int64_t clog_file_id = 1; ObValidatePrepareTask* prepare_task = NULL; ObValidateClogDataTask* clog_task = NULL; ObValidateBaseDataTask* base_task = NULL; ObValidateFinishTask* finish_task = NULL; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(parent_task), KP(child_task)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc finish task", K(ret)); } else if (OB_FAIL(build_validate_backup_ctx(finish_task->get_validate_ctx()))) { LOG_WARN("failed to build validate backup ctx", K(ret)); } else if (OB_FAIL(finish_task->init(*ctx_))) { LOG_WARN("failed to init finish task", K(ret)); } else if (OB_FAIL(finish_task->add_child(*child_task))) { LOG_WARN("failed to add child", K(ret)); } if (OB_SUCC(ret)) { if (finish_task->get_validate_ctx().get_sub_task_count() > 0) { if (OB_FAIL(dag_->alloc_task(prepare_task))) { LOG_WARN("failed to alloc prepare task", K(ret)); } else if (OB_FAIL(dag_->alloc_task(clog_task))) { LOG_WARN("failed to alloc clog task", K(ret)); } else if (OB_FAIL(dag_->alloc_task(base_task))) { LOG_WARN("failed to alloc base task", K(ret)); } else if (OB_FAIL(prepare_task->init(*ctx_, finish_task->get_validate_ctx()))) { LOG_WARN("failed to init prepare task", K(ret)); } else if (OB_FAIL(clog_task->init(clog_file_id, *ctx_, finish_task->get_validate_ctx()))) { LOG_WARN("failed to init clog task", K(ret)); } else if (OB_FAIL(base_task->init(task_idx, *ctx_, finish_task->get_validate_ctx()))) { LOG_WARN("failed to init base task", K(ret)); } else if (OB_FAIL(parent_task->add_child(*prepare_task))) { LOG_WARN("failed to add child prepare task", K(ret)); } else if (OB_FAIL(prepare_task->add_child(*clog_task))) { LOG_WARN("failed to add child clog task", K(ret)); } else if (OB_FAIL(clog_task->add_child(*base_task))) { LOG_WARN("failed to add child base task", K(ret)); } else if (OB_FAIL(base_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish task", K(ret)); } else if (OB_FAIL(dag_->add_task(*prepare_task))) { LOG_WARN("failed to add prepare task to dag", K(ret)); } else if (OB_FAIL(dag_->add_task(*clog_task))) { LOG_WARN("failed to add clog task to dag", K(ret)); } else if (OB_FAIL(dag_->add_task(*base_task))) { LOG_WARN("failed to add base task to dag", K(ret)); } } else { if (OB_FAIL(parent_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish_task for parent"); } } } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed tp add finish task to dag", K(ret)); } } return ret; } int ObMigrateTaskGeneratorTask::build_validate_backup_ctx(ObValidateBackupPGCtx& pg_ctx) { int ret = OB_SUCCESS; pg_ctx.reset(); ObBackupBaseDataPathInfo path_info; ObArray macro_index_list; const ObPartitionKey& pg_key = ctx_->replica_op_arg_.validate_arg_.pg_key_; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(ctx_->replica_op_arg_.validate_arg_.get_backup_base_data_info(path_info))) { STORAGE_LOG(WARN, "failed to get backup base data info", K(ret)); } else if (!ctx_->macro_index_store_.is_inited() && OB_FAIL(ctx_->macro_index_store_.init(path_info))) { STORAGE_LOG(WARN, "failed to init macro index store", K(ret), K(path_info)); } else if (OB_FAIL(pg_ctx.init(*ctx_, *bandwidth_throttle_))) { LOG_WARN("ctx init failed", K(ret)); } else if (OB_FAIL(fetch_pg_macro_index_list(pg_key, pg_ctx, macro_index_list))) { LOG_WARN("failed to fetch pg macro block list", K(ret)); } else if (OB_FAIL(build_validate_sub_task(macro_index_list, pg_ctx))) { LOG_WARN("failed to build validate sub task", K(ret)); } else { LOG_INFO("succeed to build validate backup pg ctx", K(pg_ctx)); } return ret; } int ObMigrateTaskGeneratorTask::fetch_pg_macro_index_list( const ObPartitionKey& pg_key, ObValidateBackupPGCtx& pg_ctx, ObIArray& macro_index_list) { int ret = OB_SUCCESS; share::ObBackupMetaIndex meta_index; ObArray* index_list = NULL; ObBackupMetaIndexStore* meta_index_store = NULL; ObBackupMacroIndexStore& macro_index_store = ctx_->macro_index_store_; ObPartGroupMigrationTask* group_task = NULL; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (FALSE_IT(group_task = reinterpret_cast(ctx_->group_task_))) { } else if (FALSE_IT(meta_index_store = &group_task->get_meta_index_store())) { } else if (OB_ISNULL(meta_index_store)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("meta index store should not be NULL", K(ret), KP(meta_index_store)); } else if (OB_FAIL(meta_index_store->get_meta_index(pg_key, ObBackupMetaType::PARTITION_GROUP_META, meta_index))) { if (OB_HASH_NOT_EXIST == ret) { pg_ctx.only_in_clog_ = true; ret = OB_SUCCESS; LOG_INFO("pkey only exists in clog", K(ret), K(pg_key)); } } else if (OB_UNLIKELY(!macro_index_store.is_inited())) { ret = OB_NOT_INIT; LOG_WARN("macro index store do not init", K(ret)); } else if (OB_FAIL(macro_index_store.get_macro_block_index(pg_key, index_list))) { LOG_WARN("failed to get macro block index", K(ret), K(pg_key)); } else if (OB_ISNULL(index_list)) { LOG_INFO("the pg is empty, skip", K(ret), K(pg_key)); } else if (OB_FAIL(macro_index_list.assign(*index_list))) { LOG_WARN("failed to assign to macro_index_list", K(ret)); } else if (FALSE_IT(pg_ctx.total_macro_block_count_ = macro_index_list.count())) { // do nothing } return ret; } int ObMigrateTaskGeneratorTask::build_validate_sub_task( const ObIArray& macro_index_list, ObValidateBackupPGCtx& ctx) { int ret = OB_SUCCESS; void* buf = NULL; int64_t max_macro_block_count_per_task = 1024 /*MAX_MACRO_BLOCK_COUNT_PER_TASK*/; LOG_INFO("start to build validate sub task"); if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (macro_index_list.empty()) { ctx.sub_task_cnt_ = 0; LOG_INFO("macro index list is empty, skip validating", K(ret)); } else { ctx.sub_task_cnt_ = macro_index_list.count() / max_macro_block_count_per_task; if (macro_index_list.count() % max_macro_block_count_per_task > 0) { ++ctx.sub_task_cnt_; } if (OB_ISNULL(buf = ctx.allocator_.alloc(sizeof(ObValidateBackupPGCtx::SubTask) * ctx.sub_task_cnt_))) { ret = OB_ALLOCATE_MEMORY_FAILED; LOG_ERROR("failed to allocate memory", K(ret)); } else { ctx.sub_tasks_ = new (buf) ObValidateBackupPGCtx::SubTask[ctx.sub_task_cnt_]; buf = NULL; } for (int64_t i = 0; OB_SUCC(ret) && i < ctx.sub_task_cnt_; ++i) { ObValidateBackupPGCtx::SubTask& sub_task = ctx.sub_tasks_[i]; sub_task.macro_block_count_ = std::min(max_macro_block_count_per_task, macro_index_list.count() - i * max_macro_block_count_per_task); sub_task.pkey_ = ctx.pg_key_; if (OB_ISNULL(buf = ctx.allocator_.alloc(sizeof(ObBackupMacroIndex) * sub_task.macro_block_count_))) { ret = OB_ALLOCATE_MEMORY_FAILED; LOG_WARN("failed to alloc memory", K(ret)); } else { sub_task.macro_block_infos_ = new (buf) ObBackupMacroIndex[sub_task.macro_block_count_]; buf = NULL; } for (int64_t j = 0; OB_SUCC(ret) && j < sub_task.macro_block_count_; ++j) { const int64_t macro_idx = i * max_macro_block_count_per_task + j; if (macro_idx >= macro_index_list.count()) { break; } const ObBackupMacroIndex& macro_index = macro_index_list.at(macro_idx); ObBackupMacroIndex& index = ctx.sub_tasks_[i].macro_block_infos_[j]; index = macro_index; } } } return ret; } int ObMigrateTaskGeneratorTask::generate_pg_backup_backupset_tasks(common::ObIArray& last_task_array) { int ret = OB_SUCCESS; ObITask* last_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_ERROR("not inited", KR(ret)); } else if (BACKUP_BACKUPSET_OP != ctx_->replica_op_arg_.type_) { ret = OB_INVALID_ARGUMENT; LOG_WARN("unexcepted op type, ", KR(ret), K(ctx_->replica_op_arg_)); } else if (OB_FAIL(generate_backup_backupset_tasks(last_task))) { LOG_WARN("fail to generate backup backupset tasks", KR(ret)); } else if (OB_FAIL(last_task_array.push_back(last_task))) { LOG_WARN("fail to push last task into last task array", KR(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_backupset_tasks(share::ObITask*& last_task) { int ret = OB_SUCCESS; ObFakeTask* wait_migrate_finish_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_ERROR("not inited", KR(ret)); } else if (OB_FAIL(generate_wait_migrate_finish_task(wait_migrate_finish_task))) { LOG_WARN("failed to generate_wait_migrate_finish_task", KR(ret)); } else if (OB_ISNULL(wait_migrate_finish_task)) { ret = OB_ERR_SYS; LOG_ERROR("wait_migrate_finish_task must not null", KR(ret)); } else if (OB_FAIL(generate_backup_backupset_tasks(*wait_migrate_finish_task))) { LOG_WARN("failed to generate_backup_backupset_tasks", K(ret)); } if (OB_SUCC(ret)) { last_task = wait_migrate_finish_task; } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_backupset_tasks(share::ObFakeTask& wait_finish_task) { int ret = OB_SUCCESS; ObITask* parent_task = this; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_ERROR("not inited", KR(ret)); } else if (OB_FAIL(generate_backup_backupset_pg_tasks(parent_task))) { LOG_WARN("failed to generate backup major tasks", K(ret)); } else if (OB_FAIL(parent_task->add_child(wait_finish_task))) { LOG_WARN("failed to add wait_migrate_finish_task", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_backupset_pg_tasks(share::ObITask*& parent_task) { int ret = OB_SUCCESS; ObFakeTask* wait_finish_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", KR(ret)); } else if (OB_ISNULL(parent_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid argument", K(ret), KP(parent_task)); } else if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_backup_backupset_pg_tasks(parent_task, wait_finish_task))) { LOG_WARN("failed to generate backup backupset pg task", KR(ret)); } else if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_INFO("succeed to generate_backup_backupset_pg_tasks"); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_backupset_pg_tasks( share::ObITask*& parent_task, share::ObITask* child_task) { int ret = OB_SUCCESS; int64_t cur_idx = 0; ObBackupBackupsetFileTask* copy_task = NULL; ObBackupBackupsetFinishTask* finish_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("not inited", KR(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", KR(ret), KP(parent_task), KP(child_task)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc finish task", KR(ret)); } else if (OB_FAIL(build_backup_backupset_ctx_v2(finish_task->get_backup_backupset_file_ctx()))) { LOG_WARN("failed to build backup backupset ctx", KR(ret)); } else if (OB_FAIL(finish_task->init(*ctx_))) { LOG_WARN("failed to init finish task", KR(ret)); } else if (OB_FAIL(finish_task->add_child(*child_task))) { LOG_WARN("failed to add child", KR(ret)); } if (OB_SUCC(ret)) { // TODO() FIX total file count const int64_t total_file_count = 0; if (total_file_count > 0) { if (OB_FAIL(dag_->alloc_task(copy_task))) { LOG_WARN("failed to alloc copy task", KR(ret)); } else if (OB_FAIL(copy_task->init(cur_idx, *ctx_, finish_task->get_backup_backupset_file_ctx()))) { LOG_WARN("failed to init copy task", KR(ret)); } else if (OB_FAIL(parent_task->add_child(*copy_task))) { LOG_WARN("failed to add child copy task", KR(ret)); } else if (OB_FAIL(copy_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish task", KR(ret)); } else if (OB_FAIL(dag_->add_task(*copy_task))) { LOG_WARN("failed to add copy task to dag", KR(ret)); } } else { if (OB_FAIL(parent_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish_task for parent", KR(ret)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish taksk to dag", KR(ret)); } } return ret; } int ObMigrateTaskGeneratorTask::build_backup_backupset_ctx_v2(ObBackupBackupsetPGFileCtx& pg_ctx) { int ret = OB_SUCCESS; ObStorageUtil util(true /*need retry*/); const share::ObBackupBackupsetArg& bb_arg = ctx_->replica_op_arg_.backup_backupset_arg_; const common::ObPGKey& pg_key = bb_arg.pg_key_; const uint64_t table_id = pg_key.get_table_id(); const int64_t partition_id = pg_key.get_partition_id(); const int64_t compatible = bb_arg.compatible_; ObBackupBaseDataPathInfo path_info; ObBackupPath major_pg_path, minor_pg_path; int64_t minor_task_id = 0; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("not init", KR(ret)); } else if (compatible <= OB_BACKUP_COMPATIBLE_VERSION_V1 || compatible >= OB_BACKUP_COMPATIBLE_VERSION_MAX) { ret = OB_ERR_UNEXPECTED; LOG_WARN("compatible version is not valid", KR(ret), K(compatible)); } else if (OB_FAIL(pg_ctx.open(bb_arg, pg_key))) { LOG_WARN("failed to open pg ctx", KR(ret), K(bb_arg), K(pg_key)); } else if (OB_FAIL(bb_arg.get_src_backup_base_data_info(path_info))) { LOG_WARN("failed to get src backup base data info", KR(ret), K(bb_arg)); } else if (OB_FAIL( ObBackupPathUtil::get_tenant_pg_major_data_path(path_info, table_id, partition_id, major_pg_path))) { LOG_WARN("failed to get tenant pg data path", KR(ret), K(path_info), K(pg_key)); // TODO()Backup backup fix it //} else if (OB_FAIL(util.list_files(major_pg_path.get_obstr(), // path_info.dest_.get_storage_info(), // pg_ctx.allocator_, // pg_ctx.major_files_))) { // LOG_WARN("failed to list files", KR(ret), K(major_pg_path), K(path_info)); // } } else { if (compatible >= ObBackupCompatibleVersion::OB_BACKUP_COMPATIBLE_VERSION_V3) { if (OB_FAIL(get_backup_backup_minor_task_id(path_info, pg_key, minor_task_id))) { LOG_WARN("failed to get backup backup minor task id", KR(ret), K(path_info), K(pg_key)); } else if (OB_FAIL(OB_FAIL(ObBackupPathUtil::get_tenant_pg_minor_data_path( path_info, table_id, partition_id, minor_task_id, minor_pg_path)))) { LOG_WARN("failed to get tenant pg minor data path", KR(ret), K(path_info), K(pg_key)); // TODO()Backup backup Fix it //} else if (OB_FAIL(util.list_files(minor_pg_path.get_obstr(), // path_info.dest_.get_storage_info(), // pg_ctx.allocator_, // pg_ctx.minor_files_))) { // LOG_WARN("failed to list files", KR(ret), K(minor_pg_path), K(path_info)); //} else { // pg_ctx.minor_task_id_ = minor_task_id; // } } } } return ret; } int ObMigrateTaskGeneratorTask::get_backup_backup_minor_task_id( const share::ObBackupBaseDataPathInfo& path_info, const common::ObPGKey& pg_key, int64_t& task_id) { int ret = OB_SUCCESS; task_id = 0; ObArenaAllocator allocator; ObStorageUtil util(true /*need retry*/); const uint64_t table_id = pg_key.get_table_id(); const int64_t partition_id = pg_key.get_partition_id(); const share::ObBackupBackupsetArg& bb_arg = ctx_->replica_op_arg_.backup_backupset_arg_; ObBackupPath pg_path; ObArray dir_list; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("not init", KR(ret)); } else if (!pg_key.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("get invalid args", KR(ret), K(pg_key)); } else if (OB_FAIL(ObBackupPathUtil::get_tenant_pg_minor_dir_path(path_info, table_id, partition_id, pg_path))) { LOG_WARN("failed to get tenant pg minor data path", KR(ret), K(path_info), K(pg_key)); } else if (OB_FAIL(util.list_directories(pg_path.get_obstr(), path_info.dest_.get_storage_info(), allocator, dir_list))) { LOG_WARN("failed to list files", KR(ret), K(pg_path), K(path_info)); } else if (dir_list.empty()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("minor directory is empty", KR(ret), K(path_info), K(pg_key)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < dir_list.count(); ++i) { int64_t tmp_task_id = 0; const ObString &dir= dir_list.at(i); const char *str = dir.ptr(); for (int64_t j = 0; OB_SUCC(ret) && j < dir.length(); ++j) { const char end_flag = '/'; if (end_flag == str[j]) { break; } else if (!isdigit(str[j])) { ret = OB_ERR_UNEXPECTED; LOG_WARN("not digit number", KR(ret), K(j), K(str[j])); } else { if (tmp_task_id > INT64_MAX / 10 || (tmp_task_id == INT64_MAX / 10 && (str[j] - '0') > INT64_MAX % 10)) { ret = OB_DECIMAL_OVERFLOW_WARN; LOG_WARN("task id is not valid", KR(ret), K(tmp_task_id)); } else { tmp_task_id = tmp_task_id * 10 + (str[j] - '0'); } } } if (OB_SUCC(ret) && task_id < tmp_task_id) { task_id = tmp_task_id; } } } return ret; } int ObMigrateTaskGeneratorTask::generate_pg_backup_archive_log_tasks(common::ObIArray& last_task_array) { int ret = OB_SUCCESS; ObITask* last_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_ERROR("not inited", KR(ret)); } else if (BACKUP_ARCHIVELOG_OP != ctx_->replica_op_arg_.type_) { ret = OB_INVALID_ARGUMENT; LOG_WARN("unexcepted op type, ", KR(ret), K(ctx_->replica_op_arg_)); } else if (OB_FAIL(generate_backup_archive_log_tasks(last_task))) { LOG_WARN("fail to generate backup backupset tasks", KR(ret)); } else if (OB_FAIL(last_task_array.push_back(last_task))) { LOG_WARN("fail to push last task into last task array", KR(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_archive_log_tasks(share::ObITask*& last_task) { int ret = OB_SUCCESS; ObFakeTask* wait_migrate_finish_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_ERROR("not inited", KR(ret)); } else if (OB_FAIL(generate_wait_migrate_finish_task(wait_migrate_finish_task))) { LOG_WARN("failed to generate_wait_migrate_finish_task", KR(ret)); } else if (OB_ISNULL(wait_migrate_finish_task)) { ret = OB_ERR_SYS; LOG_ERROR("wait_migrate_finish_task must not null", KR(ret)); } else if (OB_FAIL(generate_backup_archive_log_tasks(*wait_migrate_finish_task))) { LOG_WARN("failed to generate_backup_backupset_tasks", K(ret)); } if (OB_SUCC(ret)) { last_task = wait_migrate_finish_task; } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_archive_log_tasks(share::ObFakeTask& wait_finish_task) { int ret = OB_SUCCESS; ObITask* parent_task = this; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_ERROR("not inited", KR(ret)); } else if (OB_FAIL(generate_backup_archive_log_pg_tasks(parent_task))) { LOG_WARN("failed to generate backup major tasks", K(ret)); } else if (OB_FAIL(parent_task->add_child(wait_finish_task))) { LOG_WARN("failed to add wait_migrate_finish_task", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_archive_log_pg_tasks(share::ObITask*& parent_task) { int ret = OB_SUCCESS; ObFakeTask* wait_finish_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", KR(ret)); } else if (OB_ISNULL(parent_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid argument", K(ret), KP(parent_task)); } else if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_backup_archive_log_pg_tasks(parent_task, wait_finish_task))) { LOG_WARN("failed to generate backup backupset pg task", KR(ret)); } else if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_INFO("succeed to generate_backup_backupset_pg_tasks"); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_archive_log_pg_tasks( share::ObITask*& parent_task, share::ObITask* child_task) { int ret = OB_SUCCESS; ObBackupArchiveLogPGTask* copy_task = NULL; ObBackupArchiveLogFinishTask* finish_task = NULL; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("not inited", KR(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", KR(ret), KP(parent_task), KP(child_task)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc finish task", KR(ret)); } else if (OB_FAIL(build_backup_archive_log_ctx(finish_task->get_backup_archivelog_ctx()))) { LOG_WARN("failed to build backup archive log ctx", KR(ret)); } else if (OB_FAIL(finish_task->init(*ctx_))) { LOG_WARN("failed to init finish task", KR(ret)); } else if (OB_FAIL(finish_task->add_child(*child_task))) { LOG_WARN("failed to add child", KR(ret)); } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->alloc_task(copy_task))) { LOG_WARN("failed to alloc copy task", KR(ret)); } else if (OB_FAIL(copy_task->init(*ctx_, finish_task->get_backup_archivelog_ctx()))) { LOG_WARN("failed to init copy task", KR(ret)); } else if (OB_FAIL(parent_task->add_child(*copy_task))) { LOG_WARN("failed to add child copy task", KR(ret)); } else if (OB_FAIL(copy_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish task", KR(ret)); } else if (OB_FAIL(dag_->add_task(*copy_task))) { LOG_WARN("failed to add copy task to dag", KR(ret)); } } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish taksk to dag", KR(ret)); } } return ret; } int ObMigrateTaskGeneratorTask::build_backup_archive_log_ctx(ObBackupArchiveLogPGCtx& ctx) { int ret = OB_SUCCESS; const common::ObPGKey& pg_key = ctx_->replica_op_arg_.backup_archive_log_arg_.pg_key_; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("not init", KR(ret)); } else if (OB_FAIL(ctx.open(*ctx_, pg_key, *bandwidth_throttle_))) { LOG_WARN("failed to init ctx", KR(ret), K(pg_key)); } else { LOG_INFO("build backup archive log ctx success", KR(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_pg_backup_tasks(ObIArray& last_task_array) { int ret = OB_SUCCESS; ObITask* last_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (BACKUP_REPLICA_OP != ctx_->replica_op_arg_.type_) { ret = OB_INVALID_ARGUMENT; LOG_WARN("unexcepted op type, ", K(ret), K(ctx_->replica_op_arg_)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(ctx_->replica_op_arg_.src_.get_replica_type())) { ret = OB_REPLICA_CANNOT_BACKUP; LOG_WARN("replica has no sstore.", K(ret), K(ctx_->replica_op_arg_)); } else if (ctx_->partition_guard_.get_partition_group()->is_removed()) { ret = OB_PG_IS_REMOVED; LOG_WARN("partition is already removed, can not backup", K(ret), K(ctx_->replica_op_arg_.key_)); } else if (OB_FAIL(generate_backup_tasks(last_task))) { LOG_WARN("fail to generate backup tasks", K(ret)); } else if (OB_FAIL(last_task_array.push_back(last_task))) { LOG_WARN("fail to push last task into last task array", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_tasks(ObITask*& last_task) { int ret = OB_SUCCESS; ObFakeTask* wait_migrate_finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(generate_wait_migrate_finish_task(wait_migrate_finish_task))) { LOG_WARN("failed to generate_wait_migrate_finish_task", K(ret)); } else if (OB_ISNULL(wait_migrate_finish_task)) { ret = OB_ERR_SYS; LOG_ERROR("wait_rebuild_finish_task must not null", K(ret)); } else if (OB_FAIL(generate_backup_tasks(*wait_migrate_finish_task))) { LOG_WARN("failed to generate_backup_tasks", K(ret)); } if (OB_SUCC(ret)) { last_task = wait_migrate_finish_task; } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_tasks(ObFakeTask& wait_migrate_finish_task) { int ret = OB_SUCCESS; ObITask* parent_task = this; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(generate_backup_major_tasks(parent_task))) { LOG_WARN("failed to generate backup major tasks", K(ret)); } else if (OB_FAIL(parent_task->add_child(wait_migrate_finish_task))) { LOG_WARN("failed to add wait_migrate_finish_task", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_major_tasks(share::ObITask*& parent_task) { int ret = OB_SUCCESS; ObFakeTask* wait_finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (OB_ISNULL(parent_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid argument", K(ret), KP(parent_task)); } else if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_backup_major_copy_task(parent_task, wait_finish_task))) { LOG_WARN("failed to generate major sstable copy task", K(ret)); } else if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_DEBUG("succeed to generate_major_sstable_copy_task"); } return ret; } int ObMigrateTaskGeneratorTask::generate_backup_major_copy_task(ObITask* parent_task, ObITask* child_task) { int ret = OB_SUCCESS; ObBackupCopyPhysicalTask* copy_task = NULL; ObBackupFinishTask* finish_task = NULL; const int64_t task_idx = 0; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(parent_task), KP(child_task)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc finish task", K(ret)); } else if (OB_FAIL(build_backup_physical_ctx(ctx_->physical_backup_ctx_))) { LOG_WARN("failed to build physical sstable ctx", K(ret)); } else if (OB_FAIL(finish_task->init(*ctx_))) { LOG_WARN("failed to init finish task", K(ret)); } else if (OB_FAIL(finish_task->add_child(*child_task))) { LOG_WARN("failed to add child", K(ret)); } if (OB_SUCC(ret)) { if (ctx_->physical_backup_ctx_.get_task_count() > 0) { // parent->copy->finish->child if (OB_FAIL(dag_->alloc_task(copy_task))) { LOG_WARN("failed to alloc copy task", K(ret)); } else if (OB_FAIL(copy_task->init(task_idx, *ctx_))) { LOG_WARN("failed to init copy task", K(ret)); } else if (OB_FAIL(parent_task->add_child(*copy_task))) { LOG_WARN("failed to add child copy task", K(ret)); } else if (OB_FAIL(copy_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish task", K(ret)); } else if (OB_FAIL(dag_->add_task(*copy_task))) { LOG_WARN("failed to add copy task to dag", K(ret)); } } else { // parent->finish->child if (OB_FAIL(parent_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish_task for parent", K(ret)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish task to dag", K(ret)); } } return ret; } int ObMigrateTaskGeneratorTask::build_backup_physical_ctx(ObBackupPhysicalPGCtx& physical_backup_ctx) { int ret = OB_SUCCESS; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(physical_backup_ctx.open())) { LOG_WARN("failed to open physical backup ctx", K(ret), K(ctx_->pg_meta_)); } else if (OB_FAIL(fetch_backup_sstables(physical_backup_ctx.table_keys_))) { LOG_WARN("failed to build sub task", K(ret)); } else if (0 == physical_backup_ctx.table_keys_.count()) { LOG_INFO("pg don't have major sstable, ", K(ret), K(ctx_->pg_meta_)); } else if (OB_FAIL(build_backup_sub_task(physical_backup_ctx))) { LOG_WARN("failed to build sub task", K(ret)); } else { ATOMIC_AAF(&ctx_->data_statics_.total_macro_block_, physical_backup_ctx.macro_block_count_); LOG_INFO("succeed to build physical sstable ctx", K(physical_backup_ctx)); } return ret; } int ObMigrateTaskGeneratorTask::fetch_backup_sstables(ObIArray& table_keys) { int ret = OB_SUCCESS; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else { // partition for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { ObPartitionMigrateCtx& part_migrate_ctx = ctx_->part_ctx_array_.at(i); if (OB_LIKELY(!part_migrate_ctx.is_valid())) { ret = OB_INVALID_ARGUMENT; LOG_WARN("part_migrate_ctx is invalid", K(ret), K(part_migrate_ctx)); } else { // tables const ObArray& table_infos = part_migrate_ctx.copy_info_.table_infos_; for (int64_t i = 0; OB_SUCC(ret) && i < table_infos.count(); ++i) { const ObMigrateTableInfo& table_info = table_infos.at(i); // major sstables for (int64_t sstable_idx = 0; OB_SUCC(ret) && sstable_idx < table_info.major_sstables_.count(); ++sstable_idx) { const ObITable::TableKey& major_table_key = table_info.major_sstables_.at(sstable_idx).src_table_key_; if (OB_UNLIKELY(!major_table_key.is_valid())) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid major table key", K(ret), K(major_table_key)); } else if (OB_UNLIKELY(!ObITable::is_major_sstable(major_table_key.table_type_))) { ret = OB_ERR_SYS; LOG_ERROR("table type is not major sstable", K(ret), K(major_table_key), K(table_info)); } else { table_keys.push_back(major_table_key); } } } } } } return ret; } int ObMigrateTaskGeneratorTask::build_backup_sub_task(ObBackupPhysicalPGCtx& ctx) { int ret = OB_SUCCESS; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else { for (int i = 0; OB_SUCC(ret) && i < ctx.table_keys_.count(); ++i) { ObTableHandle tmp_handle; ObSSTable* sstable = NULL; int64_t sstable_macro_count = 0; ObITable::TableKey& major_table_key = ctx.table_keys_.at(i); if (OB_FAIL(ObPartitionService::get_instance().acquire_sstable(major_table_key, tmp_handle))) { STORAGE_LOG(WARN, "failed to get table", K(major_table_key), K(ret)); } else if (OB_FAIL(tmp_handle.get_sstable(sstable))) { STORAGE_LOG(WARN, "failed to get table", K(major_table_key), K(ret)); } else if (OB_ISNULL(sstable)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "sstable should not be null", K(ret)); } else if (0 == (sstable_macro_count = sstable->get_meta().get_total_macro_block_count())) { // empty sstable } else { ObBackupMacroBlockInfo block_info; block_info.table_key_ = major_table_key; block_info.start_index_ = 0; block_info.cur_block_count_ = sstable_macro_count; block_info.total_block_count_ = sstable_macro_count; if (OB_FAIL(ctx.add_backup_macro_block_info(block_info))) { STORAGE_LOG(WARN, "add backup block info failed", K(ret), K(block_info)); } } } } return ret; } int ObMigrateTaskGeneratorTask::generate_pg_migrate_tasks(ObIArray& last_task_array) { int ret = OB_SUCCESS; ObITask* last_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(ctx_->replica_op_arg_.dst_.get_replica_type())) { ret = OB_ERR_SYS; LOG_WARN("no need to generate migrate tasks", K(ret), K(ctx_->replica_op_arg_)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { ObPartitionMigrateCtx& part_migrate_ctx = ctx_->part_ctx_array_.at(i); if (part_migrate_ctx.copy_info_.meta_.pkey_.is_trans_table()) { continue; } else if (OB_FAIL(generate_migrate_tasks(part_migrate_ctx, last_task))) { LOG_WARN("fail to generate migrate tasks", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(last_task_array.push_back(last_task))) { LOG_WARN("fail to push last task into last task array", K(ret)); } } } return ret; } int ObMigrateTaskGeneratorTask::generate_pg_rebuild_tasks(ObIArray& last_task_array) { int ret = OB_SUCCESS; ObITask* last_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(ctx_->replica_op_arg_.dst_.get_replica_type())) { ret = OB_ERR_SYS; LOG_WARN("no need to generate rebuild tasks", K(ret), K(ctx_->replica_op_arg_)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { ObPartitionMigrateCtx& part_migrate_ctx = ctx_->part_ctx_array_.at(i); if (part_migrate_ctx.copy_info_.meta_.pkey_.is_trans_table()) { continue; } else if (OB_FAIL(generate_rebuild_tasks(part_migrate_ctx, last_task))) { LOG_WARN("fail to generate rebuild tasks", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(last_task_array.push_back(last_task))) { LOG_WARN("fail to push last task into last task array", K(ret)); } } } return ret; } int ObMigrateTaskGeneratorTask::generate_migrate_tasks(ObPartitionMigrateCtx& part_migrate_ctx, ObITask*& last_task) { int ret = OB_SUCCESS; ObFakeTask* wait_migrate_finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (!part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("copy info is invalid", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(generate_wait_migrate_finish_task(wait_migrate_finish_task))) { LOG_WARN("failed to generate_wait_migrate_finish_task", K(ret)); } else if (OB_ISNULL(wait_migrate_finish_task)) { ret = OB_ERR_SYS; LOG_ERROR("wait_rebuild_finish_task must not null", K(ret)); } else { const ObMigratePartitionInfo& copy_info = part_migrate_ctx.copy_info_; const ObArray& table_infos = copy_info.table_infos_; for (int64_t i = 0; OB_SUCC(ret) && i < table_infos.count(); ++i) { const ObMigrateTableInfo& table_info = table_infos.at(i); if (OB_FAIL(generate_migrate_tasks( ctx_->migrate_src_info_, part_migrate_ctx, table_info, *wait_migrate_finish_task))) { LOG_WARN("failed to generate_migrate_tasks", K(ret), K(i), K(table_info), K(part_migrate_ctx)); } } } if (OB_SUCC(ret)) { last_task = wait_migrate_finish_task; } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_GEN_REBUILD_TASK) OB_SUCCESS; if (OB_FAIL(ret)) { STORAGE_LOG(ERROR, "fake EN_GEN_REBUILD_TASK", K(ret)); } } #endif return ret; } int ObMigrateTaskGeneratorTask::generate_rebuild_tasks(ObPartitionMigrateCtx& part_migrate_ctx, ObITask*& last_task) { int ret = OB_SUCCESS; ObFakeTask* wait_rebuild_finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (!part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("copy info is invalid", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(generate_wait_migrate_finish_task(wait_rebuild_finish_task))) { LOG_WARN("failed to generate_wait_rebuild_finish_task", K(ret)); } else if (OB_ISNULL(wait_rebuild_finish_task)) { ret = OB_ERR_SYS; LOG_ERROR("wait_rebuild_finish_task must not null", K(ret)); } else { const ObMigratePartitionInfo& copy_info = part_migrate_ctx.copy_info_; const ObArray& table_infos = copy_info.table_infos_; for (int64_t i = 0; OB_SUCC(ret) && i < table_infos.count(); ++i) { const ObMigrateTableInfo& table_info = table_infos.at(i); if (OB_FAIL(generate_rebuild_tasks( ctx_->migrate_src_info_, part_migrate_ctx, table_info, *wait_rebuild_finish_task))) { LOG_WARN("failed to generate_rebuild_tasks", K(ret), K(i), K(table_info), K(part_migrate_ctx)); } } } if (OB_SUCC(ret)) { last_task = wait_rebuild_finish_task; } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_GEN_REBUILD_TASK) OB_SUCCESS; if (OB_FAIL(ret)) { STORAGE_LOG(ERROR, "fake EN_GEN_REBUILD_TASK", K(ret)); } } #endif return ret; } int ObITableTaskGeneratorTask::generate_wait_migrate_finish_task(ObFakeTask*& wait_migrate_finish_task) { int ret = OB_SUCCESS; wait_migrate_finish_task = NULL; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret), KP(ctx_)); } else if (OB_FAIL(dag_->alloc_task(wait_migrate_finish_task))) { LOG_WARN("failed to alloc wait_rebuild_finish_task", K(ret)); } else if (OB_FAIL(add_child(*wait_migrate_finish_task))) { LOG_WARN("failed to add child wait_rebuild_finish_task", K(ret)); } else if (OB_FAIL(dag_->add_task(*wait_migrate_finish_task))) { LOG_WARN("failed to add wait_rebuild_finish_task", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_rebuild_tasks(const ObMigrateSrcInfo& src_info, ObPartitionMigrateCtx& part_migrate_ctx, const ObMigrateTableInfo& table_info, ObFakeTask& wait_rebuild_finish_task) { int ret = OB_SUCCESS; ObITask* parent_task = this; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if ((!src_info.is_valid() && RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_) || !part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("generate migrate tasks get invalid argument", K(ret), K(src_info), K(part_migrate_ctx)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(ctx_->replica_op_arg_.dst_.get_replica_type())) { ret = OB_ERR_SYS; LOG_WARN("no need to generate minor task", K(ret), K(ctx_->replica_op_arg_)); } else if (OB_FAIL(generate_major_tasks(src_info, part_migrate_ctx, table_info, parent_task))) { LOG_WARN("failed to generate rebuild major tasks", K(ret), K(table_info), K(part_migrate_ctx)); } else if (OB_FAIL(generate_minor_tasks(src_info, part_migrate_ctx, table_info, parent_task))) { LOG_WARN("failed to generate rebuild minor tasks", K(ret), K(table_info), K(part_migrate_ctx)); } else if (OB_FAIL(parent_task->add_child(wait_rebuild_finish_task))) { LOG_WARN("failed to add wait_rebuild_finish_task", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_migrate_tasks(const ObMigrateSrcInfo& src_info, ObPartitionMigrateCtx& part_migrate_ctx, const ObMigrateTableInfo& table_info, ObFakeTask& wait_migrate_finish_task) { int ret = OB_SUCCESS; ObITask* parent_task = this; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if ((RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && !src_info.is_valid()) || !part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("generate tasks get invalid argument", K(ret), K(src_info), K(part_migrate_ctx)); } else if (!ObReplicaTypeCheck::is_replica_with_ssstore(ctx_->replica_op_arg_.dst_.get_replica_type())) { ret = OB_ERR_SYS; LOG_WARN("no need to generate minor task", K(ret), K(ctx_->replica_op_arg_)); } else if (OB_FAIL(generate_minor_tasks(src_info, part_migrate_ctx, table_info, parent_task))) { LOG_WARN("failed to generate migrate minor tasks", K(ret), K(table_info), K(part_migrate_ctx)); } else if (OB_FAIL(generate_major_tasks(src_info, part_migrate_ctx, table_info, parent_task))) { LOG_WARN("failed to generate migrate major tasks", K(ret), K(table_info), K(part_migrate_ctx)); } else if (OB_FAIL(parent_task->add_child(wait_migrate_finish_task))) { LOG_WARN("failed to add wait_migrate_finish_task", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_major_tasks(const ObMigrateSrcInfo& src_info, ObPartitionMigrateCtx& part_migrate_ctx, const ObMigrateTableInfo& table_info, share::ObITask*& parent_task) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if ((!src_info.is_valid() && RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_) || OB_ISNULL(parent_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("generate migrate major task get invalid argument", K(ret), K(src_info), KP(parent_task)); } else { // add major sstables for (int64_t sstable_idx = 0; OB_SUCC(ret) && sstable_idx < table_info.major_sstables_.count(); ++sstable_idx) { const ObMigrateTableInfo::SSTableInfo& major_table_info = table_info.major_sstables_.at(sstable_idx); ObFakeTask* wait_finish_task = NULL; if (!ObITable::is_major_sstable(major_table_info.src_table_key_.table_type_)) { ret = OB_ERR_SYS; LOG_ERROR("table type not match major sstable", K(ret), K(major_table_info), K(table_info)); } else { if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_major_sstable_copy_task( src_info, part_migrate_ctx, major_table_info, parent_task, wait_finish_task))) { LOG_WARN("failed to generate major sstable copy task", K(ret), K(major_table_info)); } else if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_INFO("succeed to generate_major_sstable_copy_task", K(src_info), K(major_table_info)); } } } } return ret; } int ObMigrateTaskGeneratorTask::generate_minor_tasks(const ObMigrateSrcInfo& src_info, ObPartitionMigrateCtx& part_migrate_ctx, const ObMigrateTableInfo& table_info, share::ObITask*& parent_task) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if ((!src_info.is_valid() && RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_) || OB_ISNULL(parent_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("generate migrate minor task get invalid argument", K(ret), K(src_info), KP(parent_task)); } else { for (int64_t sstable_idx = 0; OB_SUCC(ret) && sstable_idx < table_info.minor_sstables_.count(); ++sstable_idx) { const ObMigrateTableInfo::SSTableInfo& minor_sstable_info = table_info.minor_sstables_.at(sstable_idx); ObFakeTask* wait_finish_task = NULL; if (!ObITable::is_minor_sstable(minor_sstable_info.src_table_key_.table_type_) && !ObITable::is_memtable(minor_sstable_info.src_table_key_.table_type_)) { ret = OB_ERR_SYS; LOG_ERROR("table type not match minor sstable", K(ret), K(minor_sstable_info), K(table_info)); } else { if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_minor_sstable_copy_task( parent_task, wait_finish_task, src_info, minor_sstable_info, part_migrate_ctx))) { LOG_WARN("failed to generate minor sstable copy task", K(ret), K(minor_sstable_info)); } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_INFO("succeed to generate_minor_sstable_copy_task", K(src_info), K(minor_sstable_info)); } } } } } return ret; } // now only support rebuild from old server, // if want support more condition, modified it int ObMigrateTaskGeneratorTask::generate_physic_minor_sstable_copy_task(const ObMigrateSrcInfo& src_info, ObPartitionMigrateCtx& part_migrate_ctx, const ObMigrateTableInfo::SSTableInfo& minor_sstable_info, ObITask* parent_task, ObITask* child_task) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task) || (RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && !src_info.is_valid()) || !minor_sstable_info.is_valid() || !part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(parent_task), KP(child_task), K(src_info), K(minor_sstable_info), K(ctx_->replica_op_arg_), K(part_migrate_ctx)); } else if (OB_FAIL(generate_physical_sstable_copy_task( src_info, part_migrate_ctx, minor_sstable_info, parent_task, child_task))) { LOG_WARN("fail to generate physic sstable copy task", K(ret), K(src_info), K(minor_sstable_info)); } return ret; } int ObITableTaskGeneratorTask::generate_logic_minor_sstable_copy_task(ObITask* parent_task, ObITask* child_task, const ObMigrateSrcInfo& src_info, const ObMigrateTableInfo::SSTableInfo& minor_sstable_info, ObPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; ObMigrateCopyLogicTask* copy_task = NULL; ObMigrateFinishLogicTask* finish_task = NULL; const int64_t task_idx = 0; const ObITable::TableKey& table_key = minor_sstable_info.src_table_key_; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task) || !src_info.is_valid() || !minor_sstable_info.is_valid() || !part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(parent_task), KP(child_task), K(src_info), K(minor_sstable_info), K(part_migrate_ctx)); } else if (table_key.pkey_.get_partition_id() != part_migrate_ctx.copy_info_.meta_.pkey_.get_partition_id()) { ret = OB_NOT_SUPPORTED; LOG_ERROR("cannot migrate partition with different partition id minor sstable", K(ret), K(table_key), "pkey", part_migrate_ctx.copy_info_.meta_.pkey_, "pg_key", ctx_->replica_op_arg_.key_); } else if (OB_FAIL(dag_->alloc_task(copy_task))) { LOG_WARN("failed to alloc copy task", K(ret)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc finish task", K(ret)); } else if (OB_FAIL(build_logic_sstable_ctx(src_info, minor_sstable_info, finish_task->get_sstable_ctx()))) { LOG_WARN("failed to build logic sstable ctx", K(ret), K(minor_sstable_info)); } else if (OB_FAIL(copy_task->init(task_idx, finish_task->get_sstable_ctx(), *ctx_))) { LOG_WARN("failed to init copy task", K(ret)); } else if (OB_FAIL(finish_task->init(part_migrate_ctx))) { LOG_WARN("failed to init finish task", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(parent_task->add_child(*copy_task))) { LOG_WARN("failed to add child copy task", K(ret)); } else if (OB_FAIL(copy_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish task", K(ret)); } else if (OB_FAIL(finish_task->add_child(*child_task))) { LOG_WARN("failed to add child", K(ret)); } else if (OB_FAIL(dag_->add_task(*copy_task))) { LOG_WARN("failed to add copy task to dag", K(ret)); } else if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish task to dag", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_minor_sstable_copy_task(share::ObITask* parent_task, share::ObITask* child_task, const ObMigrateSrcInfo& src_info, const ObMigrateTableInfo::SSTableInfo& minor_sstable_info, ObPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task) || (RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && !src_info.is_valid()) || !minor_sstable_info.is_valid() || !part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(parent_task), KP(child_task), K(src_info), K(minor_sstable_info), K(part_migrate_ctx)); } else { if (ObITable::is_minor_sstable(minor_sstable_info.src_table_key_.table_type_)) { if (OB_FAIL(generate_physic_minor_sstable_copy_task( src_info, part_migrate_ctx, minor_sstable_info, parent_task, child_task))) { LOG_WARN("failed to generate physic minor sstable copy task", K(ret), K(src_info), K(minor_sstable_info)); } } else { if (OB_FAIL(generate_logic_minor_sstable_copy_task( parent_task, child_task, src_info, minor_sstable_info, part_migrate_ctx))) { LOG_WARN("fail to generate logic minor sstable copy task", K(ret), K(src_info), K(minor_sstable_info)); } } } return ret; } int ObITableTaskGeneratorTask::build_logic_sstable_ctx(const ObMigrateSrcInfo& src_info, const ObMigrateTableInfo::SSTableInfo& minor_sstable_info, ObMigrateLogicSSTableCtx& ctx) { int ret = OB_SUCCESS; obrpc::ObFetchLogicBaseMetaArg arg; ObLogicBaseMetaReader reader; obrpc::ObFetchLogicRowArg logic_row_arg; ObLogicDataChecksumReader logic_data_checksum_reader; ObSchemaGetterGuard schema_guard; common::ObArray end_key_list; ctx.reset(); if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret), KP(ctx_)); } else { arg.table_key_ = minor_sstable_info.src_table_key_; arg.task_count_ = ObMigratePrepareTask::MAX_LOGIC_TASK_COUNT_PER_SSTABLE; } if (OB_FAIL(ret)) { } else if (OB_FAIL( reader.init(*srv_rpc_proxy_, *bandwidth_throttle_, src_info.src_addr_, arg, src_info.cluster_id_))) { LOG_WARN("failed to init logic meta reader", K(ret), K(src_info), K(arg)); } else if (OB_FAIL(reader.fetch_end_key_list(end_key_list))) { LOG_WARN("failed to fetch_end_key_list", K(ret), K(src_info), K(arg)); } else if (OB_FAIL(ctx.build_sub_task(end_key_list))) { LOG_WARN("failed to build sub task", K(ret)); } else if (FALSE_IT(ctx.meta_.schema_version_ = ctx_->pg_meta_.storage_info_.get_data_info().get_schema_version())) { // get remote memtable's max table_schema_version } else if (OB_FAIL(MIGRATOR.get_schema_service()->retry_get_schema_guard(ctx.meta_.schema_version_, minor_sstable_info.src_table_key_.table_id_, schema_guard, ctx.meta_.schema_version_))) { LOG_WARN("failed to get schema guard", K(minor_sstable_info), K(ret), K(ctx)); } else { ctx.src_info_ = src_info; ctx.table_key_ = minor_sstable_info.src_table_key_; ctx.dest_base_version_ = minor_sstable_info.dest_base_version_; LOG_INFO("succ to get logic migrate context", K(ctx)); } // fetch logic data checksum if (OB_FAIL(ret)) { } else { logic_row_arg.table_key_ = minor_sstable_info.src_table_key_; logic_row_arg.key_range_.set_whole_range(); logic_row_arg.schema_version_ = ctx.meta_.schema_version_; if (OB_FAIL(logic_data_checksum_reader.init( *srv_rpc_proxy_, *bandwidth_throttle_, src_info.src_addr_, src_info.cluster_id_, logic_row_arg))) { LOG_WARN("failed to init logic data checksum reader", K(ret)); } else if (OB_FAIL(logic_data_checksum_reader.get_data_checksum(ctx.data_checksum_))) { LOG_WARN("fail to get logic data checksum", K(ret)); } else { LOG_INFO("succ to get logic data checksum", K(ctx.data_checksum_)); } } return ret; } int ObITableTaskGeneratorTask::generate_physical_sstable_copy_task(const ObMigrateSrcInfo& src_info, ObIPartitionMigrateCtx& part_migrate_ctx, const ObMigrateTableInfo::SSTableInfo& sstable_info, ObITask* parent_task, ObITask* child_task) { int ret = OB_SUCCESS; ObMigrateCopyPhysicalTask* copy_task = NULL; ObMigrateFinishPhysicalTask* finish_task = NULL; const int64_t task_idx = 0; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret), KP(ctx_)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task) || (RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && !src_info.is_valid()) || !sstable_info.is_valid() || !part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN( "invalid args", K(ret), KP(parent_task), KP(child_task), K(src_info), K(sstable_info), K(part_migrate_ctx)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc finish task", K(ret)); } else if (OB_FAIL(build_physical_sstable_ctx(src_info, sstable_info, finish_task->get_sstable_ctx()))) { LOG_WARN("failed to build physical sstable ctx", K(ret), K(src_info), K(sstable_info)); } else if (OB_FAIL(finish_task->init(part_migrate_ctx))) { LOG_WARN("failed to init finish task", K(ret), K(part_migrate_ctx)); } else if (OB_FAIL(finish_task->add_child(*child_task))) { LOG_WARN("failed to add child", K(ret)); } if (OB_SUCC(ret)) { if (finish_task->get_sstable_ctx().task_count_ > 0) { // parent->copy->finish->child if (OB_FAIL(dag_->alloc_task(copy_task))) { LOG_WARN("failed to alloc copy task", K(ret)); } else if (OB_FAIL(copy_task->init(task_idx, finish_task->get_sstable_ctx(), *ctx_))) { LOG_WARN("failed to init copy task", K(ret)); } else if (OB_FAIL(parent_task->add_child(*copy_task))) { LOG_WARN("failed to add child copy task", K(ret)); } else if (OB_FAIL(copy_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish task", K(ret)); } else if (OB_FAIL(dag_->add_task(*copy_task))) { LOG_WARN("failed to add copy task to dag", K(ret)); } } else { // parent->finish->child if (OB_FAIL(parent_task->add_child(*finish_task))) { LOG_WARN("failed to add child finish_task for parent", K(ret)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish task to dag", K(ret)); } } return ret; } int ObMigrateTaskGeneratorTask::generate_major_sstable_copy_task(const ObMigrateSrcInfo& src_info, ObPartitionMigrateCtx& part_migrate_ctx, const ObMigrateTableInfo::SSTableInfo& major_sstable_info, ObITask* parent_task, ObITask* child_task) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(parent_task) || OB_ISNULL(child_task) || (RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && !src_info.is_valid()) || !major_sstable_info.is_valid() || !part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(parent_task), KP(child_task), K(src_info), K(major_sstable_info), K(part_migrate_ctx)); } else if (OB_FAIL(generate_physical_sstable_copy_task( src_info, part_migrate_ctx, major_sstable_info, parent_task, child_task))) { LOG_WARN("fail to generate physic sstable copy task", K(ret), K(major_sstable_info), K(src_info)); } return ret; } int ObMigrateTaskGeneratorTask::generate_finish_migrate_task( const ObIArray& last_task_array, share::ObITask*& last_task) { int ret = OB_SUCCESS; ObMigrateFinishTask* finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc task", K(ret)); } else if (OB_FAIL(finish_task->init(*ctx_))) { LOG_WARN("failed to init finish task", K(ret)); } if (OB_SUCC(ret)) { for (int64_t i = 0; OB_SUCC(ret) && i < last_task_array.count(); ++i) { ObITask* last_task = last_task_array.at(i); if (NULL != last_task) { if (OB_FAIL(last_task->add_child(*finish_task))) { LOG_WARN("failed to add child of last task", K(ret)); } } else { if (OB_FAIL(add_child(*finish_task))) { LOG_WARN("failed to add child of prepare task", K(ret)); } } } } if (OB_SUCC(ret)) { if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish task", K(ret)); } else { last_task = finish_task; } } return ret; } int ObITableTaskGeneratorTask::get_base_meta_reader( const ObMigrateSrcInfo& src_info, const obrpc::ObFetchPhysicalBaseMetaArg& arg, ObIPhysicalBaseMetaReader*& reader) { int ret = OB_SUCCESS; const int64_t compatible = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.compatible_; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret), KP(ctx_)); } else if (ctx_->replica_op_arg_.is_physical_restore_leader()) { if (OB_BACKUP_COMPATIBLE_VERSION_V1 == compatible || OB_BACKUP_COMPATIBLE_VERSION_V2 == compatible) { if (OB_FAIL(get_base_meta_restore_reader_v1(arg.table_key_, reader))) { STORAGE_LOG(WARN, "fail to get_base_meta_restore_reader_v1", K(ret)); } } else if (OB_BACKUP_COMPATIBLE_VERSION_V3 == compatible) { if (OB_FAIL(get_base_meta_restore_reader_v2(arg.table_key_, reader))) { STORAGE_LOG(WARN, "fail to get_base_meta_restore_reader_v2", K(ret)); } } else { ret = OB_ERR_UNEXPECTED; LOG_WARN("restore compatible version is expected", K(ret), K(compatible), K(arg)); } } else if (RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_NOT_SUPPORTED; LOG_ERROR("not support logical restore", K(ret), "replica_op_arg", ctx_->replica_op_arg_); } else if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(get_base_meta_backup_reader(arg.table_key_, reader))) { STORAGE_LOG(WARN, "fail to get_base_meta_backup_reader", K(ret)); } } else { if (OB_FAIL(get_base_meta_ob_reader(src_info, arg, reader))) { STORAGE_LOG(WARN, "failed to get_macro_block_ob_reader", K(ret)); } } return ret; } int ObITableTaskGeneratorTask::get_base_meta_restore_reader_v1( const ObITable::TableKey& table_key, ObIPhysicalBaseMetaReader*& reader) { int ret = OB_SUCCESS; ObPhysicalBaseMetaRestoreReaderV1* tmp_reader = NULL; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret), KP(ctx_)); } else if (OB_ISNULL(ctx_->restore_meta_reader_)) { ret = OB_ERR_SYS; LOG_ERROR("restore meta reader must not null", K(ret)); } else if (NULL == (tmp_reader = cp_fty_->get_base_data_meta_restore_reader_v1())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_FAIL(tmp_reader->init(*bandwidth_throttle_, ctx_->replica_op_arg_.phy_restore_arg_, table_key, *ctx_->restore_meta_reader_))) { STORAGE_LOG(WARN, "failed to init ob reader", K(ret)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != reader) { cp_fty_->free(reader); reader = NULL; } } if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObITableTaskGeneratorTask::get_base_meta_restore_reader_v2( const ObITable::TableKey& table_key, ObIPhysicalBaseMetaReader*& reader) { int ret = OB_SUCCESS; ObPhysicalBaseMetaRestoreReaderV2* tmp_reader = NULL; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret), KP(ctx_)); } else if (OB_ISNULL(ctx_->restore_meta_reader_)) { ret = OB_ERR_SYS; LOG_ERROR("restore meta reader must not null", K(ret)); } else if (NULL == (tmp_reader = cp_fty_->get_base_data_meta_restore_reader_v2())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_FAIL(tmp_reader->init(*bandwidth_throttle_, ctx_->replica_op_arg_.phy_restore_arg_, table_key, *ctx_->restore_meta_reader_))) { STORAGE_LOG(WARN, "failed to init ob reader", K(ret)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != reader) { cp_fty_->free(reader); reader = NULL; } } if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } // TODO() delete it later int ObITableTaskGeneratorTask::get_base_meta_backup_reader( const ObITable::TableKey& table_key, ObIPhysicalBaseMetaReader*& reader) { int ret = OB_SUCCESS; ObPhysicalBaseMetaBackupReader* tmp_reader = NULL; UNUSED(table_key); if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret), KP(ctx_)); } else if (OB_ISNULL(tmp_reader = cp_fty_->get_base_data_meta_backup_reader())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get base data meta backup reader", K(ret)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != reader) { cp_fty_->free(reader); reader = NULL; } } if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObITableTaskGeneratorTask::get_base_meta_ob_reader( const ObMigrateSrcInfo& src_info, const obrpc::ObFetchPhysicalBaseMetaArg& arg, ObIPhysicalBaseMetaReader*& reader) { int ret = OB_SUCCESS; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret), KP(ctx_)); } else if (OB_FAIL(get_base_meta_rpc_reader(src_info, arg, reader))) { STORAGE_LOG(WARN, "fail to get base meta rpc reader", K(ret), K(src_info), K(arg)); } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != reader) { cp_fty_->free(reader); reader = NULL; } } } return ret; } int ObITableTaskGeneratorTask::get_base_meta_rpc_reader( const ObMigrateSrcInfo& src_info, const obrpc::ObFetchPhysicalBaseMetaArg& arg, ObIPhysicalBaseMetaReader*& reader) { int ret = OB_SUCCESS; ObPhysicalBaseMetaReader* tmp_reader = NULL; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret), KP(ctx_)); } else if (NULL == (tmp_reader = cp_fty_->get_base_data_meta_ob_reader())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_FAIL(tmp_reader->init( *srv_rpc_proxy_, *bandwidth_throttle_, src_info.src_addr_, arg, src_info.cluster_id_))) { STORAGE_LOG(WARN, "failed to init ob reader", K(ret)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } } return ret; } int ObITableTaskGeneratorTask::build_physical_sstable_ctx(const ObMigrateSrcInfo& src_info, const ObMigrateTableInfo::SSTableInfo& sstable_info, ObMigratePhysicalSSTableCtx& ctx) { int ret = OB_SUCCESS; obrpc::ObFetchPhysicalBaseMetaArg arg; common::ObArray macro_block_list; ObTablesHandle handle; ctx.reset(); ObIPhysicalBaseMetaReader* reader = NULL; arg.table_key_ = sstable_info.src_table_key_; ctx.sstable_info_ = sstable_info; ctx.src_info_ = src_info; ctx.is_leader_restore_ = ctx_->replica_op_arg_.is_physical_restore_leader(); ctx.restore_version_ = ctx_->replica_op_arg_.phy_restore_arg_.restore_data_version_; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret), KP(ctx_)); } else if (OB_FAIL(get_base_meta_reader(src_info, arg, reader))) { LOG_WARN("fail to get base meta reader", K(ret)); } else if (OB_ISNULL(reader)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("reader should not be NULL", K(ret), KP(reader)); } if (OB_FAIL(ret)) { } else if (OB_FAIL(reader->fetch_sstable_meta(ctx.meta_))) { LOG_WARN("failed to fetch physic table meta", K(ret), K(src_info), K(arg)); } else if (OB_FAIL(reader->fetch_macro_block_list(macro_block_list))) { LOG_WARN("failed to fetch_macro_block_list", K(ret), K(src_info), K(arg)); } else if (OB_FAIL(calc_macro_block_reuse(sstable_info.src_table_key_, macro_block_list))) { LOG_WARN("fail to calc macro block reuse", K(ret)); } else if (OB_FAIL(build_sub_physical_task(ctx, macro_block_list))) { LOG_WARN("failed to build sub task", K(ret)); } else { ATOMIC_AAF(&ctx_->data_statics_.total_macro_block_, macro_block_list.count()); LOG_INFO("succeed to build physical sstable ctx", K(ctx)); } if (NULL != reader) { cp_fty_->free(reader); } return ret; } int ObITableTaskGeneratorTask::build_sub_physical_task( ObMigratePhysicalSSTableCtx& ctx, common::ObIArray& macro_block_list) { int ret = OB_SUCCESS; void* buf = NULL; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret), KP(ctx_)); } else if (0 == macro_block_list.count()) { ctx.task_count_ = 0; LOG_INFO("no macro block need copy", K(ret)); } else { const int64_t max_macro_block_count_per_task = ObMigratePrepareTask::MAX_MACRO_BLOCK_COUNT_PER_TASK; ctx.task_count_ = macro_block_list.count() / max_macro_block_count_per_task; if (macro_block_list.count() % max_macro_block_count_per_task > 0) { ++ctx.task_count_; } if (OB_ISNULL(buf = ctx.allocator_.alloc(sizeof(ObMigratePhysicalSSTableCtx::SubTask) * ctx.task_count_))) { ret = OB_ALLOCATE_MEMORY_FAILED; LOG_WARN("failed to alloc buf", K(ret)); } else { ctx.tasks_ = new (buf) ObMigratePhysicalSSTableCtx::SubTask[ctx.task_count_]; buf = NULL; } for (int64_t task_idx = 0; OB_SUCC(ret) && task_idx < ctx.task_count_; ++task_idx) { ObMigratePhysicalSSTableCtx::SubTask& sub_task = ctx.tasks_[task_idx]; sub_task.pkey_ = ctx.sstable_info_.src_table_key_.pkey_; sub_task.block_count_ = std::min( max_macro_block_count_per_task, macro_block_list.count() - task_idx * max_macro_block_count_per_task); if (OB_ISNULL(buf = ctx.allocator_.alloc(sizeof(ObMigrateMacroBlockInfo) * sub_task.block_count_))) { ret = OB_ALLOCATE_MEMORY_FAILED; LOG_WARN("failed to alloc buf", K(ret)); } else { sub_task.block_info_ = new (buf) ObMigrateMacroBlockInfo[sub_task.block_count_]; buf = NULL; } for (int64_t i = 0; OB_SUCC(ret) && i < sub_task.block_count_; ++i) { const int64_t macro_idx = task_idx * max_macro_block_count_per_task + i; if (macro_idx >= macro_block_list.count()) { break; } const blocksstable::ObSSTablePair& pair = macro_block_list.at(macro_idx); ObMigrateMacroBlockInfo& info = ctx.tasks_[task_idx].block_info_[i]; info.pair_ = pair; info.need_copy_ = true; } } } return ret; } int ObITableTaskGeneratorTask::calc_macro_block_reuse( const ObITable::TableKey& table_key, ObIArray& macro_block_list) { int ret = OB_SUCCESS; if (OB_ISNULL(ctx_)) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret), KP(ctx_)); } else if (0 == macro_block_list.count()) { // do nothing } else if (OB_FAIL(ctx_->reuse_block_mgr_.build_reuse_macro_map(*ctx_, table_key, macro_block_list))) { LOG_WARN("fail to build reuse macro map", K(ret)); } return ret; } int ObMigratePrepareTask::build_index_partition_info( const ObReplicaOpArg& arg, ObIPGPartitionBaseDataMetaObReader* reader) { int ret = OB_SUCCESS; ObPartitionMigrateCtx part_migrate_ctx; ObArray table_info_res; ObArray table_id_list; bool found = false; ObPGPartitionMetaInfo partition_meta_info; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate preprae task do not init", K(ret)); } else if (!arg.is_valid() || OB_ISNULL(reader)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("build index partition info get invalid argument", K(ret), K(arg), KP(reader)); } else if (COPY_LOCAL_INDEX_OP != arg.type_) { ret = OB_ERR_SYS; LOG_WARN("replica op type do not match", K(ret), K(arg)); } else { while (OB_SUCC(ret)) { partition_meta_info.reset(); table_info_res.reuse(); table_id_list.reuse(); part_migrate_ctx.reset(); found = false; if (OB_FAIL(reader->fetch_pg_partition_meta_info(partition_meta_info))) { if (OB_ITER_END != ret) { LOG_WARN("fail to fetch pg partition meta info", K(ret), K(ctx_->replica_op_arg_)); } else { ret = OB_SUCCESS; break; } } else { for (int64_t i = 0; OB_SUCC(ret) && i < partition_meta_info.table_id_list_.count() && !found; ++i) { found = arg.index_id_ == partition_meta_info.table_id_list_.at(i); if (found) { if (OB_FAIL(table_id_list.push_back(partition_meta_info.table_id_list_.at(i)))) { LOG_WARN("failed to push table id into array", K(ret), K(arg)); } else if (OB_FAIL(table_info_res.push_back(partition_meta_info.table_info_.at(i)))) { LOG_WARN("failed to push table info into array", K(ret), K(partition_meta_info), K(i)); } } } } if (OB_SUCC(ret)) { if (found) { if (OB_FAIL( build_migrate_partition_info(partition_meta_info, table_info_res, table_id_list, part_migrate_ctx))) { LOG_WARN("fail to build migrate partition info", K(ret), K(partition_meta_info)); } else if (OB_FAIL(ctx_->part_ctx_array_.push_back(part_migrate_ctx))) { LOG_WARN("fail to push part migrate ctx into array", K(ret), K(part_migrate_ctx)); } else { break; } } } } if (OB_SUCC(ret)) { if (0 == ctx_->part_ctx_array_.count()) { ret = OB_ERR_SYS; FLOG_WARN("local index can not find part ctx", K(ret), K(*ctx_)); } } } return ret; } int ObMigratePrepareTask::build_table_partition_info( const ObReplicaOpArg& arg, ObIPGPartitionBaseDataMetaObReader* reader) { int ret = OB_SUCCESS; ObPartitionMigrateCtx part_migrate_ctx; ObArray table_info_res; ObArray table_id_list; ObPGPartitionMetaInfo partition_meta_info; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (!arg.is_valid() || OB_ISNULL(reader)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("build table partition info get invalid argument", K(ret), K(arg), KP(reader)); } else if (OB_UNLIKELY(COPY_LOCAL_INDEX_OP == arg.type_)) { ret = OB_ERR_SYS; LOG_WARN("replica op type do not match", K(ret), K(arg)); } else { while (OB_SUCC(ret)) { partition_meta_info.reset(); table_info_res.reuse(); table_id_list.reuse(); part_migrate_ctx.reset(); if (OB_FAIL(reader->fetch_pg_partition_meta_info(partition_meta_info))) { if (OB_ITER_END != ret) { LOG_WARN("fail to fetch pg partition meta info", K(ret), K(ctx_->replica_op_arg_)); } else { ret = OB_SUCCESS; break; } } else if (OB_FAIL(table_id_list.assign(partition_meta_info.table_id_list_))) { LOG_WARN("failed to assign table id list", K(ret)); } else if (OB_FAIL(table_info_res.assign(partition_meta_info.table_info_))) { LOG_WARN("failed to copy needed table info", K(ret), K(partition_meta_info)); } else if (OB_FAIL(build_migrate_partition_info( partition_meta_info, table_info_res, table_id_list, part_migrate_ctx))) { LOG_WARN("fail to build migrate partition info", K(ret), K(partition_meta_info)); } else if (OB_FAIL(ctx_->part_ctx_array_.push_back(part_migrate_ctx))) { LOG_WARN("fail to push part migrate ctx into array", K(ret), K(part_migrate_ctx)); } if (OB_SUCC(ret) && ctx_->replica_op_arg_.is_physical_restore_follower()) { if (OB_FAIL(create_partition_if_needed(ctx_->partition_guard_, partition_meta_info.meta_))) { LOG_WARN("fail to create partition for follower restore", K(ret), K(arg), K(partition_meta_info.meta_)); } } } } return ret; } int ObMigratePrepareTask::remove_uncontinue_local_tables(const ObPartitionKey& pkey, const uint64_t table_id) { int ret = OB_SUCCESS; ObPGPartitionGuard guard; ObIPartitionGroup* partition = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (!pkey.is_valid() || 0 == table_id || OB_INVALID_ID == table_id) { ret = OB_INVALID_ARGUMENT; LOG_WARN("remove uncontinue local tables get invalid argument", K(ret), K(pkey), K(table_id)); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group())) { // do nothing LOG_DEBUG("partition not exist, skip remove uncontinue local tables", K(pkey), K(table_id)); } else if (OB_FAIL(partition->get_pg_storage().remove_uncontinues_inc_tables(pkey, table_id))) { LOG_WARN("failed to remove uncontinue local tables", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::check_partition_integrity() { int ret = OB_SUCCESS; ObPGPartitionGuard guard; ObIPartitionGroup* partition = NULL; ObPartitionArray local_pkeys; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (COPY_LOCAL_INDEX_OP == ctx_->replica_op_arg_.type_ || BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_ || VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_ || BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_ || BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need check partition integrity", "arg", ctx_->replica_op_arg_); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_WARN("partition should not be NULL", K(ret), KP(partition)); } else if (OB_FAIL(partition->get_all_pg_partition_keys(local_pkeys, true /*include_trans_table*/))) { LOG_WARN("failed to get pg partition keys", K(ret), K(*ctx_)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { ObPartitionMigrateCtx& pctx = ctx_->part_ctx_array_.at(i); const ObPartitionKey& remote_pkey = pctx.copy_info_.meta_.pkey_; pctx.is_partition_exist_ = false; for (int64_t j = 0; OB_SUCC(ret) && j < local_pkeys.count() && !pctx.is_partition_exist_; ++j) { const ObPartitionKey& local_pkey = local_pkeys.at(j); if (local_pkey == remote_pkey) { pctx.is_partition_exist_ = true; } } LOG_INFO("check partition exist", K(remote_pkey), "is_exist", pctx.is_partition_exist_); } } return ret; } int ObMigratePrepareTask::create_pg_partition( const ObPGPartitionStoreMeta& meta, ObIPartitionGroup* partition, const bool in_slog_trans) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (!meta.is_valid() || OB_ISNULL(partition)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("create pg partition get invalid argument", K(ret), K(meta), KP(partition)); } else { ObTablesHandle sstables_handle; // empty obrpc::ObCreatePartitionArg arg; const ObSavedStorageInfoV2& saved_storage_info = ctx_->pg_meta_.storage_info_; arg.schema_version_ = saved_storage_info.get_data_info().get_schema_version(); arg.lease_start_ = meta.create_timestamp_; arg.restore_ = ctx_->is_restore_; const int64_t multi_version_start = meta.multi_version_start_; const int64_t data_table_id = meta.data_table_id_; const uint64_t unused = 0; const bool is_replay = false; if (OB_FAIL(partition->create_pg_partition( meta.pkey_, multi_version_start, data_table_id, arg, in_slog_trans, is_replay, unused, sstables_handle))) { LOG_WARN("fail to create pg partition", K(ret), K(meta)); } } return ret; } int ObMigratePrepareTask::inner_get_partition_table_info_reader( const ObMigrateSrcInfo& src_info, ObIPGPartitionBaseDataMetaObReader*& reader) { int ret = OB_SUCCESS; UNUSED(src_info); const bool is_copy_index = COPY_LOCAL_INDEX_OP == ctx_->replica_op_arg_.type_ || COPY_GLOBAL_INDEX_OP == ctx_->replica_op_arg_.type_ || LINK_SHARE_MAJOR_OP == ctx_->replica_op_arg_.type_; ObPGPartitionBaseDataMetaObReader* tmp_reader = NULL; obrpc::ObFetchPGPartitionInfoArg rpc_arg; rpc_arg.pg_key_ = ctx_->replica_op_arg_.key_; rpc_arg.snapshot_version_ = ctx_->pg_meta_.storage_info_.get_data_info().get_publish_version(); rpc_arg.log_ts_ = ctx_->pg_meta_.storage_info_.get_data_info().get_last_replay_log_ts(); rpc_arg.is_only_major_sstable_ = is_copy_index; if (OB_ISNULL(cp_fty_)) { ret = OB_ERR_SYS; LOG_ERROR("cp fty should not be NULL", K(ret), KP(cp_fty_)); } else if (NULL == (tmp_reader = cp_fty_->get_pg_info_reader())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_FAIL(tmp_reader->init( *srv_rpc_proxy_, *bandwidth_throttle_, src_info.src_addr_, rpc_arg, src_info.cluster_id_))) { LOG_WARN("fail to init partition meta info reader", K(ret), K(src_info)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObMigratePrepareTask::inner_get_partition_table_info_restore_reader_v1( const ObMigrateSrcInfo& src_info, ObIPGPartitionBaseDataMetaObReader*& reader) { int ret = OB_SUCCESS; UNUSED(src_info); ObPGPartitionBaseDataMetaRestoreReaderV1* tmp_reader = NULL; ObPartitionGroupMetaRestoreReaderV1* restore_meta_reader = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(ctx_->restore_meta_reader_)) { ret = OB_ERR_SYS; LOG_ERROR("restore_meta_reader_ must not null", K(ret)); } else if (OB_ISNULL(cp_fty_)) { ret = OB_ERR_SYS; LOG_ERROR("cp fty should not be NULL", K(ret), KP(cp_fty_)); } else if (OB_ISNULL(tmp_reader = cp_fty_->get_pg_info_restore_reader_v1())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_ISNULL(ctx_->restore_meta_reader_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "phaysical restore macro index should not be NULL", K(ret), KP(ctx_->restore_meta_reader_)); } else if (ObIPartitionGroupMetaRestoreReader::PG_META_RESTORE_READER_V1 != ctx_->restore_meta_reader_->get_type()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "restore meta reader type is unexpected", K(ret), K(ctx_->restore_meta_reader_->get_type())); } else if (FALSE_IT(restore_meta_reader = reinterpret_cast(ctx_->restore_meta_reader_))) { } else if (OB_FAIL(tmp_reader->init(ctx_->pg_meta_.partitions_, restore_meta_reader))) { LOG_WARN("fail to init partition table info restore reader", K(ret)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObMigratePrepareTask::inner_get_partition_table_info_restore_reader_v2( const ObMigrateSrcInfo& src_info, ObIPGPartitionBaseDataMetaObReader*& reader) { int ret = OB_SUCCESS; UNUSED(src_info); ObPGPartitionBaseDataMetaRestoreReaderV2* tmp_reader = NULL; ObPartitionGroupMetaRestoreReaderV2* restore_meta_reader = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(ctx_->restore_meta_reader_)) { ret = OB_ERR_SYS; LOG_ERROR("restore_meta_reader_ must not null", K(ret)); } else if (OB_ISNULL(cp_fty_)) { ret = OB_ERR_SYS; LOG_ERROR("cp fty should not be NULL", K(ret), KP(cp_fty_)); } else if (OB_ISNULL(tmp_reader = cp_fty_->get_pg_info_restore_reader_v2())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_ISNULL(ctx_->restore_meta_reader_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "phaysical restore macro index should not be NULL", K(ret), KP(restore_meta_reader)); } else if (ObIPartitionGroupMetaRestoreReader::PG_META_RESTORE_READER_V2 != ctx_->restore_meta_reader_->get_type()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "restore meta reader type is unexpected", K(ret), K(ctx_->restore_meta_reader_->get_type())); } else if (FALSE_IT(restore_meta_reader = reinterpret_cast(ctx_->restore_meta_reader_))) { } else if (OB_FAIL(tmp_reader->init(ctx_->pg_meta_.partitions_, restore_meta_reader))) { LOG_WARN("fail to init partition table info restore reader", K(ret)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } // TODO(muwei.ym) delete it later int ObMigratePrepareTask::inner_get_partition_table_info_backup_reader( const ObMigrateSrcInfo& src_info, ObIPGPartitionBaseDataMetaObReader*& reader) { int ret = OB_SUCCESS; UNUSED(src_info); ObPGPartitionBaseDataMetaBackupReader* tmp_reader = NULL; if (OB_ISNULL(cp_fty_)) { ret = OB_ERR_SYS; LOG_ERROR("cp fty should not be NULL", K(ret), KP(cp_fty_)); } else if (OB_ISNULL(tmp_reader = cp_fty_->get_pg_info_backup_reader())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObMigratePrepareTask::get_partition_table_info_reader( const ObMigrateSrcInfo& src_info, ObIPGPartitionBaseDataMetaObReader*& reader) { int ret = OB_SUCCESS; reader = NULL; const int64_t compatible = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.compatible_; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (!src_info.is_valid() && RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_) { ret = OB_INVALID_ARGUMENT; LOG_WARN( "get partition table info reader get invalid argument", K(ret), K(src_info), K(ctx_->replica_op_arg_.type_)); } else if (ctx_->replica_op_arg_.is_physical_restore_leader() && share::REPLICA_RESTORE_DATA == ctx_->is_restore_) { if (OB_BACKUP_COMPATIBLE_VERSION_V1 == compatible || OB_BACKUP_COMPATIBLE_VERSION_V2 == compatible) { if (OB_FAIL(inner_get_partition_table_info_restore_reader_v1(src_info, reader))) { LOG_WARN("failed to get partition table info restore reader", K(ret)); } } else if (OB_BACKUP_COMPATIBLE_VERSION_V3 == compatible) { if (OB_FAIL(inner_get_partition_table_info_restore_reader_v2(src_info, reader))) { LOG_WARN("failed to get partition table info restore reader", K(ret)); } } else { ret = OB_ERR_UNEXPECTED; LOG_WARN("restore compatible version is expected", K(ret), K(compatible)); } } else if (RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_ && share::REPLICA_LOGICAL_RESTORE_DATA == ctx_->is_restore_) { ret = OB_NOT_SUPPORTED; LOG_ERROR("not support logical restore", K(ret)); } else if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(inner_get_partition_table_info_backup_reader(src_info, reader))) { LOG_WARN("failed to get partition table info backup reader", K(ret)); } } else { if (OB_FAIL(inner_get_partition_table_info_reader(src_info, reader))) { LOG_WARN("failed to get partition table info reader", K(ret)); } } return ret; } int ObMigratePrepareTask::check_backup_data_continues() { int ret = OB_SUCCESS; ObIPartitionGroup* partition_group = NULL; clog::ObPGLogArchiveStatus pg_log_archive_status; ObLogArchiveBackupInfo archive_backup_info; const int64_t start_ts = ObTimeUtility::current_time(); static const int64_t OB_MAX_RETRY_TIME = 60 * 1000 * 1000; // 60s static const int64_t OB_SINGLE_SLEEP_US = 1 * 1000 * 1000; // 1s const int64_t last_replay_log_id = ctx_->pg_meta_.storage_info_.get_clog_info().get_last_replay_log_id(); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (BACKUP_REPLICA_OP != ctx_->replica_op_arg_.type_) { // do nothing } else if (OB_ISNULL(partition_group = ctx_->get_partition())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("backup local partition should not be NULL", K(ret), KP(partition_group), K(*ctx_)); } else { while (OB_SUCC(ret)) { if (OB_FAIL(partition_group->get_log_archive_status(pg_log_archive_status))) { LOG_WARN("failed to get log archive status", K(ret), K(*ctx_)); } else if (OB_FAIL(ObBackupInfoMgr::get_instance().get_log_archive_backup_info(archive_backup_info))) { LOG_WARN("failed to get log archive backup info", K(ret)); } else if (ObLogArchiveStatus::INVALID == pg_log_archive_status.status_) { if (ObLogArchiveStatus::DOING == archive_backup_info.status_.status_) { LOG_WARN("partition group log archive status is invalid, need retry", K(pg_log_archive_status), K(archive_backup_info)); ret = OB_EAGAIN; } else { ret = OB_ERR_UNEXPECTED; LOG_WARN( "server do not archive log, can not backup", K(ret), K(pg_log_archive_status), K(archive_backup_info)); } } else if (ObLogArchiveStatus::DOING == pg_log_archive_status.status_) { if (pg_log_archive_status.archive_incarnation_ != archive_backup_info.status_.incarnation_ || pg_log_archive_status.log_archive_round_ != archive_backup_info.status_.round_) { ret = OB_ERR_UNEXPECTED; LOG_WARN("pg archive round or incarnation not equal to server archive info", K(ret), K(pg_log_archive_status), K(archive_backup_info)); } else if (pg_log_archive_status.round_start_log_id_ < 1) { ret = OB_LOG_ARCHIVE_STAT_NOT_MATCH; LOG_WARN("log archive status is not match", K(ret), K(pg_log_archive_status)); } else if (last_replay_log_id + 1 < pg_log_archive_status.round_start_log_id_) { ret = OB_ARCHIVE_LOG_NOT_CONTINUES_WITH_DATA; LOG_WARN("base data and log archive data do not continues", K(ret), K(pg_log_archive_status), K(ctx_->pg_meta_), K(last_replay_log_id)); } else { break; } } else { ret = OB_LOG_ARCHIVE_STAT_NOT_MATCH; LOG_WARN("log archive status is unexpected", K(ret), K(pg_log_archive_status)); } if (OB_EAGAIN == ret) { const int64_t cost_ts = ObTimeUtility::current_time() - start_ts; if (cost_ts < OB_MAX_RETRY_TIME) { usleep(OB_SINGLE_SLEEP_US); ret = OB_SUCCESS; } else { ret = OB_PG_LOG_ARCHIVE_STATUS_NOT_INIT; LOG_WARN("server start do archive log, but partition retry many times status still wrong", K(ret), K(pg_log_archive_status), K(archive_backup_info), K(*ctx_)); } } } } return ret; } ObMigrateLogicSSTableCtx::SubLogicTask::SubLogicTask() : end_key_(), block_write_ctx_(), lob_block_write_ctx_() {} ObMigrateLogicSSTableCtx::SubLogicTask::~SubLogicTask() {} ObMigrateLogicSSTableCtx::ObMigrateLogicSSTableCtx() : tasks_(NULL), task_count_(0), meta_(), allocator_(ObModIds::OB_PARTITION_MIGRATOR), src_info_(), table_key_(), dest_base_version_(-1), data_checksum_(0) {} ObMigrateLogicSSTableCtx::~ObMigrateLogicSSTableCtx() { reset(); } void ObMigrateLogicSSTableCtx::reset() { for (int64_t i = 0; i < task_count_; ++i) { tasks_[i].~SubLogicTask(); } tasks_ = NULL; task_count_ = 0; meta_.reset(); allocator_.reset(); src_info_.reset(); table_key_.reset(); dest_base_version_ = -1; data_checksum_ = 0; } bool ObMigrateLogicSSTableCtx::is_valid() { bool valid = true; if (NULL == tasks_ || task_count_ <= 0 || !meta_.is_valid() || !src_info_.is_valid() || !table_key_.is_valid() || dest_base_version_ < table_key_.trans_version_range_.base_version_ || dest_base_version_ >= table_key_.trans_version_range_.snapshot_version_) { valid = false; } else { for (int64_t i = 0; valid && i < task_count_; ++i) { if (!tasks_[i].end_key_.is_valid()) { valid = false; } } } return valid; } int ObMigrateLogicSSTableCtx::build_sub_task(const common::ObIArray& end_key_list) { int ret = OB_SUCCESS; void* buf = NULL; if (NULL != tasks_) { ret = OB_INIT_TWICE; LOG_WARN("cannot init twice", K(ret)); } else if (end_key_list.empty()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), K(end_key_list)); } else if (OB_ISNULL(buf = allocator_.alloc(sizeof(SubLogicTask) * end_key_list.count()))) { ret = OB_ALLOCATE_MEMORY_FAILED; LOG_WARN("failed to alloc sub logic task buf", K(ret)); } else { task_count_ = end_key_list.count(); tasks_ = new (buf) SubLogicTask[task_count_]; for (int64_t i = 0; OB_SUCC(ret) && i < task_count_; ++i) { if (OB_FAIL(end_key_list.at(i).deep_copy(tasks_[i].end_key_, allocator_))) { LOG_WARN("failed to deep copy end key", K(ret)); } if (OB_SUCC(ret) && i == task_count_ - 1) { if (!tasks_[i].end_key_.is_max()) { ret = OB_ERR_SYS; LOG_ERROR("the last end key must be max row", K(ret), K(tasks_[i].end_key_), K(end_key_list)); } } } } return ret; } int ObMigrateLogicSSTableCtx::get_sub_task(const int64_t idx, SubLogicTask*& sub_task) { int ret = OB_SUCCESS; sub_task = NULL; if (OB_ISNULL(tasks_)) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret), K(*this)); } else if (idx < 0 || idx >= task_count_) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), K(idx), K(*this)); } else { sub_task = tasks_ + idx; } return ret; } int ObMigrateLogicSSTableCtx::get_dest_table_key(ObITable::TableKey& dest_table_key) { int ret = OB_SUCCESS; if (!is_valid()) { ret = OB_NOT_INIT; LOG_WARN("invalid ctx", K(ret), K(*this)); } else if (table_key_.trans_version_range_.base_version_ == dest_base_version_) { dest_table_key = table_key_; dest_table_key.table_type_ = ObITable::MINI_MINOR_SSTABLE; } else { dest_table_key = table_key_; dest_table_key.table_type_ = ObITable::MINI_MINOR_SSTABLE; dest_table_key.trans_version_range_.base_version_ = dest_base_version_; dest_table_key.version_ = 0; if (dest_base_version_ > dest_table_key.trans_version_range_.multi_version_start_) { dest_table_key.trans_version_range_.multi_version_start_ = dest_base_version_; } LOG_INFO( "base version not match, rewrite trans version range", K(dest_base_version_), K_(table_key), K(dest_table_key)); } return ret; } ObMigrateCopyLogicTask::ObMigrateCopyLogicTask() : ObITask(TASK_TYPE_MIGRATE_COPY_LOGIC), is_inited_(false), ctx_(NULL), task_idx_(-1), sstable_ctx_(NULL), sub_task_(NULL) {} ObMigrateCopyLogicTask::~ObMigrateCopyLogicTask() {} int ObMigrateCopyLogicTask::init(const int64_t task_idx, ObMigrateLogicSSTableCtx& sstable_ctx, ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (task_idx < 0 || task_idx >= sstable_ctx.task_count_ || !sstable_ctx.is_valid() || !ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), K(task_idx), K(sstable_ctx), K(ctx)); } else { ctx_ = &ctx; task_idx_ = task_idx; sstable_ctx_ = &sstable_ctx; sub_task_ = sstable_ctx.tasks_ + task_idx_; is_inited_ = true; } return ret; } int ObMigrateCopyLogicTask::generate_next_task(ObITask*& next_task) { int ret = OB_SUCCESS; const int64_t next_task_idx = task_idx_ + 1; ObMigrateCopyLogicTask* tmp_next_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not init", K(ret)); } else if (task_idx_ + 1 == sstable_ctx_->task_count_) { ret = OB_ITER_END; } else if (OB_FAIL(dag_->alloc_task(tmp_next_task))) { LOG_WARN("failed to alloc task", K(ret)); } else if (OB_FAIL(tmp_next_task->init(next_task_idx, *sstable_ctx_, *ctx_))) { LOG_WARN("failed o init next task", K(ret)); } else { next_task = tmp_next_task; } return ret; } int ObMigrateCopyLogicTask::process() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObLogicRowFetcher fetcher; obrpc::ObFetchLogicRowArg reader_arg; ObMigrateLogicRowWriter writer; obrpc::ObPartitionServiceRpcProxy* srv_rpc_proxy = MIGRATOR.get_svr_rpc_proxy(); common::ObInOutBandwidthThrottle* bandwidth_throttle = MIGRATOR.get_bandwidth_throttle(); DEBUG_SYNC(BEFORE_MIGRATE_COPY_LOGIC_DATA); if (NULL != sstable_ctx_) { LOG_INFO("start ObMigrateCopyLogicTask process", "table_key", sstable_ctx_->table_key_, "dest_base_version", sstable_ctx_->dest_base_version_); } if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_ISNULL(srv_rpc_proxy) || OB_ISNULL(bandwidth_throttle)) { ret = OB_ERR_SYS; LOG_ERROR("srv_rpc_proxy or bandwidth_throttle must not null", K(ret), KP(srv_rpc_proxy), KP(bandwidth_throttle)); } else { ctx_->action_ = ObMigrateCtx::COPY_MINOR; if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { LOG_WARN("failed to update_partition_migration_status", K(tmp_ret)); } reader_arg.table_key_ = sstable_ctx_->table_key_; reader_arg.key_range_.set_left_open(); reader_arg.schema_version_ = sstable_ctx_->meta_.schema_version_; if (task_idx_ == 0) { reader_arg.key_range_.set_start_key(ObStoreRowkey::MIN_STORE_ROWKEY); } else { reader_arg.key_range_.set_start_key(sstable_ctx_->tasks_[task_idx_ - 1].end_key_); } if (task_idx_ == sstable_ctx_->task_count_ - 1) { reader_arg.key_range_.set_end_key(ObStoreRowkey::MAX_STORE_ROWKEY); reader_arg.key_range_.set_right_open(); } else { reader_arg.key_range_.set_end_key(sub_task_->end_key_); reader_arg.key_range_.set_right_closed(); } reader_arg.data_checksum_ = sstable_ctx_->data_checksum_; if (OB_FAIL(fetcher.init(*srv_rpc_proxy, *bandwidth_throttle, sstable_ctx_->src_info_.src_addr_, reader_arg, ctx_->replica_op_arg_.cluster_id_))) { LOG_WARN("failed to init reader", K(ret)); } else if (OB_FAIL(writer.init( &fetcher, ctx_->replica_op_arg_.key_, ctx_->get_partition()->get_storage_file_handle()))) { LOG_WARN("failed to init writer", K(ret)); } else if (OB_FAIL(writer.process(sub_task_->block_write_ctx_, sub_task_->lob_block_write_ctx_))) { LOG_WARN("failed to process writer", K(ret)); } else { const int64_t data_checksum = writer.get_data_checksum(); if (data_checksum != sstable_ctx_->data_checksum_) { ret = OB_ERR_UNEXPECTED; LOG_ERROR("logic migrate data checksum is not equal", K(ret), "dst data_checksum", data_checksum, "src data_checksum", sstable_ctx_->data_checksum_); } } } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::EN_MIGRATE_LOGIC_TASK) OB_SUCCESS; if (OB_FAIL(ret)) { LOG_ERROR("fake logic migrate fail", K(ret)); } } #endif if (OB_FAIL(ret) && NULL != ctx_) { if (ctx_->is_need_retry(ret)) { ctx_->need_rebuild_ = true; LOG_INFO("migreate copy logic task need retry", K(ret), K(*ctx_)); } ctx_->set_result_code(ret); } if (NULL != sstable_ctx_) { LOG_INFO("end ObMigrateCopyLogicTask process", "table_key", sstable_ctx_->table_key_, "dest_base_version", sstable_ctx_->dest_base_version_); } return ret; } ObMigrateFinishLogicTask::ObMigrateFinishLogicTask() : ObITask(TASK_TYPE_MIGRATE_FINISH_LOGIC), is_inited_(false), part_migrate_ctx_(NULL), sstable_ctx_() {} ObMigrateFinishLogicTask::~ObMigrateFinishLogicTask() {} int ObMigrateFinishLogicTask::init(ObPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (!part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), K(part_migrate_ctx)); } else { is_inited_ = true; part_migrate_ctx_ = &part_migrate_ctx; } return ret; } int ObMigrateFinishLogicTask::process() { int ret = OB_SUCCESS; storage::ObCreateSSTableParamWithTable create_sstable_param; ObSSTable* sstable = NULL; ObSchemaGetterGuard schema_guard; ObTableHandle handle; ObMigrateCtx* ctx = part_migrate_ctx_->ctx_; bool has_lob_column = false; const uint64_t table_id = sstable_ctx_.table_key_.table_id_; const uint64_t tenant_id = is_inner_table(table_id) ? OB_SYS_TENANT_ID : extract_tenant_id(table_id); const bool check_formal = extract_pure_id(table_id) > OB_MAX_CORE_TABLE_ID; // Avoid the problem of circular dependency LOG_INFO("start ObMigrateFinishLogicTask process", "table_key", sstable_ctx_.table_key_, "dest_base_version", sstable_ctx_.dest_base_version_); if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(sstable_ctx_.get_dest_table_key(create_sstable_param.table_key_))) { LOG_WARN("failed to get dest table key", K(ret), K(sstable_ctx_)); } else { create_sstable_param.schema_version_ = sstable_ctx_.meta_.schema_version_; create_sstable_param.logical_data_version_ = create_sstable_param.table_key_.trans_version_range_.snapshot_version_; if (OB_FAIL(MIGRATOR.get_schema_service()->get_tenant_schema_guard( tenant_id, schema_guard, sstable_ctx_.meta_.schema_version_, OB_INVALID_VERSION))) { LOG_WARN("failed to get schema guard", K(ret), K(sstable_ctx_.meta_)); } else if (check_formal && OB_FAIL(schema_guard.check_formal_guard())) { LOG_WARN("schema_guard is not formal", K(ret), K(tenant_id)); } else if (OB_FAIL( schema_guard.get_table_schema(sstable_ctx_.table_key_.table_id_, create_sstable_param.schema_))) { LOG_WARN("Fail to get table schema, ", K(ret), K(sstable_ctx_.table_key_)); } else if (OB_ISNULL(create_sstable_param.schema_)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("Unexpected null table schema", K(ret), K(sstable_ctx_.table_key_)); } else if (OB_FAIL(create_sstable_param.schema_->has_lob_column(has_lob_column, true))) { LOG_WARN("Failed to check lob column in table schema", K(ret)); } else { ObPGCreateSSTableParam param; param.with_table_param_ = &create_sstable_param; for (int64_t task_idx = 0; OB_SUCC(ret) && task_idx < sstable_ctx_.task_count_; ++task_idx) { ObMigrateLogicSSTableCtx::SubLogicTask& sub_task = sstable_ctx_.tasks_[task_idx]; if (!sub_task.block_write_ctx_.file_handle_.is_valid() && OB_FAIL(sub_task.block_write_ctx_.file_handle_.assign( part_migrate_ctx_->ctx_->partition_guard_.get_partition_group()->get_storage_file_handle()))) { LOG_WARN("fail to assign file handle", K(ret)); } else if (!sub_task.lob_block_write_ctx_.file_handle_.is_valid() && OB_FAIL(sub_task.lob_block_write_ctx_.file_handle_.assign( part_migrate_ctx_->ctx_->partition_guard_.get_partition_group()->get_storage_file_handle()))) { LOG_WARN("fail to assign file handle", K(ret)); } else if (OB_FAIL(param.data_blocks_.push_back(&sub_task.block_write_ctx_))) { LOG_WARN("fail to push back data block write ctx", K(ret)); } else if (has_lob_column && sub_task.lob_block_write_ctx_.is_valid() && !sub_task.lob_block_write_ctx_.is_empty()) { if (OB_FAIL(param.lob_blocks_.push_back(&sub_task.lob_block_write_ctx_))) { LOG_WARN("fail to push back lob block write ctx", K(ret)); } } } if (OB_SUCC(ret)) { ObIPartitionGroup* pg = nullptr; if (OB_ISNULL(pg = part_migrate_ctx_->ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_WARN("partition must not null", K(ret)); } else if (OB_FAIL(pg->create_sstable(param, handle))) { LOG_WARN("fail to create sstable", K(ret)); } else if (OB_FAIL(handle.get_sstable(sstable))) { LOG_WARN("fail to get sstable", K(ret)); } else if (OB_FAIL(part_migrate_ctx_->add_sstable(*sstable))) { LOG_WARN("failed to add sstable", K(ret)); } } } } if (OB_FAIL(ret) && NULL != ctx) { ctx->set_result_code(ret); } LOG_INFO("end ObMigrateFinishLogicTask process", "table_key", sstable_ctx_.table_key_, "dest_base_version", sstable_ctx_.dest_base_version_); return ret; } ObMigratePhysicalSSTableCtx::SubTask::SubTask() : block_info_(NULL), block_count_(0), pkey_(), allocator_(ObModIds::OB_PARTITION_MIGRATOR) {} ObMigratePhysicalSSTableCtx::SubTask::~SubTask() { int ret = OB_SUCCESS; storage::ObIPartitionGroupGuard pg_guard; blocksstable::ObStorageFile* pg_file = NULL; if (OB_FAIL(ObFileSystemUtil::get_pg_file_with_guard(pkey_, pg_guard, pg_file))) { LOG_ERROR("meta must not null", K(ret), K_(pkey)); } else { if (NULL != block_info_ && block_count_ > 0) { for (int64_t i = 0; i < block_count_; ++i) { if (nullptr != block_info_[i].full_meta_.meta_) { block_info_[i].full_meta_.meta_->~ObMacroBlockMetaV2(); block_info_[i].full_meta_.meta_ = nullptr; } if (nullptr != block_info_[i].full_meta_.schema_) { block_info_[i].full_meta_.schema_->~ObMacroBlockSchemaInfo(); block_info_[i].full_meta_.schema_ = nullptr; } if (block_info_[i].macro_block_id_.is_valid()) { if (OB_FAIL(pg_file->dec_ref(block_info_[i].macro_block_id_))) { LOG_WARN("fail to dec ref, ", K(ret)); } block_info_[i].need_copy_ = false; block_info_[i].macro_block_id_.reset(); } } } } allocator_.clear(); } ObMigratePhysicalSSTableCtx::ObMigratePhysicalSSTableCtx() : tasks_(NULL), task_count_(0), allocator_(ObModIds::OB_PARTITION_MIGRATOR), meta_(allocator_), src_info_(), sstable_info_(), is_leader_restore_(false), restore_version_(0) {} ObMigratePhysicalSSTableCtx::~ObMigratePhysicalSSTableCtx() { reset(); } bool ObMigratePhysicalSSTableCtx::is_valid() { bool valid = true; if (NULL == tasks_ || task_count_ <= 0 || !meta_.is_valid() || (!is_leader_restore_ && !src_info_.is_valid()) || !sstable_info_.is_valid()) { valid = false; LOG_WARN("ctx not valid", K(task_count_), KP(tasks_), K(meta_.is_valid()), K(src_info_.is_valid()), K(sstable_info_.is_valid())); } else { for (int64_t i = 0; valid && i < task_count_; ++i) { if (tasks_[i].block_count_ <= 0 || NULL == tasks_[i].block_info_) { valid = false; LOG_WARN("sub task not valid", K(i), K(task_count_), K(tasks_[i])); } } } return valid; } void ObMigratePhysicalSSTableCtx::reset() { if (NULL != tasks_ && task_count_ > 0) { for (int64_t task_idx = 0; task_idx < task_count_; ++task_idx) { ObMigratePhysicalSSTableCtx::SubTask& task = tasks_[task_idx]; task.~SubTask(); } } tasks_ = NULL; task_count_ = 0; meta_.reset(); allocator_.reset(); src_info_.reset(); sstable_info_.reset(); is_leader_restore_ = false; restore_version_ = 0; } int ObMigratePhysicalSSTableCtx::get_dest_table_key(ObITable::TableKey& dest_table_key) { int ret = OB_SUCCESS; if (OB_FAIL(ObDestTableKeyManager::convert_sstable_info_to_table_key(sstable_info_, dest_table_key))) { LOG_WARN("failed to conver sstable info to table key", K(ret), K(sstable_info_)); } else if (ObITable::is_major_sstable(sstable_info_.src_table_key_.table_type_)) { dest_table_key = sstable_info_.src_table_key_; if (is_leader_restore_) { dest_table_key.version_ = restore_version_; } } return ret; } ObMigrateCopyPhysicalTask::ObMigrateCopyPhysicalTask() : ObITask(TASK_TYPE_MIGRATE_COPY_PHYSICAL), is_inited_(false), ctx_(NULL), task_idx_(-1), sstable_ctx_(NULL), sub_task_(NULL), cp_fty_(NULL) {} ObMigrateCopyPhysicalTask::~ObMigrateCopyPhysicalTask() {} int ObMigrateCopyPhysicalTask::init(const int64_t task_idx, ObMigratePhysicalSSTableCtx& sstable_ctx, ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret), K(is_inited_)); } else if (task_idx < 0 || task_idx >= sstable_ctx.task_count_ || !sstable_ctx.is_valid() || !ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), K(task_idx), K(sstable_ctx.is_valid()), K(ctx.is_valid()), K(sstable_ctx), K(ctx)); } else { is_inited_ = true; ctx_ = &ctx; task_idx_ = task_idx; sstable_ctx_ = &sstable_ctx; sub_task_ = sstable_ctx.tasks_ + task_idx_; cp_fty_ = MIGRATOR.get_cp_fty(); } return ret; } int ObMigrateCopyPhysicalTask::get_macro_block_reader(obrpc::ObPartitionServiceRpcProxy* srv_rpc_proxy, common::ObInOutBandwidthThrottle* bandwidth_throttle, const ObMigrateSrcInfo& src_info, common::ObIArray& list, ObITable::TableKey& table_key, const ObRestoreInfo* restore_info, ObIPartitionMacroBlockReader*& reader) { int ret = OB_SUCCESS; UNUSED(restore_info); if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (ctx_->replica_op_arg_.is_physical_restore_leader() && share::REPLICA_RESTORE_DATA == ctx_->is_restore_) { const int64_t compatible = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.compatible_; if (OB_BACKUP_COMPATIBLE_VERSION_V1 == compatible || OB_BACKUP_COMPATIBLE_VERSION_V2 == compatible) { if (OB_FAIL(get_macro_block_restore_reader_v1( *bandwidth_throttle, list, ctx_->replica_op_arg_.phy_restore_arg_, table_key, reader))) { STORAGE_LOG(WARN, "get_macro_block_restore_reader_v1 fail", K(ret), K(table_key)); } } else if (OB_FAIL(get_macro_block_restore_reader_v2( *bandwidth_throttle, list, ctx_->replica_op_arg_.phy_restore_arg_, table_key, reader))) { STORAGE_LOG(WARN, "get_macro_block_restore_reader_v2 fail", K(ret), K(table_key)); } } else if (RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_NOT_SUPPORTED; LOG_ERROR("cannot support logical backup data anymore", K(ret), "ARG", ctx_->replica_op_arg_); } else { if (OB_FAIL(get_macro_block_ob_reader(*srv_rpc_proxy, *bandwidth_throttle, src_info, table_key, list, reader))) { STORAGE_LOG(WARN, "failed to get_macro_block_ob_reader", K(ret)); } } return ret; } int ObMigrateCopyPhysicalTask::get_macro_block_restore_reader_v1(common::ObInOutBandwidthThrottle& bandwidth_throttle, common::ObIArray& list, const share::ObPhysicalRestoreArg& restore_info, const ObITable::TableKey& table_key, ObIPartitionMacroBlockReader*& reader) { int ret = OB_SUCCESS; ObPartitionMacroBlockRestoreReaderV1* tmp_reader = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (ctx_->macro_indexs_->get_type() != ObIPhyRestoreMacroIndexStore::PHY_RESTORE_MACRO_INDEX_STORE_V1) { ret = OB_ERR_UNEXPECTED; LOG_ERROR("invlaid macro index type", K(ret), "type", ctx_->macro_indexs_->get_type()); } else if (OB_ISNULL(tmp_reader = cp_fty_->get_macro_block_restore_reader_v1())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_FAIL(tmp_reader->init(bandwidth_throttle, list, restore_info, *static_cast(ctx_->macro_indexs_), table_key))) { STORAGE_LOG(WARN, "failed to init restore reader", K(ret), K(table_key)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != reader) { cp_fty_->free(reader); reader = NULL; } } if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObMigrateCopyPhysicalTask::get_macro_block_restore_reader_v2(common::ObInOutBandwidthThrottle& bandwidth_throttle, common::ObIArray& list, const share::ObPhysicalRestoreArg& restore_info, const ObITable::TableKey& table_key, ObIPartitionMacroBlockReader*& reader) { int ret = OB_SUCCESS; ObPartitionMacroBlockRestoreReaderV2* tmp_reader = NULL; ObPhyRestoreMacroIndexStoreV2* macro_index = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_ISNULL(tmp_reader = cp_fty_->get_macro_block_restore_reader_v2())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_ISNULL(ctx_->macro_indexs_)) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "macro index should not be NULL", K(ret), KP(ctx_->macro_indexs_)); } else if (ObIPhyRestoreMacroIndexStore::PHY_RESTORE_MACRO_INDEX_STORE_V2 != ctx_->macro_indexs_->get_type()) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "macro index type is unexpected", K(ret), "type", ctx_->macro_indexs_->get_type()); } else if (FALSE_IT(macro_index = reinterpret_cast(ctx_->macro_indexs_))) { } else if (OB_FAIL(tmp_reader->init(bandwidth_throttle, list, restore_info, *macro_index, table_key))) { STORAGE_LOG(WARN, "failed to init restore reader", K(ret), K(table_key)); } else { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != reader) { cp_fty_->free(reader); reader = NULL; } } if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObMigrateCopyPhysicalTask::get_macro_block_ob_reader(obrpc::ObPartitionServiceRpcProxy& srv_rpc_proxy, common::ObInOutBandwidthThrottle& bandwidth_throttle, const ObMigrateSrcInfo& src_info, ObITable::TableKey& table_key, common::ObIArray& list, ObIPartitionMacroBlockReader*& reader) { int ret = OB_SUCCESS; ObPartitionMacroBlockObReader* tmp_reader = NULL; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!is_replica_op_valid(ctx_->replica_op_arg_.type_) || RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "unexpected type", K(ret), K(*this), K(*ctx_)); } else if (NULL == (tmp_reader = cp_fty_->get_macro_block_ob_reader())) { ret = OB_ALLOCATE_MEMORY_FAILED; STORAGE_LOG(WARN, "failed to get macro block ob reader", K(ret)); } else if (OB_FAIL(tmp_reader->init( srv_rpc_proxy, bandwidth_throttle, src_info.src_addr_, table_key, list, src_info.cluster_id_))) { STORAGE_LOG(WARN, "failed to init ob reader", K(ret)); } if (OB_SUCC(ret)) { reader = tmp_reader; tmp_reader = NULL; } if (NULL != cp_fty_) { if (OB_FAIL(ret)) { if (NULL != reader) { cp_fty_->free(reader); reader = NULL; } } if (NULL != tmp_reader) { cp_fty_->free(tmp_reader); tmp_reader = NULL; } } return ret; } int ObMigrateCopyPhysicalTask::generate_next_task(ObITask*& next_task) { int ret = OB_SUCCESS; const int64_t next_task_idx = task_idx_ + 1; ObMigrateCopyPhysicalTask* tmp_next_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (task_idx_ + 1 == sstable_ctx_->task_count_) { ret = OB_ITER_END; } else if (OB_FAIL(dag_->alloc_task(tmp_next_task))) { LOG_WARN("failed to alloc task", K(ret)); } else if (OB_FAIL(tmp_next_task->init(next_task_idx, *sstable_ctx_, *ctx_))) { LOG_WARN("failed o init next task", K(ret)); } else { next_task = tmp_next_task; } return ret; } int ObMigrateCopyPhysicalTask::process() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; common::ObArray list; ObMigrateArgMacroBlockInfo arg_info; ObMacroBlockInfoPair info; ObMacroBlocksWriteCtx copied_ctx; ObMajorMacroBlockKey key; int64_t copy_count = 0; int64_t reuse_count = 0; storage::ObIPartitionGroupGuard pg_guard; blocksstable::ObStorageFile* pg_file = NULL; DEBUG_SYNC(BEFORE_MIGRATE_COPY_BASE_DATA); if (NULL != sstable_ctx_) { LOG_INFO("start ObMigrateCopyPhysicalTask process", "table_key", sstable_ctx_->sstable_info_.src_table_key_); } if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(ObFileSystemUtil::get_pg_file_with_guard( sstable_ctx_->sstable_info_.src_table_key_.pkey_, pg_guard, pg_file))) { LOG_ERROR("fail to get pg file", K(ret), K(sstable_ctx_->sstable_info_.src_table_key_.pkey_)); } else if (!copied_ctx.file_handle_.is_valid() && OB_FAIL(copied_ctx.file_handle_.assign(pg_guard.get_partition_group()->get_storage_file_handle()))) { LOG_WARN("fail to assign file handle", K(ret)); } for (int64_t i = 0; OB_SUCC(ret) && i < sub_task_->block_count_; ++i) { arg_info.reset(); info.reset(); key.reset(); if (!sub_task_->block_info_[i].need_copy_) { ++reuse_count; } else { key.table_id_ = sstable_ctx_->sstable_info_.src_table_key_.table_id_; key.partition_id_ = sstable_ctx_->sstable_info_.src_table_key_.pkey_.get_partition_id(); key.data_version_ = sub_task_->block_info_[i].pair_.data_version_; key.data_seq_ = sub_task_->block_info_[i].pair_.data_seq_; if (sub_task_->block_info_[i].macro_block_id_.is_valid()) { ret = OB_ERR_SYS; LOG_ERROR( "need copy block info's macro block id must be 0", K(ret), K(i), "block_info", sub_task_->block_info_[i]); } else if (key.is_valid() && BACKUP_REPLICA_OP != ctx_->replica_op_arg_.type_ && OB_SUCCESS == ctx_->reuse_block_mgr_.get_reuse_macro_meta(key, info)) { sub_task_->block_info_[i].need_copy_ = false; sub_task_->block_info_[i].macro_block_id_ = info.block_id_; sub_task_->block_info_[i].full_meta_ = info.meta_; if (OB_FAIL(pg_file->inc_ref(info.block_id_))) { LOG_ERROR("fail to inc ref, ", K(ret), K(info.block_id_)); } else { ObTaskController::get().allow_next_syslog(); LOG_INFO("reuse local macro id", K(info.block_id_), K(key), "meta", info.meta_); ++reuse_count; } if (OB_FAIL(ret)) { sub_task_->block_info_[i].macro_block_id_.reset(); } // todo ():backup macro reuse } else { const int64_t max_block_count_per_task = ObMigratePrepareTask::MAX_MACRO_BLOCK_COUNT_PER_TASK; arg_info.fetch_arg_.macro_block_index_ = max_block_count_per_task * task_idx_ + i; arg_info.fetch_arg_.data_version_ = sub_task_->block_info_[i].pair_.data_version_; arg_info.fetch_arg_.data_seq_ = sub_task_->block_info_[i].pair_.data_seq_; if (OB_FAIL(list.push_back(arg_info))) { LOG_WARN("failed to add list", K(ret)); } } } } if (OB_SUCC(ret) && list.count() > 0) { if (OB_FAIL(fetch_major_block_with_retry(list, copied_ctx))) { LOG_WARN("failed to fetch major block", K(ret), K(list)); } else if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { copy_count += list.count(); } else if (list.count() != copied_ctx.get_macro_block_count()) { ret = OB_ERR_SYS; LOG_ERROR("list count not match", K(ret), K(list.count()), K(copied_ctx.get_macro_block_count()), K(list), K(copied_ctx)); } else if (copied_ctx.macro_block_list_.count() != copied_ctx.macro_block_meta_list_.count()) { ret = OB_ERR_SYS; LOG_WARN("error sys, macro block count do not match", K(ret), K(copied_ctx)); } else { int64_t sub_task_idx = 0; bool found = false; for (int64_t i = 0; OB_SUCC(ret) && i < copied_ctx.macro_block_list_.count(); ++i) { blocksstable::ObSSTablePair pair; pair.data_version_ = list.at(i).fetch_arg_.data_version_; pair.data_seq_ = list.at(i).fetch_arg_.data_seq_; found = false; while (OB_SUCC(ret) && !found && sub_task_idx < sub_task_->block_count_) { if (sub_task_->block_info_[sub_task_idx].pair_ == pair) { found = true; sub_task_->block_info_[sub_task_idx].macro_block_id_ = copied_ctx.macro_block_list_.at(i); if (OB_FAIL(copied_ctx.macro_block_meta_list_.at(i).deep_copy( sub_task_->block_info_[sub_task_idx].full_meta_, sub_task_->allocator_))) { LOG_WARN("fail to deep copy macro block meta, ", K(ret)); } else if (OB_FAIL(pg_file->inc_ref(copied_ctx.macro_block_list_.at(i)))) { sub_task_->block_info_[sub_task_idx].macro_block_id_.reset(); LOG_WARN("fail to inc ref", K(ret)); } else if (OB_FAIL(ctx_->add_major_block_id(copied_ctx.macro_block_list_.at(i)))) { LOG_WARN("fail to add major block info", K(ret)); } } ++sub_task_idx; ++copy_count; } if (!found && OB_SUCC(ret)) { ret = OB_ERR_SYS; LOG_ERROR("no macro block task found", K(ret), K(i), K(pair), K(*sub_task_), K(copied_ctx)); } } } } if (OB_SUCCESS != (tmp_ret = calc_migrate_data_statics(copy_count, reuse_count))) { LOG_WARN("failed to calc migrate data statics", K(ret)); } if (OB_FAIL(ret) && NULL != ctx_) { if (ctx_->is_need_retry(ret)) { ctx_->need_rebuild_ = true; LOG_INFO("migreate copy physical task need retry", K(ret), K(*ctx_)); } ctx_->set_result_code(ret); } if (NULL != sstable_ctx_) { LOG_INFO("end ObMigrateCopyPhysicalTask process", "table_key", sstable_ctx_->sstable_info_.src_table_key_); } return ret; } int ObMigrateCopyPhysicalTask::fetch_major_block_with_retry( ObIArray& list, ObMacroBlocksWriteCtx& copied_ctx) { int ret = OB_SUCCESS; int64_t retry_times = 0; obrpc::ObPartitionServiceRpcProxy* srv_rpc_proxy = MIGRATOR.get_svr_rpc_proxy(); common::ObInOutBandwidthThrottle* bandwidth_throttle = MIGRATOR.get_bandwidth_throttle(); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate copy physical task do not init", K(ret)); } else if (OB_ISNULL(srv_rpc_proxy) || OB_ISNULL(bandwidth_throttle)) { ret = OB_ERR_SYS; LOG_ERROR("srv_rpc_proxy or bandwidth_throttle must not null", K(ret), KP(srv_rpc_proxy), KP(bandwidth_throttle)); } else if (0 == list.count()) { LOG_INFO("no macro block need fetch", K(list.count())); } else { while (retry_times < MAX_RETRY_TIMES) { if (retry_times > 0) { LOG_INFO("retry get major block", K(retry_times)); } if (OB_FAIL(fetch_major_block(list, copied_ctx))) { STORAGE_LOG(WARN, "failed to fetch major block", K(ret), K(retry_times)); } if (OB_SUCC(ret)) { break; } if (OB_FAIL(ret)) { copied_ctx.clear(); retry_times++; usleep(OB_FETCH_MAJOR_BLOCK_RETRY_INTERVAL); } } } return ret; } int ObMigrateCopyPhysicalTask::alloc_migrate_writer( ObIPartitionMacroBlockReader* reader, ObIMigrateMacroBlockWriter*& writer) { int ret = OB_SUCCESS; writer = nullptr; if (OB_UNLIKELY(!is_inited_)) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (OB_ISNULL(reader)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid argument", K(ret), KP(reader)); } else if (OB_ISNULL(sstable_ctx_) || OB_ISNULL(ctx_)) { ret = OB_ERR_SYS; LOG_WARN("invalid inner state", K(ret), KP(sstable_ctx_), KP(ctx_)); } else { void* buf = nullptr; const uint64_t tenant_id = sstable_ctx_->sstable_info_.src_table_key_.get_tenant_id(); ObStorageFile* dest_pg_file = ctx_->get_partition()->get_pg_storage().get_storage_file(); if (OB_ISNULL(buf = ob_malloc(sizeof(ObMigrateMacroBlockWriter), ObModIds::OB_PARTITION_MIGRATE))) { ret = OB_ALLOCATE_MEMORY_FAILED; LOG_WARN("fail to alloc migrate macro block writer", K(ret)); } else { ObMigrateMacroBlockWriter* tmp_writer = new (buf) ObMigrateMacroBlockWriter; if (OB_FAIL(tmp_writer->init(reader, tenant_id, dest_pg_file))) { LOG_WARN("failed to init writer", K(ret), KP(reader), K(tenant_id), KP(dest_pg_file)); tmp_writer->~ObMigrateMacroBlockWriter(); } else { writer = tmp_writer; } } if (OB_FAIL(ret) && NULL != buf) { ob_free(buf); } } return ret; } void ObMigrateCopyPhysicalTask::free_migrate_writer(ObIMigrateMacroBlockWriter*& writer) { if (NULL != writer) { writer->~ObIMigrateMacroBlockWriter(); ob_free(writer); writer = nullptr; } } int ObMigrateCopyPhysicalTask::fetch_major_block( ObIArray& list, ObMacroBlocksWriteCtx& copied_ctx) { int ret = OB_SUCCESS; obrpc::ObPartitionServiceRpcProxy* srv_rpc_proxy = MIGRATOR.get_svr_rpc_proxy(); common::ObInOutBandwidthThrottle* bandwidth_throttle = MIGRATOR.get_bandwidth_throttle(); ObIMigrateMacroBlockWriter* writer = NULL; ObIPartitionMacroBlockReader* reader = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate copy physical task do not init", K(ret)); } else if (OB_ISNULL(srv_rpc_proxy) || OB_ISNULL(bandwidth_throttle)) { ret = OB_ERR_SYS; LOG_ERROR("srv_rpc_proxy or bandwidth_throttle must not null", K(ret), KP(srv_rpc_proxy), KP(bandwidth_throttle)); } else if (0 == list.count()) { LOG_INFO("no macro block need fetch", K(list.count())); } else { LOG_INFO("init reader", K(list)); if (OB_FAIL(get_macro_block_reader(srv_rpc_proxy, bandwidth_throttle, sstable_ctx_->src_info_, list, sstable_ctx_->sstable_info_.src_table_key_, &ctx_->restore_info_, reader))) { LOG_WARN("fail to get macro block reader", K(ret)); } else if (OB_FAIL(alloc_migrate_writer(reader, writer))) { LOG_WARN("fail to alloc migrate macro block writer", K(ret), KP(reader)); } else if (OB_FAIL(writer->process(copied_ctx))) { LOG_WARN("failed to process writer", K(ret)); } else if (list.count() != copied_ctx.get_macro_block_count()) { ret = OB_ERR_SYS; LOG_ERROR("list count not match", K(ret), K(list.count()), K(copied_ctx.get_macro_block_count()), K(list), K(copied_ctx)); } if (NULL != reader) { cp_fty_->free(reader); } if (NULL != writer) { free_migrate_writer(writer); } } return ret; } int ObMigrateCopyPhysicalTask::calc_migrate_data_statics(const int64_t copy_count, const int64_t reuse_count) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate copy physic task do not init", K(ret)); } else { const ObITable::TableKey& src_table_key = sstable_ctx_->sstable_info_.src_table_key_; ctx_->action_ = src_table_key.is_major_sstable() ? ObMigrateCtx::COPY_MAJOR : ObMigrateCtx::COPY_MINOR; ATOMIC_AAF(&ctx_->data_statics_.ready_macro_block_, sub_task_->block_count_); if (src_table_key.is_major_sstable()) { ATOMIC_AAF(&ctx_->data_statics_.major_count_, copy_count); } else if (src_table_key.is_mini_minor_sstable()) { ATOMIC_AAF(&ctx_->data_statics_.mini_minor_count_, copy_count); } else if (src_table_key.is_normal_minor_sstable()) { ATOMIC_AAF(&ctx_->data_statics_.normal_minor_count_, copy_count); } ATOMIC_AAF(&ctx_->data_statics_.reuse_count_, reuse_count); if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { LOG_WARN("failed to update_partition_migration_status", K(tmp_ret)); } } return ret; } ObMigrateFinishPhysicalTask::ObMigrateFinishPhysicalTask() : ObITask(TASK_TYPE_MIGRATE_FINISH_PHYSICAL), is_inited_(false), part_migrate_ctx_(NULL), sstable_ctx_() {} ObMigrateFinishPhysicalTask::~ObMigrateFinishPhysicalTask() {} int ObMigrateFinishPhysicalTask::init(ObIPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (!part_migrate_ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), K(part_migrate_ctx)); } else { is_inited_ = true; part_migrate_ctx_ = &part_migrate_ctx; } return ret; } int ObMigrateFinishPhysicalTask::process() { int ret = OB_SUCCESS; ObSSTable* sstable = NULL; ObTableHandle handle; ObIPartitionGroup* partition = NULL; ObMigrateCtx* ctx = part_migrate_ctx_->ctx_; ObITable::TableKey dest_table_key; blocksstable::ObMacroBlocksWriteCtx write_ctx, lob_write_ctx; // TODO(): fix it for ofs ObPGCreateSSTableParam param; ObTimeGuard tg("ObMigrateFinishPhysicalTask", 1000L * 1000L); LOG_INFO("start ObMigrateFinishPhysicalTask process", "table_key", sstable_ctx_.sstable_info_.src_table_key_); if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(sstable_ctx_.get_dest_table_key(dest_table_key))) { LOG_WARN("failed to get dest table key", K(ret), K(dest_table_key)); } else if (!dest_table_key.is_valid()) { ret = OB_ERR_SYS; LOG_WARN("table key is invalid", K(ret), K(dest_table_key)); } else if (OB_FAIL(OB_FILE_SYSTEM.init_file_ctx(STORE_FILE_MACRO_BLOCK, write_ctx.file_ctx_))) { LOG_WARN("Failed to init file ctx", K(ret)); } else if (sstable_ctx_.meta_.lob_macro_block_count_ > 0) { if (OB_FAIL(OB_FILE_SYSTEM.init_file_ctx(STORE_FILE_MACRO_BLOCK, lob_write_ctx.file_ctx_))) { LOG_WARN("Failed to init lob file ctx", K(ret)); } } if (OB_SUCC(ret)) { if (!write_ctx.file_handle_.is_valid() && OB_FAIL( write_ctx.file_handle_.assign(ctx->partition_guard_.get_partition_group()->get_storage_file_handle()))) { LOG_WARN("fail to assign file handle", K(ret)); } else if (!lob_write_ctx.file_handle_.is_valid() && OB_FAIL(lob_write_ctx.file_handle_.assign( ctx->partition_guard_.get_partition_group()->get_storage_file_handle()))) { LOG_WARN("fail to assign file handle", K(ret)); } tg.click("init"); for (int64_t task_idx = 0; OB_SUCC(ret) && task_idx < sstable_ctx_.task_count_; ++task_idx) { ObMigratePhysicalSSTableCtx::SubTask& sub_task = sstable_ctx_.tasks_[task_idx]; for (int64_t i = 0; OB_SUCC(ret) && i < sub_task.block_count_; ++i) { const MacroBlockId macro_block_id = sub_task.block_info_[i].macro_block_id_; const ObFullMacroBlockMeta& macro_meta = sub_task.block_info_[i].full_meta_; if (sstable_ctx_.meta_.lob_macro_block_count_ > 0) { if (OB_UNLIKELY(!macro_meta.is_valid())) { ret = OB_ERR_SYS; LOG_WARN("Invalid macro meta", "macro_block_id", macro_block_id, K(macro_meta), K(ret)); } else if (macro_meta.meta_->is_lob_data_block()) { if (OB_FAIL(lob_write_ctx.add_macro_block(macro_block_id, macro_meta))) { LOG_WARN("Failed to add lob macro block to write ctx", K(ret)); } } else if (OB_FAIL(write_ctx.add_macro_block(macro_block_id, macro_meta))) { LOG_WARN("Failed to add macro block to write ctx", K(ret)); } } else if (OB_FAIL(write_ctx.add_macro_block(macro_block_id, macro_meta))) { LOG_WARN("Failed to add macro block to write ctx", K(ret)); } } } tg.click("for_loop"); if (OB_SUCC(ret)) { if (OB_FAIL(param.data_blocks_.push_back(&write_ctx))) { LOG_WARN("fail to push back data block", K(ret)); } else if (sstable_ctx_.meta_.lob_macro_block_count_ > 0) { if (OB_FAIL(param.lob_blocks_.push_back(&lob_write_ctx))) { LOG_WARN("fail to push back lob macro block", K(ret)); } } } } if (OB_SUCC(ret)) { param.table_key_ = &dest_table_key; param.meta_ = &sstable_ctx_.meta_; if (OB_ISNULL(partition = ctx->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret)); } else if (OB_FAIL(partition->create_sstable(param, handle))) { LOG_WARN("fail to create sstable", K(ret)); if (OB_ENTRY_EXIST == ret) { // cover ret if (!dest_table_key.is_trans_sstable() && OB_FAIL(acquire_sstable(dest_table_key, handle))) { LOG_WARN("failed to acquire sstables", K(ret), K(dest_table_key)); } } } tg.click("create_sstable"); if (OB_SUCC(ret)) { if (OB_FAIL(handle.get_sstable(sstable))) { LOG_WARN("fail to get sstable", K(ret)); } else if (OB_ISNULL(sstable)) { ret = OB_ERR_SYS; LOG_WARN("sstable should not be null", K(ret)); } else if (ObIPartitionMigrateCtx::RECOVERY_POINT_CTX == part_migrate_ctx_->get_type()) { if (OB_FAIL(part_migrate_ctx_->add_sstable(*sstable))) { LOG_WARN("failed to add sstable", K(ret)); } } else { if (dest_table_key.is_trans_sstable()) { if (OB_FAIL(ctx->trans_table_handle_.set_table(sstable))) { LOG_WARN("failed to set trans table", K(ret), K(dest_table_key)); } } else if (OB_FAIL(part_migrate_ctx_->add_sstable(*sstable))) { LOG_WARN("failed to add sstable", K(ret)); } } tg.click("add_sstable"); } } if (OB_FAIL(ret) && NULL != ctx) { ctx->set_result_code(ret); } tg.click("set_result_code"); LOG_INFO("end ObMigrateFinishPhysicalTask process", "table_key", dest_table_key, "sub_task_cnt", sstable_ctx_.task_count_, K(tg)); return ret; } int ObMigrateFinishPhysicalTask::acquire_sstable(const ObITable::TableKey& dest_table_key, ObTableHandle& handle) { int ret = OB_SUCCESS; handle.reset(); ObSSTable* sstable = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("finish physical task do not init", K(ret)); } else if (!dest_table_key.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("acquire sstable get invalid argument", K(ret), K(dest_table_key)); } else if (OB_FAIL(ObPartitionService::get_instance().acquire_sstable(dest_table_key, handle))) { LOG_WARN("failed to acquire sstable", K(ret), K(dest_table_key)); } else if (OB_FAIL(handle.get_sstable(sstable))) { LOG_WARN("failed to get sstable", K(ret), K(handle)); } else if (OB_FAIL(check_sstable_meta(sstable_ctx_.meta_, sstable->get_meta()))) { LOG_WARN("failed to check sstable meta", K(ret), K(sstable_ctx_.meta_), "new_meta", sstable->get_meta()); } return ret; } ObMigrateFinishTask::ObMigrateFinishTask() : ObITask(TASK_TYPE_MIGRATE_FINISH), is_inited_(false), ctx_(NULL) {} ObMigrateFinishTask::~ObMigrateFinishTask() {} int ObMigrateFinishTask::init(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else { is_inited_ = true; ctx_ = &ctx; } return ret; } int ObMigrateFinishTask::process() { int ret = OB_SUCCESS; bool is_change_replica_with_data = true; bool is_add_replica_during_restore = false; ObIPartitionGroup* partition = NULL; ObPGStorage* pg_storage = NULL; DEBUG_SYNC(BEFORE_FINISH_MIGRATE_TASK); if (NULL != ctx_) { LOG_INFO("start ObMigrateFinishTask process", "pkey", ctx_->replica_op_arg_.key_, K(ctx_->replica_op_arg_.type_)); } if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_UNLIKELY(BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_)) { ret = OB_ERR_UNEXPECTED; LOG_ERROR("migrate op type is unexpected", K(ctx_->replica_op_arg_)); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret)); } else if (OB_ISNULL(pg_storage = &(partition->get_pg_storage()))) { ret = OB_ERR_SYS; LOG_ERROR("storage must not null", K(ret), KP(pg_storage)); } else if (ctx_->is_leader_restore_archive_data()) { LOG_INFO("no need check read for leader_restore_archive_data", "pkey", ctx_->replica_op_arg_.key_, "type", ctx_->replica_op_arg_.type_); } else if (OB_FAIL(ctx_->change_replica_with_data(is_change_replica_with_data))) { LOG_WARN("failed to check replica", K(ret), K(*ctx_)); } else if (is_change_replica_with_data) { LOG_INFO("no need check read for change replica", K(*ctx_)); } else if (OB_FAIL(ObMigrateUtil::merge_trans_table(*ctx_))) { LOG_WARN("failed to merge trans table", K(ret)); } else if (OB_FAIL(create_pg_partition_if_need())) { LOG_WARN("Failed to create pg partition if need", K(ret)); } else if (ADD_REPLICA_OP == ctx_->replica_op_arg_.type_ && ((ctx_->is_restore_ > REPLICA_NOT_RESTORE && ctx_->is_restore_ < REPLICA_RESTORE_MAX) || REPLICA_RESTORE_STANDBY == ctx_->is_restore_)) { is_add_replica_during_restore = true; LOG_INFO("no need check read add replica during restore", K(ctx_->is_restore_), "pkey", ctx_->replica_op_arg_.key_); } else { if (ObReplicaTypeCheck::is_replica_with_ssstore(ctx_->replica_op_arg_.dst_.get_replica_type())) { if (OB_FAIL(check_pg_partition_ready_for_read())) { LOG_WARN("Failed to check partition ready for read", K(ret)); } else if (OB_FAIL(check_pg_available_index_all_exist())) { if (ctx_->replica_op_arg_.is_physical_restore()) { ret = OB_SUCCESS; // TODO: fix restore schema, some index status should be set unavailable } LOG_WARN("failed to check_available_index_all_exist", K(ret)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(enable_replay())) { LOG_WARN("failed to enable replay", K(ret), K(ctx_->replica_op_arg_)); } } if (OB_SUCC(ret)) { int64_t retry_times = 0; while (OB_SUCC(ret) && retry_times < ObMigrateUtil::OB_MIGRATE_ONLINE_RETRY_COUNT) { if (OB_FAIL(update_split_state())) { LOG_WARN("failed to update split state", K(ret)); } retry_times++; if (OB_FAIL(ret) && retry_times < ObMigrateUtil::OB_MIGRATE_ONLINE_RETRY_COUNT) { ret = OB_SUCCESS; usleep(ObMigrateUtil::RETRY_TASK_SLEEP_INTERVAL_S); } else { break; } } } if (OB_SUCC(ret)) { if (is_change_replica_with_data || is_add_replica_during_restore || REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_ || ctx_->is_only_copy_sstable()) { LOG_INFO("no need to wait log sync", K(is_change_replica_with_data), "arg", ctx_->replica_op_arg_); } else if (OB_FAIL(wait_log_sync())) { LOG_WARN("failed to check is log sync", K(ret)); } } if (OB_SUCC(ret)) { if (OB_FAIL(update_pg_partition_report_status())) { LOG_WARN("fail to update pg partition report status", K(ret)); } } if (OB_SUCC(ret)) { const int64_t cur_ts = ObTimeUtility::current_time(); if (OB_FAIL(pg_storage->set_pg_migrate_status(OB_MIGRATE_STATUS_HOLD, cur_ts))) { LOG_WARN("failed to set migrate status to OB_MIGRATE_STATUS_HOLD", K(ret), K(*ctx_)); } else { LOG_INFO("succeed to set migrate status to OB_MIGRATE_STATUS_HOLD", K(*ctx_)); } } if (OB_SUCC(ret) && RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_ && REPLICA_RESTORE_DATA == ctx_->is_restore_) { if (OB_FAIL(partition->set_storage_info(ctx_->pg_meta_.storage_info_))) { STORAGE_LOG(WARN, "failed to set storage info", K(ret), "pg_key", ctx_->replica_op_arg_.key_, "storage_info", ctx_->pg_meta_.storage_info_); } } if (OB_SUCC(ret) && ((RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_ && (REPLICA_RESTORE_DATA == ctx_->is_restore_ || REPLICA_RESTORE_CUT_DATA == ctx_->is_restore_)) || (RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_ && REPLICA_RESTORE_STANDBY == ctx_->is_restore_) || RESTORE_FOLLOWER_REPLICA_OP == ctx_->replica_op_arg_.type_)) { if (ObReplicaTypeCheck::is_replica_with_memstore(partition->get_replica_type()) && OB_FAIL(partition->get_pg_storage().restore_mem_trans_table())) { STORAGE_LOG(WARN, "failed to restore mem trans table", K(ret)); } } if (OB_FAIL(ret) && NULL != ctx_) { if (OB_LOG_NOT_SYNC == ret) { if (ADD_REPLICA_OP == ctx_->replica_op_arg_.type_ || MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_ || FAST_MIGRATE_REPLICA_OP == ctx_->replica_op_arg_.type_ || REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { ctx_->need_rebuild_ = true; } } ctx_->set_result_code(ret); } if (NULL != ctx_) { LOG_INFO("end ObMigrateFinishTask process", "pkey", ctx_->replica_op_arg_.key_, K(ctx_->replica_op_arg_.type_)); } #ifdef ERRSIM if (OB_SUCC(ret)) { ret = E(EventTable::AFTER_MIGRATE_FINISH_TASK) OB_SUCCESS; if (OB_FAIL(ret)) { STORAGE_LOG(ERROR, "fake AFTER_MIGRATE_FINISH_TASK", K(ret)); } } #endif return ret; } int ObMigrateFinishTask::lock_pg_owner(common::ObMySQLTransaction& trans, ObIPartitionGroup& pg, const common::ObAddr& data_src_addr, const int64_t epoch_number) { int ret = OB_SUCCESS; if (OB_UNLIKELY(!pg.is_valid() || !data_src_addr.is_valid() || OB_INVALID_VERSION == epoch_number)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid argument", K(ret), K(pg), K(data_src_addr), K(epoch_number)); } else { const ObPGKey& pg_key = pg.get_partition_key(); ObSqlString sql; SMART_VAR(ObISQLClient::ReadResult, result) { sqlclient::ObMySQLResult* mysql_result = nullptr; char ip_str[32] = {0}; const char* sql_template = " SELECT is_owner, epoch_num " " FROM __all_tenant_meta_table " " WHERE tenant_id=%lu AND table_id=%lu AND partition_id=%ld " " AND svr_ip='%s' AND svr_port=%d AND epoch_num=%ld and is_owner=%d" " FOR UPDATE; "; if (OB_UNLIKELY(!data_src_addr.ip_to_string(ip_str, sizeof(ip_str)))) { ret = OB_ERR_UNEXPECTED; LOG_WARN("fail to get ip string", K(ret), K(data_src_addr)); } else if (OB_FAIL(sql.append_fmt(sql_template, pg_key.get_tenant_id(), pg_key.get_table_id(), pg_key.get_partition_id(), ip_str, data_src_addr.get_port(), epoch_number, true /*is_owner*/))) { LOG_WARN("fail to generate sql string", K(ret), K(pg_key)); } else if (OB_FAIL(trans.read(result, pg_key.get_tenant_id(), sql.ptr()))) { LOG_WARN("fail to execute read sql", K(ret), K(sql), K(pg_key)); } else if (OB_ISNULL(mysql_result = result.get_result())) { // empty result, owner maybe changed or src server reboot ret = OB_EMPTY_RESULT; LOG_WARN("query get empty result", K(ret), K(data_src_addr), K(pg_key), K(epoch_number)); } } } return ret; } int ObMigrateFinishTask::enable_replay() { int ret = OB_SUCCESS; if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(enable_replay_for_rebuild())) { LOG_WARN("failed to enable_replay_for_rebuild", K(ret), "arg", ctx_->replica_op_arg_); } } else { if (ctx_->create_new_pg_) { if (OB_FAIL(ObMigrateUtil::enable_replay_with_new_partition(*ctx_))) { LOG_WARN("failed to enable replay with new partition", K(ret), KPC(ctx_)); } } else if (ctx_->get_partition()->get_pg_storage().is_paused()) { if (OB_FAIL(ObMigrateUtil::enable_replay_with_old_partition(*ctx_))) { LOG_WARN("failed to enable replay with old partition", K(ret), KPC(ctx_)); } } } return ret; } int ObMigrateFinishTask::create_pg_partition_if_need() { int ret = OB_SUCCESS; ObIPartitionGroup* pg = NULL; const bool in_slog_trans = false; const int64_t max_kept_major_version_number = 0; LOG_INFO("start create_pg_partition_if_need"); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(pg = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_WARN("pg should not be NULL", K(ret), KP(pg)); } else if (ctx_->is_copy_index()) { // no need to replace store map, just skip it. LOG_INFO("replica op tyoe is no need batch replace sstable, skip it"); } else { const ObSavedStorageInfoV2& saved_storage_info = ctx_->pg_meta_.storage_info_; if (OB_FAIL(pg->get_pg_storage().batch_replace_store_map(ctx_->part_ctx_array_, saved_storage_info.get_data_info().get_schema_version(), ctx_->is_restore_, ctx_->old_trans_table_seq_))) { LOG_WARN("failed to batch replace store map", K(ret)); } } return ret; } int ObMigrateFinishTask::check_pg_partition_ready_for_read() { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not init", K(ret)); } else if (OB_FAIL(check_partition_ready_for_read_in_remote())) { LOG_WARN("failed to check partition ready for read in remote", K(ret)); } else if (OB_FAIL(check_partition_ready_for_read_out_remote())) { LOG_WARN("failed to check partition ready for read out remote", K(ret)); } return ret; } int ObMigrateFinishTask::check_partition_ready_for_read_in_remote() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPGPartition* pg_partition = NULL; ObIPartitionStorage* storage = NULL; ObPartitionArray pkeys; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret)); } else if (OB_FAIL(partition->get_all_pg_partition_keys(pkeys, true /*include_trans_table*/))) { LOG_WARN("fail to get all pg partition keys", K(ret)); } else if (pkeys.count() < ctx_->part_ctx_array_.count()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("can not remove partition during migration", K(ret), K(pkeys.count()), K(ctx_->part_ctx_array_.count())); } else { for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { ObPGPartitionGuard guard; storage = NULL; const ObMigratePartitionInfo& copy_info = ctx_->part_ctx_array_.at(i).copy_info_; const ObPartitionKey& pkey = copy_info.meta_.pkey_; if (pkey.is_trans_table()) { continue; } else if (OB_FAIL(partition->get_pg_partition(pkey, guard))) { LOG_ERROR("failed to get pg partition", K(ret), K(pkey)); } else if (OB_ISNULL(pg_partition = guard.get_pg_partition())) { ret = OB_ERR_SYS; LOG_ERROR("pg partition should not be NULL", K(ret), KP(pg_partition)); } else if (OB_ISNULL(storage = pg_partition->get_storage())) { ret = OB_ERR_SYS; LOG_ERROR("storage must not null", K(ret)); } else if (OB_FAIL(check_partition_ready_for_read(copy_info, storage))) { LOG_WARN("failed to check partition ready for read", K(ret), K(copy_info)); } } } return ret; } int ObMigrateFinishTask::check_partition_ready_for_read( const ObMigratePartitionInfo& copy_info, ObIPartitionStorage* storage) { int ret = OB_SUCCESS; ObTablesHandle handle; bool is_ready_for_read = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (!copy_info.is_valid() || OB_ISNULL(storage)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("check partition ready for read get invalid argument", K(ret), K(copy_info), KP(storage)); } else { ObPartitionStore& store = static_cast(storage)->get_partition_store(); for (int64_t i = 0; OB_SUCC(ret) && i < copy_info.table_infos_.count(); ++i) { const ObMigrateTableInfo& info = copy_info.table_infos_[i]; handle.reset(); if (OB_FAIL(store.get_effective_tables(info.table_id_, handle, is_ready_for_read))) { LOG_WARN("failed to check ready for read", K(ret), K(info.table_id_)); } else if (info.ready_for_read_ && !is_ready_for_read) { const uint64_t tenant_id = extract_tenant_id(info.table_id_); share::schema::ObSchemaGetterGuard schema_guard; share::schema::ObMultiVersionSchemaService& schema_service = share::schema::ObMultiVersionSchemaService::get_instance(); bool is_exist = true; if (OB_FAIL(schema_service.get_tenant_schema_guard(tenant_id, schema_guard))) { LOG_WARN("failed to get schema guard", K(ret), K(tenant_id)); } else if (OB_FAIL(schema_guard.check_table_exist(info.table_id_, is_exist))) { LOG_WARN("failed to check table exist", K(ret)); } else if (!is_exist) { ret = OB_TABLE_IS_DELETED; LOG_WARN("table is not exist, can not check is ready for read", K(ret), K(info)); } else { ret = OB_ERR_SYS; LOG_ERROR("src is ready for read, but dest is not, migrate failed", K(ret), K(info), K(handle)); } } } } return ret; } int ObMigrateFinishTask::check_partition_ready_for_read_out_remote() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPGPartition* pg_partition = NULL; ObIPartitionStorage* storage = NULL; ObTablesHandle handle; ObPartitionArray pkeys; ObHashSet remote_pkeys; int hash_ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret)); } else if (OB_FAIL(partition->get_pg_storage().get_all_pg_partition_keys_with_lock( pkeys, true /*include_trans_table*/))) { LOG_WARN("fail to get all pg partition keys", K(ret)); } else { const int64_t OB_MAX_BUCKET = std::max(pkeys.count(), ctx_->part_ctx_array_.count()); if (0 == OB_MAX_BUCKET) { // do nothing } else if (pkeys.count() < ctx_->part_ctx_array_.count()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("can not remove partition during migration", K(ret), K(pkeys.count()), K(ctx_->part_ctx_array_.count())); } else if (OB_FAIL(remote_pkeys.create(OB_MAX_BUCKET))) { LOG_WARN("failed to create remote pkeys set", K(ret), K(ctx_->part_ctx_array_.count())); } else { for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { const ObMigratePartitionInfo& copy_info = ctx_->part_ctx_array_.at(i).copy_info_; const ObPartitionKey& pkey = copy_info.meta_.pkey_; if (OB_FAIL(remote_pkeys.set_refactored(pkey))) { LOG_WARN("failed to set pkey into remote pkey set", K(ret), K(pkey)); } } for (int64_t i = 0; OB_SUCC(ret) && i < pkeys.count(); ++i) { const ObPartitionKey& pkey = pkeys.at(i); hash_ret = remote_pkeys.exist_refactored(pkey); if (OB_HASH_EXIST == hash_ret) { // do nothing } else if (OB_HASH_NOT_EXIST == hash_ret) { ObPGPartitionGuard guard; storage = NULL; if (pkey.is_trans_table()) { continue; } else if (OB_FAIL(partition->get_pg_partition(pkey, guard))) { LOG_ERROR("failed to get pg partition", K(ret), K(pkey)); } else if (OB_ISNULL(pg_partition = guard.get_pg_partition())) { ret = OB_ERR_SYS; LOG_ERROR("pg partition should not be NULL", K(ret), KP(pg_partition)); } else if (OB_ISNULL(storage = pg_partition->get_storage())) { ret = OB_ERR_SYS; LOG_ERROR("storage must not null", K(ret)); } else if (OB_FAIL(check_partition_ready_for_read(storage))) { LOG_WARN("failed to check partition out remote ready for read", K(ret), K(pkey)); } } else { ret = hash_ret; LOG_WARN("failed to check pkey into set", K(ret), K(pkey)); } } } } return ret; } int ObMigrateFinishTask::check_partition_ready_for_read(ObIPartitionStorage* storage) { int ret = OB_SUCCESS; ObTablesHandle handle; bool is_ready_for_read = false; ObArray table_ids; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(storage)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("check partition ready for read get invalid argument", K(ret), KP(storage)); } else { ObPartitionStore& store = static_cast(storage)->get_partition_store(); if (OB_FAIL(store.get_all_table_ids(table_ids))) { LOG_WARN("failed to get all table ids", K(ret), K(storage->get_partition_key())); } else { for (int64_t i = 0; OB_SUCC(ret) && i < table_ids.count(); ++i) { const uint64_t table_id = table_ids.at(i); handle.reset(); if (OB_FAIL(store.get_effective_tables(table_id, handle, is_ready_for_read))) { LOG_WARN("failed to check ready for read", K(ret), K(table_id)); } else if (!is_ready_for_read) { ret = OB_VERSION_RANGE_NOT_CONTINUES; LOG_WARN("partition which not in remote has uncontinues tables", K(ret), K(table_id), K(handle), K(storage->get_partition_key())); } } } } return ret; } int ObMigrateFinishTask::check_pg_available_index_all_exist() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPGPartition* pg_partition = NULL; ObPartitionStorage* storage = NULL; ObPartitionArray pkeys; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret)); } else if (OB_FAIL(partition->get_all_pg_partition_keys(pkeys))) { LOG_WARN("failed to get all pg partition keys", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < pkeys.count(); ++i) { ObPGPartitionGuard guard; storage = NULL; const ObPartitionKey& pkey = pkeys.at(i); if (OB_FAIL(partition->get_pg_partition(pkey, guard))) { LOG_ERROR("failed to get pg partition", K(ret), K(pkey)); } else if (OB_ISNULL(pg_partition = guard.get_pg_partition())) { ret = OB_ERR_SYS; LOG_WARN("pg partition should not be NULL", K(ret), KP(pg_partition)); } else if (OB_ISNULL(storage = static_cast(pg_partition->get_storage()))) { ret = OB_ERR_SYS; LOG_WARN("fail to get partition storage", K(ret), KP(storage)); } else if (OB_FAIL(check_available_index_all_exist(pkey, storage))) { LOG_WARN("fail to check available index all exist", K(ret)); } } } return ret; } int ObMigrateFinishTask::check_available_index_all_exist(const ObPartitionKey& pkey, ObPartitionStorage* storage) { int ret = OB_SUCCESS; ObSchemaGetterGuard schema_guard; ObArray index_stats; ObTableHandle handle; const uint64_t tenant_id = is_inner_table(pkey.get_table_id()) ? OB_SYS_TENANT_ID : pkey.get_tenant_id(); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(storage)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("check available index get invalid argument", K(ret), KP(storage)); } else if (OB_FAIL(MIGRATOR.get_schema_service()->get_tenant_full_schema_guard(tenant_id, schema_guard))) { LOG_WARN("fail to get schema schema guard", K(ret)); } else if (OB_FAIL(schema_guard.get_index_status(pkey.get_table_id(), false /*with global index*/, index_stats))) { LOG_WARN("fail to get index status", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < index_stats.count(); ++i) { if (is_available_index_status(index_stats.at(i).index_status_)) { if (OB_FAIL(storage->get_partition_store().get_last_major_sstable(index_stats.at(i).index_id_, handle))) { LOG_WARN("failed to get last major sstable for index", K(ret), "index_id", index_stats.at(i).index_id_); } else if (NULL == handle.get_table()) { ret = OB_ENTRY_NOT_EXIST; LOG_WARN("has no major sstable for index", K(ret), "index_id", index_stats.at(i).index_id_); } } } // TODO(): should rebuild migrate } return ret; } int ObMigrateFinishTask::update_pg_partition_report_status() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; const int64_t version = ctx_->pg_meta_.report_status_.data_version_ > 0 ? ctx_->pg_meta_.report_status_.data_version_ : ObPartitionScheduler::get_instance().get_frozen_version(); const common::ObVersion report_version(version); bool need_report = false; bool check_finished = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_ISNULL(partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_ERROR("partition must not null", K(ret)); } else if (OB_ISNULL(GCTX.ob_service_)) { // skip it } else if (OB_SUCCESS != (tmp_ret = partition->get_pg_storage().try_update_report_status( *GCTX.ob_service_, report_version, check_finished, need_report))) { LOG_WARN("failed to try_update_report_status", K(ret), K(partition->get_partition_key())); } return ret; } int64_t ObMigratePhysicalSSTableCtx::SubTask::to_string(char* buf, const int64_t buf_len) const { int64_t pos = 0; if (OB_ISNULL(buf) || buf_len <= 0) { // do nothing } else { J_OBJ_START(); J_KV(KP(this), K_(block_count)); J_COMMA(); J_ARRAY_START(); for (int64_t i = 0; i < block_count_; ++i) { ObMigrateMacroBlockInfo& info = block_info_[i]; J_OBJ_START(); J_KV(K(i), K(info)); J_OBJ_END(); J_COMMA(); } J_ARRAY_END(); J_OBJ_END(); } return pos; } int64_t ObMigratePhysicalSSTableCtx::to_string(char* buf, const int64_t buf_len) const { int64_t pos = 0; if (OB_ISNULL(buf) || buf_len <= 0) { // do nothing } else { J_OBJ_START(); J_KV(KP(this), K_(src_info), K_(sstable_info), K_(task_count), KP(tasks_), K_(meta)); J_COMMA(); J_ARRAY_START(); for (int64_t i = 0; i < task_count_; ++i) { SubTask& task = tasks_[i]; J_OBJ_START(); J_KV(K(i), K(task)); J_OBJ_END(); J_COMMA(); } J_ARRAY_END(); J_OBJ_END(); } return pos; } int ObMigrateFinishTask::wait_log_sync(uint64_t max_clog_id /* = OB_INVALID_TIMESTAMP*/) { int ret = OB_SUCCESS; bool is_log_sync = false; uint64_t last_confirmed_log_id = 0; bool is_cancel = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret), "pkey", ctx_->replica_op_arg_.key_); } else if (0 == ctx_->wait_replay_start_ts_) { ctx_->wait_replay_start_ts_ = ObTimeUtility::current_time(); LOG_INFO("start wait_log_sync", "pkey", ctx_->replica_op_arg_.key_); } while (OB_SUCC(ret) && !is_log_sync) { if (ctx_->group_task_->has_error()) { ret = OB_CANCELED; STORAGE_LOG(WARN, "group task has error, cancel subtask", K(ret)); } else if (OB_FAIL(SYS_TASK_STATUS_MGR.is_task_cancel(get_dag()->get_dag_id(), is_cancel))) { STORAGE_LOG(ERROR, "failed to check is task canceled", K(ret), K(*this)); } else if (is_cancel) { ret = OB_CANCELED; STORAGE_LOG(WARN, "task is cancelled", K(ret), K(*this)); } else if (OB_FAIL(MIGRATOR.get_partition_service()->is_log_sync( ctx_->replica_op_arg_.key_, is_log_sync, last_confirmed_log_id))) { LOG_WARN("failed to check is log sync", K(ret)); } else if (OB_INVALID_TIMESTAMP != max_clog_id) { is_log_sync = last_confirmed_log_id >= max_clog_id; } if (OB_FAIL(ret)) { } else if (is_log_sync) { const int64_t cost_ts = ObTimeUtility::current_time() - ctx_->wait_replay_start_ts_; LOG_INFO( "log is sync, stop wait_log_sync", "pkey", ctx_->replica_op_arg_.key_, K(cost_ts), K(last_confirmed_log_id)); } else { if (REACH_TIME_INTERVAL(60 * 1000 * 1000)) { LOG_INFO("log is not sync, retry next loop", "pkey", ctx_->replica_op_arg_.key_, K(last_confirmed_log_id)); } usleep(OB_CHECK_LOG_SYNC_INTERVAL); // If a log is not played backup for more than 30 minutes, it will be timeout during in sync const int64_t CLOG_IN_SYNC_DELAY_TIMEOUT = 30 * 60 * 1000 * 1000; if ((ctx_->last_confirmed_log_id_ == last_confirmed_log_id) && (OB_INVALID_TIMESTAMP != ctx_->last_confirmed_log_ts_)) { if (ObTimeUtility::current_time() - ctx_->last_confirmed_log_ts_ >= CLOG_IN_SYNC_DELAY_TIMEOUT) { ctx_->result_ = OB_TIMEOUT; ret = OB_TIMEOUT; STORAGE_LOG( WARN, "failed to check is partition log replay sync. timeout, stop migrate task", K(ret), K(*ctx_)); } } else { ctx_->last_confirmed_log_id_ = last_confirmed_log_id; ctx_->last_confirmed_log_ts_ = ObTimeUtility::current_time(); } } } if (OB_SUCC(ret) && !is_log_sync) { const int64_t cost_ts = ObTimeUtility::current_time() - ctx_->wait_replay_start_ts_; ret = OB_LOG_NOT_SYNC; LOG_WARN( "log is not sync", "pkey", ctx_->replica_op_arg_.key_, "pkey", ctx_->replica_op_arg_.key_, K(cost_ts), K(ret)); } return ret; } int ObMigrateFinishTask::enable_replay_for_rebuild() { int ret = OB_SUCCESS; LOG_INFO("start enable_replay_for_rebuild"); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret), "pkey", ctx_->replica_op_arg_.key_); } else if (REBUILD_REPLICA_OP != ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; LOG_ERROR("cannot enable_replay_for_rebuild for other type", K(ret), K(*ctx_)); } else { ctx_->need_online_for_rebuild_ = false; if (OB_FAIL(ObMigrateUtil::enable_replay_with_old_partition(*ctx_))) { LOG_WARN("failed to enable_replay_with_old_partition", K(ret), K(*ctx_)); } } return ret; } int ObMigrateFinishTask::update_split_state() { int ret = OB_SUCCESS; ObPartitionGroupMeta& meta = ctx_->pg_meta_; ObBaseStorageInfo& clog_info = meta.storage_info_.get_clog_info(); ObDataStorageInfo& data_info = meta.storage_info_.get_data_info(); const ObPartitionSplitInfo& split_info = meta.split_info_; int64_t split_state = meta.saved_split_state_; // 221,222 may set spit_status = -1, and it has bug during upgrade to 224. Now set split_status = 1. if (-1 == split_state) { split_state = 1; LOG_WARN("split state was -1, should be rewritten to 1", K(meta.pg_key_)); } ObIPartitionGroup* partition = NULL; if (OB_ISNULL(partition = ctx_->get_partition())) { ret = OB_ERR_SYS; LOG_ERROR("cannot use old partition", K(ret)); } else if (need_copy_split_state(ctx_->replica_op_arg_.type_)) { if (split_info.get_src_partition() == meta.pg_key_) { const int64_t source_log_id = split_info.get_source_log_id(); const int64_t last_replay_log_id = std::min(clog_info.get_last_replay_log_id(), data_info.get_last_replay_log_id()); if (last_replay_log_id >= source_log_id && OB_INVALID_ID != source_log_id) { if (OB_FAIL( partition->shutdown(split_info.get_split_version(), source_log_id, split_info.get_schema_version()))) { LOG_WARN("failed to shutdown", K(ret), K(split_info)); } else if (OB_FAIL(partition->push_reference_tables( split_info.get_dest_partitions(), split_info.get_split_version()))) { LOG_WARN("failed to push reference tables", K(split_info)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(partition->save_split_state(split_state))) { LOG_WARN("failed to save split info and state", K(ret), K(split_state)); } } } return ret; } ObGroupMigrateDag::ObGroupMigrateDag() : ObIDag(ObIDag::DAG_TYPE_GROUP_MIGRATE, ObIDag::DAG_PRIO_GROUP_MIGRATE), is_inited_(false), group_task_(NULL), tenant_id_(0), partition_service_(NULL) {} ObGroupMigrateDag::~ObGroupMigrateDag() { int tmp_ret = OB_SUCCESS; if (NULL != group_task_) { ObArray report_list; bool is_batch_mode = false; bool could_report = false; if (OB_SUCCESS != (tmp_ret = group_task_->fill_report_list(is_batch_mode, report_list))) { LOG_WARN("failed to fill report list", K(tmp_ret), K_(tenant_id)); } else { could_report = true; } if (OB_SUCCESS != (tmp_ret = ObPartGroupMigrator::get_instance().remove_finish_task(group_task_))) { LOG_WARN("failed to remove finish task", K(tmp_ret), K_(tenant_id)); } group_task_ = NULL; if (could_report) { if (OB_SUCCESS != (tmp_ret = report_result(is_batch_mode, report_list))) { STORAGE_LOG(WARN, "failed to report result", K(tmp_ret), K_(tenant_id), K(is_batch_mode)); } } } } bool ObGroupMigrateDag::operator==(const ObIDag& other) const { bool is_same = true; if (this == &other) { // same } else if (get_type() != other.get_type()) { is_same = false; } else { const ObGroupMigrateDag& other_dag = static_cast(other); is_same = group_task_ == other_dag.group_task_; } return is_same; } int64_t ObGroupMigrateDag::hash() const { int64_t ptr = reinterpret_cast(group_task_); return common::murmurhash(&ptr, sizeof(ptr), 0); } int ObGroupMigrateDag::init(ObPartGroupTask* group_task, storage::ObPartitionService* partition_service) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (OB_ISNULL(group_task) || OB_ISNULL(partition_service)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(group_task), KP(partition_service)); } else if (OB_FAIL(set_dag_id(group_task->get_task_id()))) { LOG_WARN("failed to set dag id", K(ret)); } else { tenant_id_ = group_task->get_tenant_id(); group_task_ = group_task; partition_service_ = partition_service; is_inited_ = true; } return ret; } void ObGroupMigrateDag::clear() { is_inited_ = false; group_task_ = NULL; tenant_id_ = 0; } int ObGroupMigrateDag::fill_comment(char* buf, const int64_t buf_len) const { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (NULL == buf || NULL == group_task_) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), KP(group_task_), K(buf)); } else if (group_task_->get_task_list().count() <= 0) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(*group_task_)); } else { const ObReplicaOpArg& arg = group_task_->get_task_list().at(0).arg_; int n = snprintf(buf, buf_len, "group partition migration executor: group_task_id=%s partition_count=%ld partition_id=%ld " "key=%s, op_type=%s, src=%s, dest=%s,", to_cstring(group_task_->get_task_id()), group_task_->get_task_list().count(), arg.key_.get_partition_id(), to_cstring(arg.key_), arg.get_replica_op_type_str(), to_cstring(arg.data_src_.get_server()), to_cstring(arg.dst_.get_server())); if (n < 0 || n >= buf_len) { tmp_ret = OB_BUF_NOT_ENOUGH; STORAGE_LOG(WARN, "buf not enough, comment is truncated", K(tmp_ret), K(n)); } } return ret; } ObGroupMigrateExecuteTask::ObGroupMigrateExecuteTask() : ObITask(TASK_TYPE_GROUP_MIGRATE), is_inited_(false), group_task_(NULL) {} ObGroupMigrateExecuteTask::~ObGroupMigrateExecuteTask() {} int ObGroupMigrateExecuteTask::init(ObPartGroupTask* group_task) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (OB_ISNULL(group_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("invalid args", K(ret), KP(group_task)); } else { is_inited_ = true; group_task_ = group_task; } return ret; } int ObGroupMigrateExecuteTask::process() { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(group_task_->do_task())) { LOG_WARN("failed to do task", K(ret)); } return ret; } int ObMigrateGetLeaderUtil::get_leader( const common::ObPartitionKey& pkey, ObMigrateSrcInfo& leader_info, const bool force_renew) { int ret = OB_SUCCESS; const int64_t cluster_id = GCONF.cluster_id; leader_info.reset(); if (!pkey.is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "pkey is invalid", K(ret), K(pkey)); } else { share::ObIPartitionLocationCache* location_cache = NULL; if (NULL == (location_cache = ObPartitionService::get_instance().get_location_cache())) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "location cache must not null", K(ret)); } else if (OB_FAIL(location_cache->get_leader_by_election(pkey, leader_info.src_addr_, force_renew))) { STORAGE_LOG(WARN, "get leader address failed", K(ret), K(pkey)); } else { leader_info.cluster_id_ = cluster_id; } } return ret; } int ObMigrateGetLeaderUtil::get_clog_parent(clog::ObIPartitionLogService& log_service, ObMigrateSrcInfo& parent_info) { int ret = OB_SUCCESS; parent_info.reset(); ObAddr parent_src; int64_t cluster_id = -1; if (OB_FAIL(log_service.get_clog_parent(parent_src, cluster_id))) { STORAGE_LOG(WARN, "get parent addr failed", K(ret)); } else { parent_info.src_addr_ = parent_src; parent_info.cluster_id_ = cluster_id; } return ret; } int ObMigrateTaskGeneratorTask::schedule_migrate_tasks() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; bool need_schedule = false; ObArray last_task_array; ObITask* last_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(need_generate_sstable_migrate_tasks(need_schedule))) { LOG_WARN("failed to check generate sstable migrate task", K(ret)); } else if (!need_schedule) { LOG_INFO("no need_generate_sstable_migrate_tasks"); } else if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(generate_pg_rebuild_tasks(last_task_array))) { LOG_WARN("failed to generate rebuild task", K(ret)); } } else if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(generate_pg_backup_tasks(last_task_array))) { LOG_WARN("failed to generate backup task", K(ret)); } } else if (VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(generate_pg_validate_tasks(last_task_array))) { LOG_WARN("failed to generate validate task", K(ret)); } } else if (BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(generate_pg_backup_backupset_tasks(last_task_array))) { LOG_WARN("failed to generate backup backupset task", K(ret)); } } else if (BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(generate_pg_backup_archive_log_tasks(last_task_array))) { LOG_WARN("failed to generate backup backupset task", K(ret)); } } else if (OB_FAIL(generate_pg_migrate_tasks(last_task_array))) { LOG_WARN("failed to generate_pg_migrate_tasks", K(ret)); } if (OB_SUCC(ret)) { if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need finish task for backup"); } else if (VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need finish task for validate"); } else if (BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need finish task for backup backupset"); } else if (BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need finish task for backup archivelog"); } else if (OB_FAIL(generate_finish_migrate_task(last_task_array, last_task))) { LOG_WARN("failed to generate prepare migrate task", K(ret)); } else if (RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(generate_restore_tailored_task(last_task))) { LOG_WARN("failed to generate restore tailored task", K(ret)); } } } if (OB_SUCC(ret)) { if (OB_SUCCESS != (tmp_ret = ctx_->update_partition_migration_status())) { LOG_WARN("failed to update partition migration status", K(tmp_ret)); } } // Whether success or failure, we need to clean up the reference count of macro block to prevent leakage if (NULL != ctx_) { ctx_->free_old_macro_block(); } return ret; } int ObMigrateTaskGeneratorTask::update_multi_version_start() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObPGStorage* pg_storage = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (OB_ISNULL(partition = ctx_->get_partition())) { ret = OB_ERR_SYS; LOG_WARN("can not update multi version when pg is null", K(ret), KP(partition)); } else if (OB_ISNULL(pg_storage = &(partition->get_pg_storage()))) { ret = OB_ERR_SYS; LOG_WARN("pg storage should not be NULL", K(ret), KP(pg_storage)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { const ObPartitionMigrateCtx& pctx = ctx_->part_ctx_array_.at(i); const ObMigratePartitionInfo& copy_info = ctx_->part_ctx_array_.at(i).copy_info_; const ObPartitionKey& pkey = copy_info.meta_.pkey_; if (pctx.is_partition_exist_) { if (OB_FAIL(pg_storage->update_multi_version_start(pkey, copy_info.meta_.multi_version_start_))) { LOG_WARN("fail to update multi version start", K(ret), K(pkey), K(copy_info)); } } } } return ret; } int ObMigrateTaskGeneratorTask::deal_with_new_partition() { int ret = OB_SUCCESS; if (OB_ISNULL(ctx_->group_task_)) { ret = OB_ERR_UNEXPECTED; LOG_ERROR("group task should not be null here", K(ret)); } return ret; } int ObMigrateTaskGeneratorTask::generate_restore_tailored_task(share::ObITask* last_task) { int ret = OB_SUCCESS; ObRestoreTailoredPrepareTask* prepare_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate post prepare task do not init", K(ret)); } else if (OB_ISNULL(last_task)) { ret = OB_INVALID_ARGUMENT; LOG_WARN("generate restore tailored task get invalid argument", K(ret), KP(last_task)); } else if (RESTORE_REPLICA_OP != ctx_->replica_op_arg_.type_ && RESTORE_STANDBY_OP != ctx_->replica_op_arg_.type_) { ret = OB_ERR_UNEXPECTED; LOG_WARN("failed to generate restore tailored task", K(ret), K(*ctx_)); } else if (OB_FAIL(dag_->alloc_task(prepare_task))) { LOG_WARN("failed to allock prepare task", K(ret)); } else if (OB_FAIL(prepare_task->init())) { LOG_WARN("failed to init prepare task", K(ret)); } else if (OB_FAIL(last_task->add_child(*prepare_task))) { LOG_WARN("failed to add prepare task to child task", K(ret)); } else if (OB_FAIL(dag_->add_task(*prepare_task))) { LOG_WARN("failed to add prepare task to dag", K(ret)); } return ret; } int ObMigrateUtil::wait_trans_table_merge_finish(ObMigrateCtx& ctx) { int ret = OB_SUCCESS; const int64_t WAIT_TIMEOUT = 300L * 1000L * 1000L; // 5min const int64_t SLEEP_INTERVAL = 100L * 1000L; // 100ms ObTransTableMergeDag fake_dag; if (OB_FAIL(fake_dag.init(ctx.pg_meta_.pg_key_))) { LOG_WARN("failed to init dag", K(ret), K(ctx.pg_meta_.pg_key_)); } else if (OB_FAIL(ObDagScheduler::get_instance().cancel_dag(&fake_dag))) { LOG_WARN("failed to cancel dag", K(ret), K(fake_dag)); } else { bool exist = true; const int64_t start_ts = ObTimeUtility::current_time(); do { if (OB_FAIL(ObDagScheduler::get_instance().check_dag_exist(&fake_dag, exist))) { LOG_WARN("failed to check dag exist", K(ret), K(fake_dag)); } else if (exist) { if (ObTimeUtility::current_time() - start_ts > WAIT_TIMEOUT) { ret = OB_WAIT_TRANS_TABLE_MERGE_TIMEOUT; LOG_WARN("wait trans table merge finish timeout", K(ret), K(fake_dag)); } else { usleep(SLEEP_INTERVAL); } } } while (OB_SUCC(ret) && exist); } return ret; } ObRestoreTailoredPrepareTask::ObRestoreTailoredPrepareTask() : ObITask(TASK_TYPE_RESTORE_TAILORED_PREPARE), is_inited_(false), ctx_(NULL) {} ObRestoreTailoredPrepareTask::~ObRestoreTailoredPrepareTask() {} int ObRestoreTailoredPrepareTask::init() { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_WARN("restore tailored prepare task init twice", K(ret)); } else if (ObIDag::DAG_TYPE_MIGRATE != dag_->get_type()) { ret = OB_ERR_SYS; LOG_ERROR("dag type not match", K(ret), K(*dag_)); } else { ctx_ = static_cast(dag_)->get_ctx(); is_inited_ = true; } return ret; } int ObRestoreTailoredPrepareTask::process() { int ret = OB_SUCCESS; ObRestoreTailoredFinishTask* finish_task = NULL; ObIPartitionGroup* partition_group = NULL; bool need_generate = false; bool cleared_memstore = false; bool has_memstore = false; int16_t restore_state = ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; LOG_INFO("start ObRestoreTailoredPrepareTask process", "pg_key", ctx_->replica_op_arg_.key_); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored prepare task do not init", K(ret)); } else if (OB_ISNULL(partition_group = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("partition should not be NULL", K(ret), KP(partition_group)); } else if (FALSE_IT(restore_state = partition_group->get_pg_storage().get_restore_state())) { } else if (REPLICA_RESTORE_DATA != restore_state && REPLICA_RESTORE_CUT_DATA != restore_state && REPLICA_RESTORE_ARCHIVE_DATA != restore_state && REPLICA_RESTORE_STANDBY_CUT != restore_state) { ret = OB_ERR_UNEXPECTED; LOG_WARN("unexpected restore state during tailored task", K(ret), K(restore_state)); } else if (REPLICA_RESTORE_ARCHIVE_DATA == restore_state) { LOG_INFO("replica archive data not need cut data, skip", K(restore_state)); } else if (FALSE_IT(has_memstore = partition_group->get_pg_storage().has_memstore())) { } else if (has_memstore) { ObTablesHandle tables_handle; if (OB_FAIL(partition_group->get_pg_storage().get_reference_memtables(tables_handle))) { LOG_WARN("failed to get memtables", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < tables_handle.get_count(); ++i) { memtable::ObMemtable* memtable = static_cast(tables_handle.get_table(i)); if (OB_ISNULL(memtable)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("memtable should not be NULL", KP(memtable)); } else if (memtable->not_empty()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("memtable should be empty", K(ret), "pg_key", ctx_->replica_op_arg_.key_); } } } if (OB_SUCC(ret)) { if (OB_FAIL(partition_group->get_pg_storage().clear_non_reused_stores( ctx_->replica_op_arg_.key_, cleared_memstore))) { LOG_WARN("failed to clean non reuse stores", K(ret), K(*ctx_)); } else if (!cleared_memstore) { ret = OB_ERR_UNEXPECTED; LOG_WARN("has memstore but do do cleared memstore", K(ret), K(has_memstore), K(cleared_memstore)); } } } if (OB_FAIL(ret)) { } else if (OB_FAIL(check_need_generate_task(need_generate))) { LOG_WARN("failed to check need generate task", K(ret)); } else if (!need_generate) { // do nothing } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc finish task", K(ret)); } else if (OB_FAIL(finish_task->init())) { LOG_WARN("failed to init finish task", K(ret)); } else if (OB_FAIL(update_restore_flag_cut_data())) { LOG_WARN("failed to update restore flag cut data", K(ret)); } else if (OB_FAIL(schedule_restore_tailored_task(*finish_task))) { LOG_WARN("failed to schedule restore tailored task", K(ret)); } else if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish task", K(ret)); } if (OB_FAIL(ret) && NULL != ctx_) { ctx_->set_result_code(ret); } LOG_INFO("end ObRestoreTailoredPrepareTask process", "pg_key", ctx_->replica_op_arg_.key_); return ret; } int ObRestoreTailoredPrepareTask::schedule_restore_tailored_task(ObRestoreTailoredFinishTask& finish_task) { int ret = OB_SUCCESS; ObIPartitionGroup* partition_group = NULL; ObPartitionArray pkeys; const bool include_trans_table = false; const bool is_restore_point = false; const int64_t restore_schema_version = OB_INVALID_VERSION; const int64_t current_schema_version = OB_INVALID_VERSION; ObRecoveryPointSchemaFilter schema_filter; uint64_t tenant_id = OB_INVALID_ID; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored prepare task do not init", K(ret)); } else if (FALSE_IT(tenant_id = is_inner_table(ctx_->replica_op_arg_.key_.get_table_id()) ? OB_SYS_TENANT_ID : ctx_->replica_op_arg_.key_.get_tenant_id())) { } else if (OB_ISNULL(partition_group = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("partition group should not be NULL", K(ret), KP(partition_group)); } else if (OB_FAIL(partition_group->get_all_pg_partition_keys(pkeys, include_trans_table))) { LOG_WARN("failed to get all pg partition keys", K(ret), "pg key", ctx_->replica_op_arg_.key_); } else { ObSchemaGetterGuard schema_guard; int64_t schema_version = 0; if (OB_FAIL(MIGRATOR.get_schema_service()->get_tenant_schema_guard(tenant_id, schema_guard))) { LOG_WARN("failed to get schema guard", K(ret), K(partition_group->get_partition_key())); } else if (OB_FAIL(schema_guard.get_schema_version(tenant_id, schema_version))) { LOG_WARN("failed to get table schema version", K(ret), K(partition_group->get_partition_key())); } else if (OB_FAIL(finish_task.set_schema_version(schema_version))) { LOG_WARN("failed to set schema version", K(ret), K(schema_version), K(partition_group->get_partition_key())); } } if (OB_FAIL(ret)) { } else if (OB_FAIL(schema_filter.init(tenant_id, is_restore_point, restore_schema_version, current_schema_version))) { LOG_WARN("failed to init schema filter", K(ret), K(tenant_id), K(restore_schema_version)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < pkeys.count(); ++i) { const ObPartitionKey& pkey = pkeys.at(i); ObPGPartitionGuard pg_partition_guard; ObTablesHandle tables_handle; ObPartitionStorage* storage = NULL; ObArray index_ids; bool is_exist = false; if (schema_filter.check_partition_exist(pkey, is_exist)) { LOG_WARN("failed to check partition exist", K(ret), K(pkey)); } else if (!is_exist) { // do nothing } else if (OB_FAIL(partition_group->get_pg_partition(pkey, pg_partition_guard))) { STORAGE_LOG(WARN, "get pg partition error", K(ret), K(pkey)); } else if (OB_ISNULL(pg_partition_guard.get_pg_partition())) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "get pg partition error", K(ret), KP(pg_partition_guard.get_pg_partition())); } else if (OB_ISNULL(storage = static_cast( pg_partition_guard.get_pg_partition()->get_storage()))) { ret = OB_ERR_UNEXPECTED; STORAGE_LOG(WARN, "partition storage must not be NULL", K(ret)); } else if (OB_FAIL(storage->get_partition_store().get_all_table_ids(index_ids))) { STORAGE_LOG(WARN, "failed to get all table ids", K(ret)); } else if (OB_FAIL(filter_tailored_tables(index_ids, schema_filter))) { LOG_WARN("failed to filter tailored tables", K(ret)); } else if (OB_FAIL(schedule_restore_table_tailored_task_(index_ids, pkey, *storage, finish_task))) { STORAGE_LOG(WARN, "failed to schedule restore table tailored task", K(ret)); } } } return ret; } int ObRestoreTailoredPrepareTask::schedule_restore_table_tailored_task_(const common::ObIArray& index_ids, const ObPartitionKey& partition_key, ObPartitionStorage& storage, ObRestoreTailoredFinishTask& finish_task) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored prepare task do not init", K(ret)); } else { const ObPGKey& pg_key = ctx_->replica_op_arg_.key_; for (int64_t i = 0; OB_SUCC(ret) && i < index_ids.count(); ++i) { const uint64_t index_id = index_ids.at(i); ObTablesHandle minor_tables_handle; ObTableHandle major_handle; ObRestoreTailoredTask* task = NULL; if (OB_FAIL(storage.get_partition_store().get_latest_minor_sstables(index_id, minor_tables_handle))) { LOG_WARN("failed to get latest minor sstables", K(ret), K(index_id)); } else if (OB_FAIL(storage.get_partition_store().get_last_major_sstable(index_id, major_handle))) { LOG_WARN("failed to get lastest major sstable", K(ret), K(index_id)); } else if (OB_FAIL(dag_->alloc_task(task))) { LOG_WARN("failed to alloc task", K(ret)); } else if (OB_FAIL(task->init(index_id, minor_tables_handle, major_handle, pg_key, partition_key, finish_task))) { LOG_WARN("failed to init restore tailored prepare task", K(ret), K(partition_key)); } else if (OB_FAIL(this->add_child(*task))) { LOG_WARN("failed to add child of task", K(ret)); } else if (OB_FAIL(dag_->add_task(*task))) { LOG_WARN("failed to add task to dag", K(ret)); } else if (OB_FAIL(task->add_child(finish_task))) { LOG_WARN("failed to add finish task", K(ret)); } } } return ret; } int ObRestoreTailoredPrepareTask::filter_tailored_tables( common::ObIArray& index_ids, ObRecoveryPointSchemaFilter& schema_filter) { int ret = OB_SUCCESS; ObArray tmp_index_ids; for (int64_t i = 0; OB_SUCC(ret) && i < index_ids.count(); ++i) { const uint64_t index_id = index_ids.at(i); bool is_exist = false; if (OB_FAIL(schema_filter.check_table_exist(index_id, is_exist))) { LOG_WARN("failed to check table exist", K(ret), K(index_id)); } else if (!is_exist) { // do nothing } else if (OB_FAIL(tmp_index_ids.push_back(index_id))) { LOG_WARN("failed to push index id into array", K(ret), K(index_id)); } } if (OB_SUCC(ret)) { index_ids.reset(); if (OB_FAIL(index_ids.assign(tmp_index_ids))) { LOG_WARN("failed to assign index ids", K(ret)); } } return ret; } int ObRestoreTailoredPrepareTask::update_restore_flag_cut_data() { int ret = OB_SUCCESS; ObIPartitionGroup* partition_group = NULL; int16_t restore_status = ObReplicaRestoreStatus::REPLICA_NOT_RESTORE; const int16_t flag = REPLICA_RESTORE_CUT_DATA; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore failored prepare task do not init", K(ret)); } else if (OB_ISNULL((partition_group = ctx_->partition_guard_.get_partition_group()))) { ret = OB_ERR_UNEXPECTED; LOG_WARN("partition group should not be NULL", K(ret), KP(partition_group)); } else if (FALSE_IT(restore_status = partition_group->get_pg_storage().get_restore_state())) { } else if (RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_) { if (REPLICA_RESTORE_STANDBY_CUT != restore_status) { ret = OB_ERR_UNEXPECTED; LOG_WARN("restore standby op get invlaid restore status", K(ret), K(*ctx_), K(restore_status)); } else { // do nothing } } else if (REPLICA_RESTORE_CUT_DATA == restore_status) { // do nothing } else { const ObPGKey& pg_key = ctx_->replica_op_arg_.key_; if (OB_FAIL(MIGRATOR.get_partition_service()->set_restore_flag(pg_key, flag))) { LOG_WARN("failed to set restore flag", K(ret), K(pg_key)); } } return ret; } int ObRestoreTailoredPrepareTask::check_need_generate_task(bool& need_generate) { int ret = OB_SUCCESS; need_generate = false; ObIPartitionGroup* partition_group = NULL; ObTablesHandle tables_handle; int64_t max_upper_trans_version = 0; transaction::ObTransService* trans_service = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore failored prepare task do not init", K(ret)); } else if (OB_ISNULL(partition_group = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("partition group should not be NULL", K(ret), KP(partition_group)); } else if (OB_FAIL(partition_group->get_all_tables(tables_handle))) { LOG_WARN("failed to get all tables", K(ret), KP(partition_group)); } else if (OB_ISNULL(trans_service = GCTX.par_ser_->get_trans_service())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("trans service should not be NULL", K(ret), KP(trans_service)); } else if (RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_ && REPLICA_RESTORE_STANDBY_CUT == partition_group->get_pg_storage().get_restore_status()) { need_generate = true; LOG_INFO("standby restore cut data"); } else { const int64_t restore_snapshot_version = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.restore_snapshot_version_; for (int64_t i = 0; OB_SUCC(ret) && i < tables_handle.get_count(); ++i) { ObITable* table = tables_handle.get_table(i); ObSSTable* sstable = NULL; if (OB_ISNULL(table)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("table should not be NULL", K(ret), KP(table)); } else if (table->is_major_sstable() || table->is_trans_sstable()) { // do nothing } else if (OB_ISNULL(sstable = reinterpret_cast(table))) { ret = OB_ERR_UNEXPECTED; LOG_WARN("sstable should not be NULL", K(ret), KP(sstable)); } else { int64_t max_trans_version = sstable->get_upper_trans_version(); if (INT64_MAX == sstable->get_upper_trans_version()) { bool is_all_rollback_trans = false; // get_all_latest_minor_sstables return all minor except complement // all normal minor sstable consist of data between [start_log_ts, end_log_ts] if (OB_FAIL(trans_service->get_max_trans_version_before_given_log_ts( sstable->get_partition_key(), sstable->get_end_log_ts(), max_trans_version, is_all_rollback_trans))) { LOG_WARN("failed to get_max_trans_version_before_given_log_id", K(ret), KPC(sstable)); } else if (0 == max_trans_version && !is_all_rollback_trans) { ret = OB_ERR_UNEXPECTED; LOG_ERROR("max trans version should not be 0", K(sstable->get_partition_key()), "end_log_ts", sstable->get_end_log_ts()); } } if (OB_SUCC(ret)) { max_upper_trans_version = std::max(max_trans_version, max_upper_trans_version); } } } if (OB_SUCC(ret)) { need_generate = restore_snapshot_version < max_upper_trans_version; // do need generate task if (!need_generate) { LOG_INFO("restore snapshot version is larger than max upper trans version, no need cut", K(restore_snapshot_version), K(max_upper_trans_version), K(tables_handle)); } } } return ret; } ObRestoreTailoredTask::ObRestoreTailoredTask() : ObITask(TASK_TYPE_RESTORE_TAILORED_PROCESS), is_inited_(false), ctx_(NULL), index_id_(OB_INVALID_ID), minor_tables_handle_(), major_table_handle_(), pg_key_(), finish_task_(NULL) {} ObRestoreTailoredTask::~ObRestoreTailoredTask() {} int ObRestoreTailoredTask::init(const uint64_t index_id, const ObTablesHandle& minor_tables_handle, const ObTableHandle& major_table_handle, const ObPGKey& pg_key, const ObPartitionKey& partition_key, ObRestoreTailoredFinishTask& finish_task) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_WARN("restore tailored task has already inited", K(ret)); } else if (OB_INVALID_ID == index_id || !pg_key.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("init restore tailored task get invalid argument", K(ret), K(index_id), K(pg_key)); } else if (OB_FAIL(minor_tables_handle_.add_tables(minor_tables_handle))) { LOG_WARN("failed to add table handle", K(ret), K(index_id)); } else if (OB_FAIL(major_table_handle_.assign(major_table_handle))) { LOG_WARN("failed to add major tables handle", K(ret), K(index_id)); } else { index_id_ = index_id; pg_key_ = pg_key; partition_key_ = partition_key; ctx_ = static_cast(dag_)->get_ctx(); finish_task_ = &finish_task; is_inited_ = true; } return ret; } int ObRestoreTailoredTask::process() { int ret = OB_SUCCESS; ObITable::TableKey table_key; ObTailoredRowIterator row_iter; ObMigrateLogicRowWriter logic_row_writer; ObMacroBlocksWriteCtx block_write_ctx; ObMacroBlocksWriteCtx lob_block_write_ctx; const int64_t schema_version = OB_INVALID_VERSION; int64_t restore_snapshot = 0; LOG_INFO("start ObRestoreTailoredTask process", "table_id", index_id_, "partition_key", partition_key_); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored task do not init", K(ret)); } else if (minor_tables_handle_.empty()) { // do nothing LOG_INFO("minor tables handle is empty, no need cut", K(partition_key_), K(index_id_)); } else if (FALSE_IT( restore_snapshot = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.restore_snapshot_version_)) { } else if (OB_FAIL(get_tailored_table_key(table_key))) { LOG_WARN("failed to get tailored table key", K(ret), K(index_id_), K(pg_key_)); } else if (OB_FAIL(row_iter.init( index_id_, pg_key_, schema_version, table_key, restore_snapshot, minor_tables_handle_))) { LOG_WARN("failed to init tailored row iter", K(ret), K(index_id_), K(pg_key_), K(partition_key_)); } else if (OB_FAIL(logic_row_writer.init(&row_iter, pg_key_, ctx_->get_partition()->get_storage_file_handle()))) { LOG_WARN("failed to init logic row writer", K(ret), K(pg_key_), K(partition_key_)); } else if (OB_FAIL(logic_row_writer.process(block_write_ctx, lob_block_write_ctx))) { LOG_WARN("failed to process writer", K(ret)); } else if (OB_FAIL(generate_new_minor_sstable(table_key, block_write_ctx, lob_block_write_ctx))) { LOG_WARN("failed to generate new sstable", K(ret), K(table_key)); } else if (OB_FAIL(generate_major_sstable())) { LOG_WARN("failed to generate major sstable", K(ret), K(partition_key_), K(index_id_)); } if (OB_FAIL(ret) && NULL != ctx_) { ctx_->set_result_code(ret); } LOG_INFO("end ObRestoreTailoredTask process", "table_id", index_id_, "partition key", partition_key_); return ret; } int ObRestoreTailoredTask::get_tailored_table_key(ObITable::TableKey& table_key) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored task do not init", K(ret)); } else { int64_t base_version = INT64_MAX; int64_t multi_version_start = 0; int64_t snapshot_version = 0; int64_t start_log_ts = INT64_MAX; int64_t end_log_ts = 0; int64_t max_log_ts = 0; int64_t pre_log_ts = 0; for (int64_t i = 0; OB_SUCC(ret) && i < minor_tables_handle_.get_count(); ++i) { const ObITable* table = minor_tables_handle_.get_table(i); if (OB_ISNULL(table)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("table should not be NULL", K(ret), KP(table), K(index_id_), K(partition_key_), K(pg_key_)); } else if (!table->is_minor_sstable()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("table type is unexpected", K(ret), KPC(table)); } else if (0 == i) { pre_log_ts = table->get_end_log_ts(); multi_version_start = table->get_multi_version_start(); } else if (pre_log_ts > table->get_start_log_ts()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("minor table is not continue", K(ret), K(*table)); } else { if (table->get_base_version() != table->get_multi_version_start()) { multi_version_start = table->get_multi_version_start(); } pre_log_ts = table->get_end_log_ts(); } if (OB_SUCC(ret)) { base_version = std::min(base_version, table->get_base_version()); start_log_ts = std::min(start_log_ts, table->get_start_log_ts()); end_log_ts = std::max(end_log_ts, table->get_end_log_ts()); max_log_ts = std::max(max_log_ts, table->get_max_log_ts()); snapshot_version = std::max(snapshot_version, table->get_snapshot_version()); } } if (OB_SUCC(ret)) { table_key.pkey_ = partition_key_; table_key.table_id_ = index_id_; table_key.table_type_ = ObITable::MULTI_VERSION_MINOR_SSTABLE; table_key.trans_version_range_.base_version_ = base_version; table_key.trans_version_range_.multi_version_start_ = multi_version_start; table_key.trans_version_range_.snapshot_version_ = std::min(ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.restore_snapshot_version_, snapshot_version); table_key.version_.version_ = ctx_->replica_op_arg_.phy_restore_arg_.restore_data_version_ + 1; // table_key.upper_trans_version // table_key.max_merged_version // log ts table_key.log_ts_range_.start_log_ts_ = start_log_ts; table_key.log_ts_range_.end_log_ts_ = end_log_ts; table_key.log_ts_range_.max_log_ts_ = max_log_ts; if (!table_key.is_valid()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("tailored table key is invalid", K(ret), K(table_key)); } else if (1 == minor_tables_handle_.get_count()) { const ObITable* table = minor_tables_handle_.get_table(0); if (table->get_key() == table_key) { // same table key not allowed because can not add in table mgr table_key.table_type_ = ObITable::MINI_MINOR_SSTABLE; if (!table_key.is_valid()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("tailored table key is invalid", K(ret), K(table_key)); } } } if (OB_FAIL(ret)) { } else { LOG_INFO("new cut table key", K(table_key)); } } } return ret; } int ObRestoreTailoredTask::generate_new_minor_sstable(const ObITable::TableKey& table_key, ObMacroBlocksWriteCtx& block_write_ctx, ObMacroBlocksWriteCtx& lob_block_write_ctx) { int ret = OB_SUCCESS; storage::ObCreateSSTableParamWithTable create_sstable_param; ObSSTable* sstable = NULL; ObSchemaGetterGuard schema_guard; ObTableHandle handle; bool has_lob_column = false; const uint64_t table_id = index_id_; const uint64_t tenant_id = is_inner_table(table_id) ? OB_SYS_TENANT_ID : extract_tenant_id(table_id); const bool check_formal = extract_pure_id(table_id) > OB_MAX_CORE_TABLE_ID; // Avoid the problem of circular dependency if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored task do not init", K(ret)); } else if (!table_key.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("generate new sstable get invalid argument", K(ret), K(table_key)); } else { create_sstable_param.table_key_ = table_key; create_sstable_param.logical_data_version_ = table_key.trans_version_range_.snapshot_version_; if (OB_FAIL(MIGRATOR.get_schema_service()->get_tenant_schema_guard(tenant_id, schema_guard))) { LOG_WARN("failed to get schema guard", K(ret), K(tenant_id), K(table_key)); } else if (check_formal && OB_FAIL(schema_guard.check_formal_guard())) { LOG_WARN("schema_guard is not formal", K(ret), K(tenant_id)); } else if (OB_FAIL(schema_guard.get_table_schema(table_key.table_id_, create_sstable_param.schema_))) { LOG_WARN("Fail to get table schema, ", K(ret), K(partition_key_)); } else if (OB_ISNULL(create_sstable_param.schema_)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("Unexpected null table schema", K(ret), K(partition_key_)); } else if (OB_FAIL( schema_guard.get_table_schema_version(table_key.table_id_, create_sstable_param.schema_version_))) { LOG_WARN("failed to get table schema version", K(ret), K(table_key)); } else if (OB_FAIL(create_sstable_param.schema_->has_lob_column(has_lob_column, true))) { LOG_WARN("Failed to check lob column in table schema", K(ret)); } else { ObPGCreateSSTableParam param; param.with_table_param_ = &create_sstable_param; if (!block_write_ctx.file_handle_.is_valid() && OB_FAIL(block_write_ctx.file_handle_.assign( ctx_->partition_guard_.get_partition_group()->get_storage_file_handle()))) { LOG_WARN("fail to assign file handle", K(ret)); } else if (!lob_block_write_ctx.file_handle_.is_valid() && OB_FAIL(lob_block_write_ctx.file_handle_.assign( ctx_->partition_guard_.get_partition_group()->get_storage_file_handle()))) { LOG_WARN("fail to assign file handle", K(ret)); } else if (OB_FAIL(param.data_blocks_.push_back(&block_write_ctx))) { LOG_WARN("fail to push back data block write ctx", K(ret)); } else if (has_lob_column && lob_block_write_ctx.is_valid() && !lob_block_write_ctx.is_empty()) { if (OB_FAIL(param.lob_blocks_.push_back(&lob_block_write_ctx))) { LOG_WARN("fail to push back lob block write ctx", K(ret)); } } if (OB_SUCC(ret)) { ObIPartitionGroup* pg = nullptr; if (OB_ISNULL(pg = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_SYS; LOG_WARN("partition must not null", K(ret)); } else if (OB_FAIL(pg->create_sstable(param, handle))) { LOG_WARN("fail to create sstable", K(ret)); } else if (OB_FAIL(handle.get_sstable(sstable))) { LOG_WARN("fail to get sstable", K(ret)); } else if (OB_FAIL(finish_task_->add_sstable_handle(partition_key_, handle))) { LOG_WARN("failed to add sstable handle", K(ret)); } if (OB_SUCC(ret)) { if (OB_FAIL(finish_task_->set_schema_version(create_sstable_param.schema_version_))) { LOG_WARN("failed to set schema version", K(ret), K(create_sstable_param)); } } } } } return ret; } int ObRestoreTailoredTask::generate_major_sstable() { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored task do not init", K(ret)); } else if (OB_FAIL(finish_task_->add_sstable_handle(partition_key_, major_table_handle_))) { LOG_WARN("failed to add sstable handle", K(ret), K(partition_key_)); } return ret; } ObRestoreTailoredFinishTask::ObRestoreTailoredFinishTask() : ObITask(TASK_TYPE_RESTORE_TAILORED_FINISH), is_inited_(false), ctx_(NULL), lock_(), part_ctx_array_(), schema_version_(OB_INVALID_VERSION) {} ObRestoreTailoredFinishTask::~ObRestoreTailoredFinishTask() {} int ObRestoreTailoredFinishTask::add_sstable_handle(const ObPartitionKey& pkey, ObTableHandle& handle) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored finish task do not init", K(ret)); } else if (!pkey.is_valid() || !handle.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("add sstable handle get invalid argument", K(ret), K(handle)); } else { common::SpinWLockGuard guard(lock_); bool found = false; for (int64_t i = 0; OB_SUCC(ret) && i < part_ctx_array_.count() && !found; ++i) { ObPartitionMigrateCtx& part_ctx = part_ctx_array_.at(i); if (part_ctx.copy_info_.meta_.pkey_ == pkey) { if (OB_FAIL(part_ctx.handle_.add_table(handle))) { LOG_WARN("failed to add table handle", K(ret), K(handle)); } else { found = true; } } } if (OB_SUCC(ret) && !found) { ObPartitionMigrateCtx part_ctx; part_ctx.is_partition_exist_ = true; part_ctx.need_reuse_local_minor_ = false; part_ctx.ctx_ = ctx_; if (OB_FAIL(ctx_->partition_guard_.get_partition_group()->get_pg_storage().get_pg_partition_store_meta( pkey, part_ctx.copy_info_.meta_))) { LOG_WARN("failed to get pg partition store meta", K(ret), K(pkey)); } else if (OB_FAIL(part_ctx.handle_.add_table(handle))) { LOG_WARN("failed to add table handle", K(ret), K(handle)); } else if (OB_FAIL(part_ctx_array_.push_back(part_ctx))) { LOG_WARN("failed to push part ctx into array", K(ret)); } } } return ret; } int ObRestoreTailoredFinishTask::init() { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_WARN("restore tailored finish task init twice", K(ret)); } else { ctx_ = static_cast(dag_)->get_ctx(); is_inited_ = true; } return ret; } int ObRestoreTailoredFinishTask::process() { int ret = OB_SUCCESS; const int64_t schema_version = OB_INVALID_VERSION; ObIPartitionGroup* partition_group = NULL; ObSavedStorageInfoV2 save_info; ObPGKey pg_key; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored finish task do not init", K(ret)); } else { pg_key = ctx_->replica_op_arg_.key_; LOG_INFO("start ObRestoreTailoredFinishTask process", "pg key", pg_key); } if (OB_FAIL(ret)) { // do nothing } else if (OB_ISNULL(partition_group = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("partition group should not be NULL", K(ret), KP(partition_group)); } else { int64_t trans_table_seq = ctx_->partition_guard_.get_partition_group()->get_pg_storage().get_trans_table_seq(); const int64_t restore_snapshot_version = ctx_->replica_op_arg_.phy_restore_arg_.restore_info_.restore_snapshot_version_; if (OB_FAIL(partition_group->get_pg_storage().batch_replace_store_map( part_ctx_array_, schema_version, ctx_->is_restore_, trans_table_seq))) { LOG_WARN("failed to batch replace store map", K(ret)); } else if (OB_FAIL(partition_group->get_all_saved_info(save_info))) { LOG_WARN("failed to get all saved info", K(ret)); } else { ObDataStorageInfo& storage_info = save_info.get_data_info(); const int64_t publish_version = std::min(storage_info.get_publish_version(), restore_snapshot_version); storage_info.set_publish_version(publish_version); storage_info.set_schema_version(schema_version_); if (OB_FAIL(partition_group->set_storage_info(save_info))) { LOG_WARN("failed to set storage info", K(ret), K(save_info)); } } } if (OB_FAIL(ret) && NULL != ctx_) { ctx_->set_result_code(ret); } LOG_INFO("end ObRestoreTailoredFinishTask process", "pg key", pg_key); return ret; } int ObRestoreTailoredFinishTask::set_schema_version(const int64_t schema_version) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("restore tailored finish task do not init", K(ret)); } else if (schema_version < 0) { ret = OB_INVALID_ARGUMENT; LOG_WARN("set schema version get invalid argument", K(ret), K(schema_version)); } else { schema_version_ = std::max(schema_version_, schema_version); } return ret; } ObMigrateTaskGeneratorTask::ObMigrateTaskGeneratorTask() : ObITableTaskGeneratorTask(), is_inited_(false) {} ObMigrateTaskGeneratorTask::~ObMigrateTaskGeneratorTask() {} int ObMigrateTaskGeneratorTask::init() { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (ObIDag::DAG_TYPE_MIGRATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP != dag_->get_type() && ObIDag::DAG_TYPE_VALIDATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_BACKUPSET != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_ARCHIVELOG != dag_->get_type()) { ret = OB_ERR_SYS; LOG_ERROR("dag type not match", K(ret), K(*dag_)); } else { ctx_ = static_cast(dag_)->get_ctx(); is_inited_ = true; bandwidth_throttle_ = MIGRATOR.get_bandwidth_throttle(); partition_service_ = MIGRATOR.get_partition_service(); srv_rpc_proxy_ = MIGRATOR.get_svr_rpc_proxy(); rpc_ = MIGRATOR.get_pts_rpc(); cp_fty_ = MIGRATOR.get_cp_fty(); } return ret; } int ObMigrateTaskGeneratorTask::process() { int ret = OB_SUCCESS; if (IS_NOT_INIT) { ret = OB_NOT_INIT; LOG_WARN("task is not inited", K(ret)); } else { if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_) { LOG_INFO("no need replay for BACKUP_REPLICA_OP", "arg", ctx_->replica_op_arg_); } else if (ctx_->create_new_pg_) { if (OB_FAIL(deal_with_new_partition())) { LOG_WARN("failed to enable_replay_with_new_partition", K(ret), "arg", ctx_->replica_op_arg_); } } else if (VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_ || BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_ || BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { // do nothing } else { if (REBUILD_REPLICA_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(deal_with_rebuild_partition())) { LOG_WARN("failed to offline_rebuild_partition", K(ret), "arg", ctx_->replica_op_arg_); } } else if (RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_) { if (OB_FAIL(deal_with_standby_restore_partition())) { LOG_WARN("failed to deal with standby restore partition", K(ret), "arg", ctx_->replica_op_arg_); } } else { if (OB_FAIL(deal_with_old_partition())) { LOG_WARN("failed to deal_with_old_partition", K(ret), "arg", ctx_->replica_op_arg_); } } } if (OB_SUCC(ret)) { if (OB_FAIL(check_partition_integrity())) { LOG_WARN("failed to check partition intergrity", K(ret), "arg", ctx_->replica_op_arg_); } else if (OB_FAIL(schedule_migrate_tasks())) { LOG_WARN("failed to schedule migrate tasks", K(ret), KPC(ctx_)); } } if (OB_FAIL(ret) && NULL != ctx_) { if (ctx_->is_need_retry(ret)) { ctx_->need_rebuild_ = true; LOG_INFO("migrate post prepare task need retry", K(ret), K(*ctx_)); } ctx_->set_result_code(ret); } LOG_INFO("migrate post prepare task finish", K(ret)); } return ret; } ObMigrateRecoveryPointTaskGeneratorTask::ObMigrateRecoveryPointTaskGeneratorTask() : ObITableTaskGeneratorTask(), is_inited_(false), recovery_point_meta_info_(), is_recovery_point_exist_(false), wait_recovery_point_task_(NULL) {} ObMigrateRecoveryPointTaskGeneratorTask::~ObMigrateRecoveryPointTaskGeneratorTask() {} int ObMigrateRecoveryPointTaskGeneratorTask::init(share::ObITask& wait_recovery_point_task) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (ObIDag::DAG_TYPE_MIGRATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP != dag_->get_type() && ObIDag::DAG_TYPE_VALIDATE != dag_->get_type()) { ret = OB_ERR_SYS; LOG_ERROR("dag type not match", K(ret), K(*dag_)); } else if (FALSE_IT(ctx_ = static_cast(dag_)->get_ctx())) { } else if (ADD_REPLICA_OP != ctx_->replica_op_arg_.type_ && MIGRATE_REPLICA_OP != ctx_->replica_op_arg_.type_ && REBUILD_REPLICA_OP != ctx_->replica_op_arg_.type_ && CHANGE_REPLICA_OP != ctx_->replica_op_arg_.type_) { ret = OB_INVALID_ARGUMENT; LOG_WARN("init migrate recovery point failed", K(ret)); } else if (OB_FAIL(this->add_child(wait_recovery_point_task))) { LOG_WARN("failed to add recovery point finish task", K(ret)); } else { ctx_ = static_cast(dag_)->get_ctx(); bandwidth_throttle_ = MIGRATOR.get_bandwidth_throttle(); partition_service_ = MIGRATOR.get_partition_service(); srv_rpc_proxy_ = MIGRATOR.get_svr_rpc_proxy(); rpc_ = MIGRATOR.get_pts_rpc(); cp_fty_ = MIGRATOR.get_cp_fty(); wait_recovery_point_task_ = &wait_recovery_point_task; is_inited_ = true; } return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::process() { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_ISNULL(ctx_->get_partition())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("ctx do not hold partition", K(ret)); } else if (ctx_->recovery_point_ctx_.is_recovery_point_iter_finish()) { // do nothing } else if (OB_FAIL(build_migrate_recovery_point_task())) { LOG_WARN("failed to prepare migrate", K(ret)); } else if (OB_FAIL(schedule_migrate_recovery_point_tasks())) { LOG_WARN("failed to schedule migrate recovery point task", K(ret)); } if (OB_SUCC(ret)) { ObTaskController::get().allow_next_syslog(); LOG_INFO("finish migrate recovery point task generator task", K(*ctx_)); } if (OB_FAIL(ret) && NULL != ctx_) { if (ctx_->is_need_retry(ret)) { ctx_->need_rebuild_ = true; LOG_INFO("migrate prepare task need retry", K(ret), K(*ctx_)); } ctx_->set_result_code(ret); } LOG_INFO("end ObMigrateRecoveryPointTaskGeneratorTask process", K(ret)); return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::build_migrate_recovery_point_task() { int ret = OB_SUCCESS; const bool is_write_lock = true; bool is_exist = false; ObRecoveryPointKey recovery_point_key; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!ctx_->is_valid()) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "invalid migrate ctx", K(ret), K(*this), K(*ctx_)); } else if (1 != ctx_->doing_task_cnt_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "doing task must be 1 during prepare action", K(ret), K(*this), K(*ctx_)); } else if (REMOVE_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "should not prepare migrate for remove replica", K(ret), K(ctx_->replica_op_arg_)); } else if (!ctx_->recovery_point_ctx_.is_recovery_point_index_valid()) { ret = OB_ERR_SYS; LOG_WARN("recovery point index is invalid", K(ret), K(ctx_->recovery_point_ctx_)); } else if (ctx_->recovery_point_ctx_.is_recovery_point_iter_finish()) { // do nothing } else if (OB_FAIL(ctx_->recovery_point_ctx_.get_current_recovery_point_info(recovery_point_key))) { LOG_WARN("failed to get current recovery point info", K(ret), K(ctx_->recovery_point_ctx_)); } else { ObMigrateCtxGuard guard(is_write_lock, *ctx_); if (OB_FAIL(check_local_recovery_point_exist(recovery_point_key, is_exist))) { LOG_WARN("failed to check local recovery point exist", K(ret), K(recovery_point_key)); } else if (is_exist) { is_recovery_point_exist_ = true; LOG_INFO("local recovery point has exist ,no need generate it", K(ret), K(recovery_point_key)); } else if (OB_FAIL(build_migrate_recovery_point_info(recovery_point_key))) { LOG_WARN("failed to build migrate pg partition info", K(ret), K(*ctx_), K(ctx_->recovery_point_ctx_)); } } if (OB_SUCC(ret)) { ctx_->need_rebuild_ = false; ObTaskController::get().allow_next_syslog(); STORAGE_LOG(INFO, "finish build migrate recovery point task", "pkey", ctx_->replica_op_arg_.key_, K(*ctx_), K(ctx_->recovery_point_ctx_)); } else if (NULL != ctx_) { ctx_->result_ = ret; } return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::build_migrate_recovery_point_info( const ObRecoveryPointKey& recovery_point_key) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (OB_FAIL(get_recovery_point_meta_info(ctx_->migrate_src_info_, recovery_point_key))) { STORAGE_LOG(WARN, "failed to get recovery point meta info reader", K(ret), K(*ctx_)); } else if (!recovery_point_meta_info_.is_valid()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("recovery point meta info is invalid", K(ret), K(recovery_point_meta_info_)); } else if (OB_FAIL(inner_build_migrate_recovery_point_info(recovery_point_meta_info_))) { LOG_WARN("failed to do inner build migrate recovery point info", K(ret), K(recovery_point_meta_info_)); } return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::get_recovery_point_meta_info( const ObMigrateSrcInfo& src_info, const ObRecoveryPointKey& recovery_point_key) { int ret = OB_SUCCESS; ObRecoveryPointMetaInfoReader reader; ObFetchPGRecoveryPointMetaInfoArg arg; ObRecoveryPointMetaInfo tmp_recovery_point_meta_info; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate prepare task do not init", K(ret)); } else if (!src_info.is_valid() || !recovery_point_key.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("get recovery point meta info reader get invalid argument", K(ret), K(src_info), K(recovery_point_key)); } else if (FALSE_IT(arg.pg_key_ = ctx_->replica_op_arg_.key_)) { } else if (OB_FAIL(arg.recovery_point_key_array_.push_back(recovery_point_key))) { LOG_WARN("failed to push recovery point key into array", K(ret), K(recovery_point_key)); } else if (OB_FAIL( reader.init(*srv_rpc_proxy_, *bandwidth_throttle_, src_info.src_addr_, arg, src_info.cluster_id_))) { LOG_WARN("failed to init recovery point meta info reader", K(ret), K(src_info)); } else if (OB_FAIL(reader.fetch_recovery_point_meta_info(recovery_point_meta_info_))) { LOG_WARN("failed to fetch recovery point meta info", K(ret), K(arg)); } else if (OB_ITER_END != reader.fetch_recovery_point_meta_info(tmp_recovery_point_meta_info)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("has more recovery point meta info", K(ret), K(tmp_recovery_point_meta_info), K(arg)); } return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::inner_build_migrate_recovery_point_info( const ObRecoveryPointMetaInfo& recovery_point_meta_info) { int ret = OB_SUCCESS; int64_t cost_ts = ObTimeUtility::current_time(); if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (!recovery_point_meta_info.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("inner build migrate recovery point info get invalid argument", K(ret), K(recovery_point_meta_info)); } else if (OB_FAIL(build_migrate_recovery_point_table_info(recovery_point_meta_info))) { LOG_WARN("failed to build migrate recovery point table info", K(ret)); } cost_ts = ObTimeUtility::current_time() - cost_ts; LOG_INFO("inner_build_migrate_recovery_point_info", K(cost_ts), K(recovery_point_meta_info)); return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::check_local_recovery_point_exist( const ObRecoveryPointKey& recovery_point_key, bool& is_exist) { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; is_exist = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (NULL == (partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("local partition do not exist", K(ret), K(*ctx_)); } else if (OB_FAIL(partition->get_pg_storage().get_recovery_data_mgr().check_recovery_point_exist( recovery_point_key, is_exist))) { LOG_WARN("failed to check recovery point exist", K(ret), K(recovery_point_key)); } return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::build_migrate_recovery_point_table_info( const ObRecoveryPointMetaInfo& recovery_point_meta_info) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (!recovery_point_meta_info.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("build migrate recovery point table info get invalid argument", K(ret), K(recovery_point_meta_info)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < recovery_point_meta_info.table_keys_.count(); ++i) { const ObITable::TableKey& table_key = recovery_point_meta_info.table_keys_.at(i); ObTableHandle table_handle; ObITable* table = NULL; if (!table_key.is_valid() || table_key.is_memtable()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("table key is unexpected", K(ret), K(table_key)); } else if (OB_FAIL(partition_service_->acquire_sstable(table_key, table_handle))) { if (OB_ENTRY_NOT_EXIST != ret) { LOG_WARN("failed to get complete sstable by key", K(ret), K(table_key)); } else { ret = OB_SUCCESS; } } else if (NULL == (table = table_handle.get_table())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("table should not be NULL", K(ret), KP(table)); } else if (OB_FAIL(ctx_->recovery_point_ctx_.add_sstable(table_handle))) { LOG_WARN("failed to add table into tables handle", K(ret)); } } } return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::schedule_migrate_recovery_point_tasks() { int ret = OB_SUCCESS; ObITask* last_task = NULL; ObMigrateRecoveryPointFinishTask* finish_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(dag_->alloc_task(finish_task))) { LOG_WARN("failed to alloc task", K(ret)); } else if (OB_FAIL(finish_task->init( is_recovery_point_exist_, recovery_point_meta_info_, *ctx_, *wait_recovery_point_task_))) { LOG_WARN("failed to init finish task", K(ret)); } else if (is_recovery_point_exist_) { LOG_INFO("no need_generate_recovery point tasks"); } else if (OB_FAIL(generate_migrate_recovery_point_tasks(last_task))) { LOG_WARN("failed to generate migrate recovery point tasks", K(ret)); } if (OB_SUCC(ret)) { // generate finish task if (NULL != last_task) { if (OB_FAIL(last_task->add_child(*finish_task))) { LOG_WARN("failed to add child of last task", K(ret)); } } else { if (OB_FAIL(add_child(*finish_task))) { LOG_WARN("failed to add child of prepare task", K(ret)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(finish_task->add_child(*wait_recovery_point_task_))) { LOG_WARN("failed to add wait recovery point task", K(ret)); } else if (OB_FAIL(dag_->add_task(*finish_task))) { LOG_WARN("failed to add finish task", K(ret)); } } return ret; } int ObMigrateRecoveryPointTaskGeneratorTask::generate_migrate_recovery_point_tasks(ObITask*& last_task) { int ret = OB_SUCCESS; last_task = NULL; ObFakeTask* wait_migrate_finish_task = NULL; ObITask* parent_task = this; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (!recovery_point_meta_info_.is_valid()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("reocvery point meta info is invalid", K(ret), K(recovery_point_meta_info_)); } else if (OB_FAIL(generate_wait_migrate_finish_task(wait_migrate_finish_task))) { LOG_WARN("failed to generate_wait_migrate_finish_task", K(ret)); } else if (OB_ISNULL(wait_migrate_finish_task)) { ret = OB_ERR_SYS; LOG_ERROR("wait_rebuild_finish_task must not null", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < recovery_point_meta_info_.table_keys_.count(); ++i) { const ObITable::TableKey& table_key = recovery_point_meta_info_.table_keys_.at(i); ObFakeTask* wait_finish_task = NULL; ObMigrateTableInfo::SSTableInfo sstable_info; sstable_info.src_table_key_ = table_key; sstable_info.dest_base_version_ = table_key.trans_version_range_.base_version_; sstable_info.dest_log_ts_range_ = table_key.log_ts_range_; if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_physical_sstable_copy_task(ctx_->migrate_src_info_, ctx_->recovery_point_ctx_, sstable_info, parent_task, wait_finish_task))) { LOG_WARN("fail to generate physic sstable copy task", K(ret), K(ctx_->migrate_src_info_), K(sstable_info)); } else if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_INFO("succeed to generate sstable copy task", K(ctx_->migrate_src_info_), K(sstable_info)); } } } if (OB_SUCC(ret)) { if (OB_ISNULL(parent_task)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("parent task should not be NULL", K(ret), KP(parent_task)); } else if (OB_FAIL(parent_task->add_child(*wait_migrate_finish_task))) { LOG_WARN("failed to add wait migrate finish task ", K(ret)); } else { last_task = wait_migrate_finish_task; } } return ret; } ObMigrateRecoveryPointFinishTask::ObMigrateRecoveryPointFinishTask() : ObITask(TASK_TYPE_MIGRATE_FINISH), is_inited_(false), is_recovery_point_exist_(false), recovery_point_meta_info_(), ctx_(NULL), wait_recovery_point_finish_task_(NULL) {} ObMigrateRecoveryPointFinishTask::~ObMigrateRecoveryPointFinishTask() {} int ObMigrateRecoveryPointFinishTask::init(const bool is_recovery_point_exist, const ObRecoveryPointMetaInfo& recovery_point_meta_info, ObMigrateCtx& ctx, share::ObITask& wait_recovery_point_finish_task) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (!is_recovery_point_exist && !recovery_point_meta_info.is_valid()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("init migrate recovery point finish task failed", K(ret), K(is_recovery_point_exist), K(recovery_point_meta_info)); } else if (!is_recovery_point_exist && OB_FAIL(recovery_point_meta_info_.assign(recovery_point_meta_info))) { LOG_WARN("failed to copy recovery point meta info", K(ret), K(recovery_point_meta_info)); } else { is_recovery_point_exist_ = is_recovery_point_exist; ctx_ = &ctx; wait_recovery_point_finish_task_ = &wait_recovery_point_finish_task; is_inited_ = true; } return ret; } int ObMigrateRecoveryPointFinishTask::process() { int ret = OB_SUCCESS; ObRecoveryPointKey recovery_point_key; if (NULL != ctx_) { if (ctx_->recovery_point_ctx_.is_recovery_point_iter_finish()) { LOG_INFO("start ObMigrateRecoveryPointFinishTask process", "recovery_point index", ctx_->recovery_point_ctx_, K(ctx_->replica_op_arg_.type_)); } else if (OB_FAIL(ctx_->recovery_point_ctx_.get_current_recovery_point_info(recovery_point_key))) { LOG_WARN("failed to get recovery point info", K(ret)); } else { LOG_INFO("start ObMigrateRecoveryPointFinishTask process", "recovery_point index", ctx_->recovery_point_ctx_, "recovery point key", recovery_point_key, K(ctx_->replica_op_arg_.type_)); } } if (OB_FAIL(ret)) { } else if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (ctx_->recovery_point_ctx_.is_recovery_point_iter_finish()) { // do nothing } else if (OB_FAIL(create_recovery_point())) { LOG_WARN("failed to create recovery point", K(ret)); } else if (OB_FAIL(generate_next_task())) { LOG_WARN("failed to generate next task", K(ret)); } if (OB_FAIL(ret) && NULL != ctx_) { ctx_->need_rebuild_ = true; ctx_->set_result_code(ret); } if (NULL != ctx_) { LOG_INFO("end ObMigrateRecoveryPointFinishTask process", "pkey", ctx_->replica_op_arg_.key_, K(ctx_->replica_op_arg_.type_)); } return ret; } int ObMigrateRecoveryPointFinishTask::create_recovery_point() { int ret = OB_SUCCESS; ObIPartitionGroup* partition = NULL; ObRecoveryPointKey recovery_point_key; ObTablesHandle tables_handle; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (is_recovery_point_exist_) { // do nothing } else if (!recovery_point_meta_info_.is_valid()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("recovery point meta info should be valid", K(ret), K(recovery_point_meta_info_)); } else if (OB_FAIL(ctx_->recovery_point_ctx_.get_current_recovery_point_info(recovery_point_key))) { LOG_WARN("failed to get recovery point info", K(ret)); } else if (NULL == (partition = ctx_->partition_guard_.get_partition_group())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("local partition do not exist", K(ret), K(*ctx_)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < recovery_point_meta_info_.table_keys_.count(); ++i) { const ObITable::TableKey& table_key = recovery_point_meta_info_.table_keys_.at(i); ObTableHandle table_handle; if (OB_FAIL(partition->acquire_sstable(table_key, table_handle))) { LOG_WARN("failed to acquire sstable", K(ret), K(table_key)); } else if (OB_FAIL(tables_handle.add_table(table_handle))) { LOG_WARN("failed ot add table handle into tables handle", K(ret), K(table_handle)); } } if (OB_SUCC(ret)) { if (OB_FAIL(partition->get_pg_storage().get_recovery_data_mgr().add_recovery_point( recovery_point_key, recovery_point_meta_info_, tables_handle))) { if (OB_ENTRY_EXIST == ret) { LOG_INFO("recovery point has already exist", K(recovery_point_key)); ret = OB_SUCCESS; } else { LOG_WARN("failed to add recovery point info", K(ret), K(recovery_point_key), K(recovery_point_meta_info_), K(tables_handle)); } } } if (OB_SUCC(ret)) { ctx_->recovery_point_ctx_.reuse_tables_handle(); } } return ret; } int ObMigrateRecoveryPointFinishTask::generate_next_task() { int ret = OB_SUCCESS; ObITask* task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else if (OB_FAIL(inner_generate_next_task(task))) { LOG_WARN("failed to do inner generate next task", K(ret)); } else if (OB_ISNULL(task)) { ret = OB_ERR_UNEXPECTED; LOG_WARN("task should not be NULL", K(ret), KP(task)); } else if (OB_FAIL(dag_->add_task(*task))) { LOG_WARN("failed to add task into dag", K(ret), K(task)); } return ret; } int ObMigrateRecoveryPointFinishTask::inner_generate_next_task(ObITask*& task) { int ret = OB_SUCCESS; task = NULL; ObMigrateRecoveryPointTaskGeneratorTask* recovert_point_task_generator_task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("not inited", K(ret)); } else { // has next recovery point if (OB_FAIL(dag_->alloc_task(recovert_point_task_generator_task))) { LOG_WARN("failed to alloc copy task", K(ret)); } else if (OB_FAIL(recovert_point_task_generator_task->init(*wait_recovery_point_finish_task_))) { LOG_WARN("failed to init copy task", K(ret)); } else if (OB_FAIL(add_child(*recovert_point_task_generator_task))) { LOG_WARN("failed to add child copy task", K(ret)); } else { task = recovert_point_task_generator_task; ctx_->recovery_point_ctx_.inc_recovery_point_index(); } } return ret; } ObMigrateTransTableTaskGeneratorTask::ObMigrateTransTableTaskGeneratorTask() : ObITableTaskGeneratorTask(), is_inited_(false), wait_finish_task_(NULL) {} ObMigrateTransTableTaskGeneratorTask::~ObMigrateTransTableTaskGeneratorTask() {} int ObMigrateTransTableTaskGeneratorTask::init(share::ObITask& wait_finish_task) { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_ERROR("cannot init twice", K(ret)); } else if (ObIDag::DAG_TYPE_MIGRATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP != dag_->get_type() && ObIDag::DAG_TYPE_VALIDATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_BACKUPSET != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_ARCHIVELOG != dag_->get_type()) { ret = OB_ERR_SYS; LOG_ERROR("dag type not match", K(ret), K(*dag_)); } else { ctx_ = static_cast(dag_)->get_ctx(); bandwidth_throttle_ = MIGRATOR.get_bandwidth_throttle(); partition_service_ = MIGRATOR.get_partition_service(); srv_rpc_proxy_ = MIGRATOR.get_svr_rpc_proxy(); rpc_ = MIGRATOR.get_pts_rpc(); cp_fty_ = MIGRATOR.get_cp_fty(); wait_finish_task_ = &wait_finish_task; is_inited_ = true; } return ret; } int ObMigrateTransTableTaskGeneratorTask::process() { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (BACKUP_REPLICA_OP == ctx_->replica_op_arg_.type_ || BACKUP_BACKUPSET_OP == ctx_->replica_op_arg_.type_ || VALIDATE_BACKUP_OP == ctx_->replica_op_arg_.type_ || BACKUP_ARCHIVELOG_OP == ctx_->replica_op_arg_.type_) { // do nothing } else if (OB_ISNULL(ctx_->get_partition())) { ret = OB_ERR_UNEXPECTED; LOG_WARN("ctx do not hold partition", K(ret)); } else if (OB_FAIL(build_migrate_trans_table_task())) { LOG_WARN("failed to build migrate trans table task", K(ret)); } if (OB_SUCC(ret)) { ObTaskController::get().allow_next_syslog(); LOG_INFO("finish migrate trans table generator task", K(*ctx_)); } if (OB_FAIL(ret) && NULL != ctx_) { if (ctx_->is_need_retry(ret)) { ctx_->need_rebuild_ = true; LOG_INFO("migrate prepare task need retry", K(ret), K(*ctx_)); } ctx_->set_result_code(ret); } LOG_INFO("end ObMigrateTransTableTaskGeneratorTask process", K(ret)); return ret; } int ObMigrateTransTableTaskGeneratorTask::build_migrate_trans_table_task() { int ret = OB_SUCCESS; bool need_schedule = false; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate trans table task generator do not init", K(ret)); } else if (ObFetchPGInfoResult::FETCH_PG_INFO_RES_COMPAT_VERSION_V1 >= ctx_->fetch_pg_info_compat_version_) { LOG_INFO("compat version no need create trans table", K(ret), K(ctx_->fetch_pg_info_compat_version_)); } else if (OB_FAIL(need_generate_sstable_migrate_tasks(need_schedule))) { LOG_WARN("failed to check need generate sstable migrate tasks", K(ret)); } else if (!need_schedule) { // do nothing } else if (!need_migrate_trans_table(ctx_->replica_op_arg_.type_)) { LOG_INFO("no need to migrate trans table", K(ret), K(ctx_->replica_op_arg_.type_)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < ctx_->part_ctx_array_.count(); ++i) { ObPartitionMigrateCtx& part_migrate_ctx = ctx_->part_ctx_array_.at(i); if (!part_migrate_ctx.copy_info_.meta_.pkey_.is_trans_table()) { // do nothing } else if (OB_FAIL(inner_build_migrate_trans_table_task(part_migrate_ctx))) { LOG_WARN("failed to inner build migrate trans table task", K(ret), K(part_migrate_ctx)); } } } return ret; } int ObMigrateTransTableTaskGeneratorTask::inner_build_migrate_trans_table_task(ObPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate trans table task generator do not init", K(ret)); } else if (!part_migrate_ctx.copy_info_.meta_.pkey_.is_trans_table()) { ret = OB_INVALID_ARGUMENT; LOG_WARN("inner build migrate trans table task get invalid argument", K(ret)); } else { for (int64_t i = 0; OB_SUCC(ret) && i < part_migrate_ctx.copy_info_.table_infos_.count(); ++i) { const ObMigratePartitionInfo& copy_info = part_migrate_ctx.copy_info_; const ObArray& table_infos = copy_info.table_infos_; for (int64_t i = 0; OB_SUCC(ret) && i < table_infos.count(); ++i) { const ObMigrateTableInfo& table_info = table_infos.at(i); if (OB_FAIL(build_trans_sstable_tasks(ctx_->migrate_src_info_, table_info, part_migrate_ctx))) { LOG_WARN("failed to build trans sstable tasks", K(ret), K(i), K(table_info), K(part_migrate_ctx)); } } } } return ret; } int ObMigrateTransTableTaskGeneratorTask::build_trans_sstable_tasks( const ObMigrateSrcInfo& src_info, const ObMigrateTableInfo& table_info, ObPartitionMigrateCtx& part_migrate_ctx) { int ret = OB_SUCCESS; ObITask* parent_task = this; if (!is_inited_) { ret = OB_NOT_INIT; LOG_WARN("migrate trans table task generator do not init", K(ret)); } else if (!table_info.minor_sstables_.empty()) { ret = OB_ERR_UNEXPECTED; LOG_WARN("trans sstable has minor sstables which is unexpected", K(ret), K(table_info)); } else { // add trans sstables for (int64_t sstable_idx = 0; OB_SUCC(ret) && sstable_idx < table_info.major_sstables_.count(); ++sstable_idx) { const ObMigrateTableInfo::SSTableInfo& trans_table_info = table_info.major_sstables_.at(sstable_idx); ObFakeTask* wait_finish_task = NULL; if (!ObITable::is_trans_sstable(trans_table_info.src_table_key_.table_type_)) { ret = OB_ERR_SYS; LOG_ERROR("table type not match major sstable", K(ret), K(trans_table_info), K(table_info)); } else { if (OB_FAIL(dag_->alloc_task(wait_finish_task))) { LOG_WARN("failed to alloc wait finish task", K(ret)); } else if (OB_FAIL(generate_physical_sstable_copy_task( src_info, part_migrate_ctx, trans_table_info, parent_task, wait_finish_task))) { LOG_WARN("fail to generate physical sstable copy task", K(ret), K(trans_table_info), K(src_info)); } else if (OB_FAIL(dag_->add_task(*wait_finish_task))) { LOG_WARN("failed to add wait finish task", K(ret)); } else { parent_task = wait_finish_task; LOG_INFO("succeed to build trans sstable tasks", K(src_info), K(trans_table_info)); } } } if (OB_SUCC(ret)) { if (OB_FAIL(parent_task->add_child(*wait_finish_task_))) { LOG_WARN("failed to add wait_rebuild_finish_task", K(ret)); } } } return ret; } ObMigrateTaskSchedulerTask::ObMigrateTaskSchedulerTask() : ObITask(TASK_TYPE_MIGRATE_PREPARE), is_inited_(false), ctx_(NULL), cp_fty_(NULL), rpc_(NULL), srv_rpc_proxy_(NULL), bandwidth_throttle_(NULL), partition_service_(NULL) {} ObMigrateTaskSchedulerTask::~ObMigrateTaskSchedulerTask() {} int ObMigrateTaskSchedulerTask::init() { int ret = OB_SUCCESS; if (is_inited_) { ret = OB_INIT_TWICE; LOG_WARN("migrate task scheduler task init twice", K(ret)); } else if (ObIDag::DAG_TYPE_MIGRATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP != dag_->get_type() && ObIDag::DAG_TYPE_VALIDATE != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_BACKUPSET != dag_->get_type() && ObIDag::DAG_TYPE_BACKUP_ARCHIVELOG != dag_->get_type()) { ret = OB_ERR_SYS; LOG_ERROR("dag type not match", K(ret), K(*dag_)); } else { ctx_ = static_cast(dag_)->get_ctx(); is_inited_ = true; bandwidth_throttle_ = MIGRATOR.get_bandwidth_throttle(); partition_service_ = MIGRATOR.get_partition_service(); srv_rpc_proxy_ = MIGRATOR.get_svr_rpc_proxy(); rpc_ = MIGRATOR.get_pts_rpc(); cp_fty_ = MIGRATOR.get_cp_fty(); } return ret; } int ObMigrateTaskSchedulerTask::process() { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (NULL != ctx_) { ctx_->action_ = ObMigrateCtx::PREPARE; if (NULL != dag_) { ctx_->task_id_ = dag_->get_dag_id(); } if (OB_SUCCESS != (tmp_ret = (ctx_->trace_id_array_.push_back(*ObCurTraceId::get_trace_id())))) { STORAGE_LOG(WARN, "failed to push back trace id to array", K(tmp_ret)); } LOG_INFO("start ObMigrateTaskSchedulerTask process", "arg", ctx_->replica_op_arg_); } if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(add_migrate_status(ctx_))) { LOG_WARN("failed to add migrate status", K(ret)); } else { { const bool is_write_lock = true; ObMigrateCtxGuard guard(is_write_lock, *ctx_); if (!ctx_->is_valid()) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "invalid migrate ctx", K(ret), K(*this), K(*ctx_)); } else if (1 != ctx_->doing_task_cnt_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "doing task must be 1 during prepare action", K(ret), K(*this), K(*ctx_)); } else if (REMOVE_REPLICA_OP == ctx_->replica_op_arg_.type_) { ret = OB_ERR_SYS; STORAGE_LOG(ERROR, "should not prepare migrate for remove replica", K(ret), K(ctx_->replica_op_arg_)); } else if (OB_FAIL(try_hold_local_partition())) { LOG_WARN("failed to hold local partition", K(ret)); } } if (OB_SUCC(ret)) { if (OB_FAIL(schedule_task_by_type())) { LOG_WARN("failed to schedule task by type", K(ret)); } } } if (OB_SUCC(ret)) { ObTaskController::get().allow_next_syslog(); LOG_INFO("finish scheduler task", K(*ctx_)); } if (OB_FAIL(ret) && NULL != ctx_) { if (ctx_->is_need_retry(ret)) { ctx_->need_rebuild_ = true; LOG_INFO("migrate scheduler task need retry", K(ret), K(*ctx_)); } ctx_->set_result_code(ret); } LOG_INFO("end ObMigrateTaskSchedulerTask process", K(ret)); return ret; } int ObMigrateTaskSchedulerTask::add_partition_migration_status(const ObMigrateCtx& ctx) { int ret = OB_SUCCESS; int tmp_ret = OB_SUCCESS; if (!is_inited_) { ret = OB_NOT_INIT; STORAGE_LOG(WARN, "not inited", K(ret)); } else if (!ctx.is_valid()) { ret = OB_INVALID_ARGUMENT; STORAGE_LOG(WARN, "invalid args", K(ret), K(ctx)); } else { ObPartitionMigrationStatus task; task.task_id_ = ctx.task_id_; task.migrate_type_ = ctx.replica_op_arg_.get_replica_op_type_str(); task.pkey_ = ctx.replica_op_arg_.key_; task.clog_parent_.reset(); // we don't know it now task.src_.reset(); // we don't know it now task.dest_ = OBSERVER.get_self(); task.result_ = ctx.result_; task.start_time_ = ctx.create_ts_; task.action_ = ctx.action_; task.replica_state_ = OB_UNKNOWN_REPLICA; // we don't know it now task.doing_task_count_ = 0; task.total_task_count_ = 0; task.rebuild_count_ = 0; task.continue_fail_count_ = 0; task.data_statics_ = ctx.data_statics_; // allow comment truncation, no need to set ret (void)ctx.fill_comment(task.comment_, sizeof(task.comment_)); if (OB_SUCCESS != (tmp_ret = ObPartitionMigrationStatusMgr::get_instance().add_status(task))) { STORAGE_LOG(WARN, "failed to add partition migration status", K(tmp_ret), K(task)); } } return ret; } int ObMigrateTaskSchedulerTask::schedule_task_by_type() { int ret = OB_SUCCESS; ObIPartitionGroup* partition_group = NULL; int16_t restore_flag = REPLICA_NOT_RESTORE; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (FALSE_IT(partition_group = ctx_->partition_guard_.get_partition_group())) { } else if (OB_NOT_NULL(partition_group) && FALSE_IT(restore_flag = partition_group->get_pg_storage().get_restore_state())) { } else if ((RESTORE_REPLICA_OP == ctx_->replica_op_arg_.type_ || RESTORE_STANDBY_OP == ctx_->replica_op_arg_.type_) && (REPLICA_RESTORE_CUT_DATA == restore_flag || REPLICA_RESTORE_STANDBY_CUT == restore_flag)) { if (OB_FAIL(generate_restore_cut_prepare_task())) { LOG_WARN("failed to generate restore cut prepare task", K(ret), K(*ctx_)); } } else { if (OB_FAIL(generate_migrate_prepare_task())) { LOG_WARN("failed to generate migrate prepare task", K(ret), K(*ctx_)); } } return ret; } int ObMigrateTaskSchedulerTask::generate_restore_cut_prepare_task() { int ret = OB_SUCCESS; ObRestoreTailoredPrepareTask* task = NULL; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(dag_->alloc_task(task))) { LOG_WARN("failed to alloc dag task", K(ret), K(*ctx_)); } else if (OB_FAIL(task->init())) { LOG_WARN("failed to init restore tailored pepare task", K(ret), K(*ctx_)); } else if (OB_FAIL(this->add_child(*task))) { LOG_WARN("failed to add child task", K(ret), K(*ctx_)); } else if (OB_FAIL(dag_->add_task(*task))) { LOG_WARN("failed to add task to dag", K(ret), K(*ctx_)); } return ret; } int ObMigrateTaskSchedulerTask::generate_migrate_prepare_task() { int ret = OB_SUCCESS; ObMigratePrepareTask* task = nullptr; if (!is_inited_) { ret = OB_NOT_INIT; LOG_ERROR("not inited", K(ret)); } else if (OB_FAIL(dag_->alloc_task(task))) { LOG_WARN("failed to allock migrate prepare task", K(ret), K(*ctx_)); } else if (OB_FAIL(task->init())) { LOG_WARN("failed to init migrate prepare task", K(ret), K(*ctx_)); } else if (OB_FAIL(this->add_child(*task))) { LOG_WARN("failed to add child task", K(ret), K(*ctx_)); } else if (OB_FAIL(dag_->add_task(*task))) { LOG_WARN("failed to add task to dag", K(ret), K(*ctx_)); } return ret; } } /* namespace storage */ } /* namespace oceanbase */