From 160ff54fe602feace198b76e2ea92ee93be565d2 Mon Sep 17 00:00:00 2001
From: nroskill <nroskill@gmail.com>
Date: Wed, 7 Feb 2024 09:02:07 +0000
Subject: [PATCH] [CP] fix error log when kill -15

---
 deps/oblib/src/lib/thread/ob_pthread.cpp |  2 +-
 src/observer/omt/ob_tenant.cpp           | 18 ++++++++++--------
 2 files changed, 11 insertions(+), 9 deletions(-)

diff --git a/deps/oblib/src/lib/thread/ob_pthread.cpp b/deps/oblib/src/lib/thread/ob_pthread.cpp
index 265ab0a3a9..bfb88d32d4 100644
--- a/deps/oblib/src/lib/thread/ob_pthread.cpp
+++ b/deps/oblib/src/lib/thread/ob_pthread.cpp
@@ -37,7 +37,7 @@ int ob_pthread_create(void **ptr, void *(*start_routine) (void *), void *arg)
       OB_DELETE(ObPThread, SET_USE_500("PThread"), thread);
     }
   } else {
-    *ptr = thread;
+    ATOMIC_STORE(ptr, thread);
     OB_LOG(INFO, "ob_pthread_create succeed", KP(thread));
   }
   return ret;
diff --git a/src/observer/omt/ob_tenant.cpp b/src/observer/omt/ob_tenant.cpp
index 4ed2208f7b..49191b3441 100644
--- a/src/observer/omt/ob_tenant.cpp
+++ b/src/observer/omt/ob_tenant.cpp
@@ -869,7 +869,7 @@ void ObTenant::sleep_and_warn(ObTenant* tenant)
 {
   ob_usleep(10_ms);
   const int64_t ts = ObTimeUtility::current_time() - tenant->stopped_;
-  if (ts >= 3_min && TC_REACH_TIME_INTERVAL(3_min)) {
+  if (ts >= 3L * 60 * 1000 * 1000 && TC_REACH_TIME_INTERVAL(3L * 60 * 1000 * 1000)) {
     LOG_ERROR_RET(OB_SUCCESS, "tenant destructed for too long time.", K_(tenant->id), K(ts));
   }
 }
@@ -942,16 +942,18 @@ void* ObTenant::wait(void* t)
 int ObTenant::try_wait()
 {
   int ret = OB_SUCCESS;
-  if (nullptr == gc_thread_) {
-    if (has_created_) {
-      LOG_WARN("try_wait after wait successfully", K(id_), K(wait_mtl_finished_));
+  if (OB_ISNULL(ATOMIC_LOAD(&gc_thread_))) {
+    if (!ATOMIC_BCAS(&has_created_, false, true)) {
+      // try_wait should not return OB_SUCCESS here, but we returned OB_SUCCESS for safety quit in main thread.
+      // ret = OB_ERR_UNEXPECTED;
+      LOG_ERROR("try_wait again after wait successfully, there may be `kill -15`", K(id_), K(wait_mtl_finished_));
     } else {
       // it may takes too much time for killing session after remove_tenant, we should recalculate.
-      ATOMIC_STORE(&stopped_, ObTimeUtility::current_time());
+      ATOMIC_STORE(&stopped_, ObTimeUtility::current_time()); // update, it is not 0 before here.
       if (OB_FAIL(ob_pthread_create(&gc_thread_, wait, this))) {
+        ATOMIC_STORE(&has_created_, false);
         LOG_ERROR("tenant gc thread create failed", K(ret), K(errno), K(id_));
       } else {
-        has_created_ = true;
         ret = OB_EAGAIN;
         LOG_INFO("tenant pthread_create gc thread successfully", K(id_), K(gc_thread_));
       }
@@ -960,12 +962,12 @@ int ObTenant::try_wait()
     if (OB_FAIL(ob_pthread_tryjoin_np(gc_thread_))) {
       LOG_WARN("tenant pthread_tryjoin_np failed", K(errno), K(id_));
     } else {
-      gc_thread_ = nullptr; // avoid try_wait again after wait success
+      ATOMIC_STORE(&gc_thread_, nullptr); // avoid try_wait again after wait success
       LOG_INFO("tenant pthread_tryjoin_np successfully", K(id_));
     }
     const int64_t ts = ObTimeUtility::current_time() - stopped_;
     // only warn for one time in all tenant.
-    if (ts >= 3_min && REACH_TIME_INTERVAL(3_min)) {
+    if (ts >= 3L * 60 * 1000 * 1000 && REACH_TIME_INTERVAL(3L * 60 * 1000 * 1000)) {
       LOG_ERROR_RET(OB_SUCCESS, "tenant destructed for too long time.", K_(id), K(ts));
     }
   }