Fix ls migration online succeed but inner retry return 4016 bug

src/storage/blocksstable/ob_macro_block_checker.cpp

@@ -38,6 +38,8 @@ int ObSSTableMacroBlockChecker::check(
     ret = OB_INVALID_ARGUMENT;
     STORAGE_LOG(WARN, "Invalid argument", K(ret), KP(macro_block_buf), K(macro_block_buf_size),
         K(check_level));
+  } else if (ObMacroBlockCheckLevel::CHECK_LEVEL_NONE == check_level) {
+    //do nothing
   } else if (OB_FAIL(common_header.deserialize(macro_block_buf, macro_block_buf_size, pos))) {
     STORAGE_LOG(ERROR, "fail to deserialize common header", K(ret), KP(macro_block_buf),
         K(macro_block_buf_size), K(pos), K(common_header));

src/storage/blocksstable/ob_macro_block_checker.h

@@ -26,8 +26,9 @@ namespace blocksstable
 {
 enum ObMacroBlockCheckLevel
 {
-  CHECK_LEVEL_PHYSICAL = 0, // verify data checksum
-  CHECK_LEVEL_LOGICAL  = 1, // verify column checksum
+  CHECK_LEVEL_NONE = 0,  // no check
+  CHECK_LEVEL_PHYSICAL = 1, // verify data checksum
+  CHECK_LEVEL_LOGICAL  = 2, // verify column checksum
   CHECK_LEVEL_MAX,
 };
 

src/storage/high_availability/ob_ls_migration.cpp

@@ -2915,7 +2915,6 @@ int ObDataTabletsMigrationTask::process()
 {
   int ret = OB_SUCCESS;
   int tmp_ret = OB_SUCCESS;
-  bool is_ls_online_success = false;
   LOG_INFO("start do data tablets migration task", K(ret), KPC(ctx_));
 #ifdef ERRSIM
   SERVER_EVENT_SYNC_ADD("storage_ha", "before_data_tablets_migration_task",
@@ -2935,7 +2934,6 @@ int ObDataTabletsMigrationTask::process()
     LOG_WARN("failed to try remove unneeded tablets", K(ret), KPC(ctx_));
   } else if (OB_FAIL(ls_online_())) {
     LOG_WARN("failed to start realy log", K(ret), K(*ctx_));
-  } else if (FALSE_IT(is_ls_online_success = true)) {
   } else if (OB_FAIL(build_tablet_group_info_())) {
     LOG_WARN("failed to build tablet group info", K(ret), KPC(ctx_));
   } else {
@@ -2966,7 +2964,12 @@ int ObDataTabletsMigrationTask::process()
 
   if (OB_FAIL(ret)) {
     int tmp_ret = OB_SUCCESS;
-    const bool allow_retry = !is_ls_online_success;
+    bool allow_retry = true;
+    if (OB_SUCCESS != (tmp_ret = try_offline_ls_())) {
+      LOG_WARN("failed to try offline ls", K(tmp_ret));
+    } else if (FALSE_IT(allow_retry = OB_SUCCESS == tmp_ret)) {
+    }
+
     if (OB_SUCCESS != (tmp_ret = ObStorageHADagUtils::deal_with_fo(ret, this->get_dag(), allow_retry))) {
       LOG_WARN("failed to deal with fo", K(ret), K(tmp_ret), K(*ctx_));
     }
@@ -3303,6 +3306,25 @@ int ObDataTabletsMigrationTask::try_remove_unneeded_tablets_()
   return ret;
 }
 
+int ObDataTabletsMigrationTask::try_offline_ls_()
+{
+  int ret = OB_SUCCESS;
+  ObLSHandle ls_handle;
+  ObLS *ls = nullptr;
+
+  if (!is_inited_) {
+    ret = OB_NOT_INIT;
+    LOG_WARN("start migration task do not init", K(ret));
+  } else if (OB_FAIL(ObStorageHADagUtils::get_ls(ctx_->arg_.ls_id_, ls_handle))) {
+    LOG_WARN("failed to get ls", K(ret), KPC(ctx_));
+  } else if (OB_ISNULL(ls = ls_handle.get_ls())) {
+    ret = OB_ERR_UNEXPECTED;
+    LOG_WARN("ls should not be NULL", K(ret), KPC(ctx_));
+  } else if (OB_FAIL(ls->offline())) {
+    LOG_WARN("failed to offline ls", K(ret), KPC(ctx_));
+  }
+  return ret;
+}
 
 int ObDataTabletsMigrationTask::record_server_event_()
 {

src/storage/high_availability/ob_ls_migration.h

@@ -418,6 +418,7 @@ private:
   int build_tablet_group_info_();
   int generate_tablet_group_dag_();
   int try_remove_unneeded_tablets_();
+  int try_offline_ls_();
   int record_server_event_();
 
 private:

src/storage/high_availability/ob_storage_ha_macro_block_writer.cpp

@@ -69,10 +69,15 @@ int ObStorageHAMacroBlockWriter::check_macro_block_(
     ObMacroBlockCheckLevel check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_MAX;
     switch (migrate_verify_level) {
       case 0:
-        check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_PHYSICAL;
+        check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_NONE;
         break;
       case 1:
-        check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_LOGICAL;
+        check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_PHYSICAL;
+        break;
+      case 2:
+        //check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_LOGICAL;
+        //Here using logical has a bug.
+        check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_PHYSICAL;
         break;
       default:
         check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_MAX;
@@ -80,10 +85,9 @@ int ObStorageHAMacroBlockWriter::check_macro_block_(
         STORAGE_LOG(WARN, "invalid check level", K(ret), K(migrate_verify_level));
         break;
     }
-    //Here using logical has a bug.
-    check_level = ObMacroBlockCheckLevel::CHECK_LEVEL_PHYSICAL;
 
-    if (OB_FAIL(macro_checker_.check(data.data(), data.length(), check_level))) {
+    if (OB_FAIL(ret)) {
+    } else if (OB_FAIL(macro_checker_.check(data.data(), data.length(), check_level))) {
       STORAGE_LOG(ERROR, "failed to check macro block", K(ret), K(data), K(check_level));
     }
   }