From ae446e60cfa0b44cd158a5b8fe4cd259e8eae203 Mon Sep 17 00:00:00 2001
From: godyangfight <godyangfight@gmail.com>
Date: Wed, 23 Mar 2022 14:03:00 +0800
Subject: [PATCH] 22x Fix remote update pg backup task timeout bug

---
 .../backup/ob_pg_backup_task_updater.cpp      | 30 +++++++++++++++++--
 src/share/backup/ob_pg_backup_task_updater.h  |  5 +++-
 2 files changed, 31 insertions(+), 4 deletions(-)

diff --git a/src/share/backup/ob_pg_backup_task_updater.cpp b/src/share/backup/ob_pg_backup_task_updater.cpp
index 9b2aca2c33..c1636c9369 100644
--- a/src/share/backup/ob_pg_backup_task_updater.cpp
+++ b/src/share/backup/ob_pg_backup_task_updater.cpp
@@ -47,9 +47,11 @@ int ObPGBackupTaskUpdater::update_pg_backup_task_status(
   } else {
     while (OB_SUCC(ret) && report_idx < pkeys.count()) {
       ObMySQLTransaction trans;
+      ObTimeoutCtx timeout_ctx;
       const int64_t remain_cnt = pkeys.count() - report_idx;
       int64_t cur_batch_cnt = remain_cnt < MAX_BATCH_COUNT ? remain_cnt : MAX_BATCH_COUNT;
-      if (OB_FAIL(trans.start(sql_proxy_))) {
+
+      if (OB_FAIL(start_trans_(timeout_ctx, trans))) {
         LOG_WARN("failed to start trans", K(ret));
       } else {
         for (int64_t i = 0; OB_SUCC(ret) && i < cur_batch_cnt; ++i) {
@@ -93,9 +95,11 @@ int ObPGBackupTaskUpdater::update_pg_task_info(const common::ObIArray<common::Ob
   } else {
     while (OB_SUCC(ret) && report_idx < pkeys.count()) {
       ObMySQLTransaction trans;
+      ObTimeoutCtx timeout_ctx;
       const int64_t remain_cnt = pkeys.count() - report_idx;
       int64_t cur_batch_cnt = remain_cnt < MAX_BATCH_COUNT ? remain_cnt : MAX_BATCH_COUNT;
-      if (OB_FAIL(trans.start(sql_proxy_))) {
+
+      if (OB_FAIL(start_trans_(timeout_ctx, trans))) {
         LOG_WARN("failed to start trans", K(ret));
       } else {
         for (int64_t i = 0; OB_SUCC(ret) && i < cur_batch_cnt; ++i) {
@@ -397,9 +401,11 @@ int ObPGBackupTaskUpdater::update_status_and_result(const common::ObIArray<commo
   } else {
     while (OB_SUCC(ret) && report_idx < pkeys.count()) {
       ObMySQLTransaction trans;
+      ObTimeoutCtx timeout_ctx;
       const int64_t remain_cnt = pkeys.count() - report_idx;
       int64_t cur_batch_cnt = remain_cnt < MAX_BATCH_COUNT ? remain_cnt : MAX_BATCH_COUNT;
-      if (OB_FAIL(trans.start(sql_proxy_))) {
+
+      if (OB_FAIL(start_trans_(timeout_ctx, trans))) {
         LOG_WARN("failed to start trans", K(ret));
       } else {
         for (int64_t i = 0; OB_SUCC(ret) && i < cur_batch_cnt; ++i) {
@@ -446,3 +452,21 @@ int ObPGBackupTaskUpdater::cancel_pending_tasks(
   }
   return ret;
 }
+
+int ObPGBackupTaskUpdater::start_trans_(ObTimeoutCtx &timeout_ctx, ObMySQLTransaction &trans)
+{
+  int ret = OB_SUCCESS;
+  const int64_t MAX_EXECUTE_TIMEOUT_US = 30L * 1000 * 1000;  // 30s
+  int64_t stmt_timeout = MAX_EXECUTE_TIMEOUT_US;
+  if (!is_inited_) {
+    ret = OB_NOT_INIT;
+    LOG_WARN("pg backup task updater do not init", K(ret));
+  } else if (OB_FAIL(timeout_ctx.set_trx_timeout_us(stmt_timeout))) {
+    LOG_WARN("fail to set trx timeout", K(ret), K(stmt_timeout));
+  } else if (OB_FAIL(timeout_ctx.set_timeout(stmt_timeout))) {
+    LOG_WARN("set timeout context failed", K(ret));
+  } else if (OB_FAIL(trans.start(sql_proxy_))) {
+    LOG_WARN("failed to start trans", K(ret));
+  }
+  return ret;
+}
diff --git a/src/share/backup/ob_pg_backup_task_updater.h b/src/share/backup/ob_pg_backup_task_updater.h
index 8c593de320..b6385e0803 100644
--- a/src/share/backup/ob_pg_backup_task_updater.h
+++ b/src/share/backup/ob_pg_backup_task_updater.h
@@ -58,7 +58,10 @@ public:
   int cancel_pending_tasks(const uint64_t tenant_id, const int64_t incarnation, const int64_t backup_set_id);
 
 private:
-  static const int64_t MAX_BATCH_COUNT = 1024;
+  int start_trans_(ObTimeoutCtx &timeout_ctx, ObMySQLTransaction &trans);
+
+private:
+  static const int64_t MAX_BATCH_COUNT = 128;
   bool is_inited_;
   common::ObISQLClient* sql_proxy_;
   DISALLOW_COPY_AND_ASSIGN(ObPGBackupTaskUpdater);
-- 
GitLab