From cc7aa7132af562b35e91d95c8ff304dbcfd6cb7e Mon Sep 17 00:00:00 2001
From: hiddenbomb <bwgui203@gmail.com>
Date: Mon, 28 Aug 2023 08:47:40 +0000
Subject: [PATCH] set errno to OB_EAGAIN if replay get tablet interface returns
 OB_TIMEOUT

---
 .../ob_tablet_replay_executor.cpp             | 21 ++++++++++++++-----
 src/storage/ls/ob_ls_storage_clog_handler.cpp | 14 ++++++-------
 src/storage/ob_storage_clog_recorder.cpp      | 10 ++++++++-
 3 files changed, 32 insertions(+), 13 deletions(-)

diff --git a/src/logservice/replayservice/ob_tablet_replay_executor.cpp b/src/logservice/replayservice/ob_tablet_replay_executor.cpp
index 32cf96c777..4c46a10699 100644
--- a/src/logservice/replayservice/ob_tablet_replay_executor.cpp
+++ b/src/logservice/replayservice/ob_tablet_replay_executor.cpp
@@ -128,13 +128,24 @@ int ObTabletReplayExecutor::replay_get_tablet_(
   } else if (OB_ISNULL(ls = ls_handle.get_ls())) {
     ret = OB_ERR_UNEXPECTED;
     CLOG_LOG(WARN, "log stream should not be NULL", KR(ret), K(scn));
-  } else if (is_replay_update_tablet_status_()) {
-    if (OB_FAIL(ls->replay_get_tablet_no_check(tablet_id, scn, tablet_handle))) {
-      CLOG_LOG(WARN, "replay get table failed", KR(ret), "ls_id", ls->get_ls_id());
+  } else {
+    const share::ObLSID &ls_id = ls->get_ls_id();
+    if (is_replay_update_tablet_status_()) {
+      if (OB_FAIL(ls->replay_get_tablet_no_check(tablet_id, scn, tablet_handle))) {
+        CLOG_LOG(WARN, "replay get table failed", KR(ret), K(ls_id), K(tablet_id));
+      }
+    } else if (OB_FAIL(ls->replay_get_tablet(tablet_id, scn, tablet_handle))) {
+      CLOG_LOG(WARN, "replay get table failed", KR(ret), K(ls_id), K(tablet_id));
+    }
+
+    if (OB_FAIL(ret)) {
+      if (OB_TIMEOUT == ret) {
+        ret = OB_EAGAIN;
+        CLOG_LOG(WARN, "retry get tablet for timeout error", KR(ret), K(ls_id), K(tablet_id));
+      }
     }
-  } else if (OB_FAIL(ls->replay_get_tablet(tablet_id, scn, tablet_handle))) {
-    CLOG_LOG(WARN, "replay get table failed", KR(ret), "ls_id", ls->get_ls_id());
   }
+
   return ret;
 }
 
diff --git a/src/storage/ls/ob_ls_storage_clog_handler.cpp b/src/storage/ls/ob_ls_storage_clog_handler.cpp
index 703238505a..768d15f7ca 100644
--- a/src/storage/ls/ob_ls_storage_clog_handler.cpp
+++ b/src/storage/ls/ob_ls_storage_clog_handler.cpp
@@ -114,18 +114,18 @@ int ObMediumCompactionClogHandler::inner_replay(
     LOG_WARN("fail to deserialize tablet id", K(ret), K(buffer_size), K(pos), K(tablet_id));
   } else if (OB_FAIL(ls_->replay_get_tablet(tablet_id, scn, handle))) {
     if (OB_OBSOLETE_CLOG_NEED_SKIP == ret) {
-      LOG_INFO("clog is obsolete, should skip replay", K(ret), K(tablet_id));
+      LOG_INFO("clog is obsolete, should skip replay", K(ret), K(tablet_id), K(scn));
       ret = OB_SUCCESS;
+    } else if (OB_TIMEOUT == ret) {
+      ret = OB_EAGAIN;
+      LOG_INFO("retry get tablet for timeout error", K(ret), K(tablet_id), K(scn));
     } else {
-      LOG_WARN("failed to get tablet", K(ret), K(tablet_id));
+      LOG_WARN("failed to get tablet", K(ret), K(tablet_id), K(scn));
     }
   } else if (OB_FAIL(handle.get_obj()->replay_medium_compaction_clog(scn, buffer, buffer_size, new_pos))) {
-    LOG_WARN("failed to replay medium compaction clog", K(ret), K(tablet_id), K(buffer_size), K(new_pos));
-  }
-  if (OB_TIMEOUT == ret) {
-    LOG_INFO("replace timeout errno", KR(ret), K(scn), K(tablet_id));
-    ret = OB_EAGAIN;
+    LOG_WARN("failed to replay medium compaction clog", K(ret), K(tablet_id), K(scn), K(buffer_size), K(new_pos));
   }
+
   return ret;
 }
 
diff --git a/src/storage/ob_storage_clog_recorder.cpp b/src/storage/ob_storage_clog_recorder.cpp
index eac2d8aac0..3c17e351f6 100644
--- a/src/storage/ob_storage_clog_recorder.cpp
+++ b/src/storage/ob_storage_clog_recorder.cpp
@@ -311,7 +311,15 @@ int ObIStorageClogRecorder::replay_get_tablet_handle(
   if (OB_FAIL(MTL(ObLSService *)->get_ls(ls_id, ls_handle, ObLSGetMod::STORAGE_MOD))) {
     LOG_WARN("failed to get log stream", K(ret), K(ls_id));
   } else if (OB_FAIL(ls_handle.get_ls()->replay_get_tablet(tablet_id, scn, tablet_handle))) {
-    LOG_WARN("failed to get tablet", K(ret), K(ls_id), K(tablet_id), K(scn));
+    if (OB_OBSOLETE_CLOG_NEED_SKIP == ret) {
+      LOG_INFO("clog is obsolete, should skip replay", K(ret), K(ls_id), K(tablet_id), K(scn));
+      ret = OB_SUCCESS;
+    } else if (OB_TIMEOUT == ret) {
+      ret = OB_EAGAIN;
+      LOG_INFO("retry get tablet for timeout error", K(ret), K(ls_id), K(tablet_id), K(scn));
+    } else {
+      LOG_WARN("failed to get tablet", K(ret), K(ls_id), K(tablet_id), K(scn));
+    }
   }
   return ret;
 }