Fix px infinite retry loop in PL. && Clean up interrupt status at the beginning of the thread

src/observer/ob_inner_sql_connection.cpp

@@ -474,6 +474,13 @@ int ObInnerSQLConnection::process_retry(
     THIS_WORKER.sched_wait();
     usleep(static_cast<unsigned int>(sleep_time_us));
     THIS_WORKER.sched_run();
+  } else if (repeatable_stmt && is_scheduler_thread_not_enough_err(last_ret)) {
+    const int64_t sleep_time_us = 10 * 1000; // 10ms
+    need_retry = true;
+    LOG_WARN("scheduler thread not enough, need retry", K(ret), K(last_ret), K(retry_cnt));
+    THIS_WORKER.sched_wait();
+    usleep(static_cast<unsigned int>(sleep_time_us));
+    THIS_WORKER.sched_run();
   }
   if (get_session().is_nested_session()) {
     /**

src/observer/omt/ob_tenant.cpp

@@ -20,6 +20,7 @@
 #include "lib/stat/ob_session_stat.h"
 #include "share/config/ob_server_config.h"
 #include "sql/engine/px/ob_px_admission.h"
+#include "share/interrupt/ob_global_interrupt_call.h"
 #include "ob_th_worker.h"
 #include "ob_worker_pool.h"
 #include "ob_multi_tenant.h"
@@ -133,8 +134,9 @@ void ObPxPool::run1()
   if (OB_LIKELY(nullptr != pm)) {
     pm->set_tenant_ctx(tenant_id_, common::ObCtxIds::WORK_AREA);
   }
+  CLEAR_INTERRUPTABLE();
   ObCgroupCtrl* cgroup_ctrl = GCTX.cgroup_ctrl_;
-  LOG_INFO("XXXXX: run px pool", K(group_id_), K(tenant_id_));
+  LOG_INFO("run px pool", K(group_id_), K(tenant_id_));
   if (nullptr != cgroup_ctrl && OB_LIKELY(cgroup_ctrl->is_valid())) {
     pid_t pid = static_cast<pid_t>(syscall(__NR_gettid));
     cgroup_ctrl->add_thread_to_cgroup(tenant_id_, group_id_, pid);

src/observer/omt/ob_th_worker.cpp

@@ -392,6 +392,7 @@ void ObThWorker::worker(int64_t& tenant_id, int64_t& req_recv_timestamp, int32_t
               ret = pm->set_tenant_ctx(tenant_->id(), ObCtxIds::DEFAULT_CTX_ID);
             }
           }
+          CLEAR_INTERRUPTABLE();
           set_th_worker_thread_name(tenant_->id());
           lib::ContextTLOptGuard guard(true);
           lib::ContextParam param;

src/share/interrupt/ob_global_interrupt_call.cpp

@@ -97,6 +97,14 @@ ObGlobalInterruptManager* ObGlobalInterruptManager::getInstance()
   return instance_;
 }
 
+void ObInterruptChecker::clear_interrupt_status()
+{
+  if (ref_count_ > 0) {
+    LIB_LOG(ERROR, "invlid interrupt ref count");
+  }
+  interrupted_ = false;
+}
+
 int ObGlobalInterruptManager::init(const common::ObAddr& local, ObInterruptRpcProxy* rpc_proxy)
 {
   int ret = OB_SUCCESS;

src/share/interrupt/ob_global_interrupt_call.h

@@ -124,6 +124,7 @@ public:
   void clear_status();
 
   void interrupt(ObInterruptCode& interrupt_code);
+  void clear_interrupt_status();
 
 private:
   /*
@@ -339,6 +340,15 @@ OB_INLINE void UNSET_INTERRUPTABLE(const ObInterruptibleTaskID& tid)
   }
 }
 
+OB_INLINE void CLEAR_INTERRUPTABLE()
+{
+  if (OB_ISNULL(get_checker())) {
+    LIB_LOG(ERROR, "interrupt checker may not be set correctly");
+  } else {
+    get_checker()->clear_interrupt_status();
+  }
+}
+
 } // end namespace common
 } // end namespace oceanbase
 

src/sql/engine/px/ob_px_admission.cpp

@@ -70,8 +70,7 @@ int ObPxAdmission::enter_query_admission(
       LOG_WARN("fail check query status", K(ret));
     } else if (!ObPxAdmission::admit(req_worker_count, admit_worker_count)) {
       plan.inc_delayed_px_querys();
-      THIS_WORKER.set_retry_flag();
+      ret = OB_ERR_SCHEDULER_THREAD_NOT_ENOUGH;
-      ret = OB_EAGAIN;
       LOG_INFO("It's a px query, out of px worker resource, "
                "need delay, do not need disconnect",
           K(req_worker_count),

src/sql/ob_result_set.cpp

@@ -101,12 +101,7 @@ OB_INLINE int ObResultSet::open_plan()
       if (OB_FAIL(ObPxAdmission::enter_query_admission(
               my_session_, get_exec_context(), *get_physical_plan(), worker_count_))) {
         // query is not admitted to run
-        if (OB_EAGAIN == ret) {
-          ret = OB_ERR_SCHEDULER_THREAD_NOT_ENOUGH;
         LOG_DEBUG("Query is not admitted to run, try again", K(ret));
-        } else {
-          LOG_WARN("Fail to get admission to use px", K(ret));
-        }
       } else if (THIS_WORKER.is_timeout()) {
         ret = OB_TIMEOUT;
         LOG_WARN("query is timeout",
@@ -1108,7 +1103,6 @@ void ObResultSet::refresh_location_cache(ObTaskExecutorCtx &task_exec_ctx, bool
   }
 }
 
-// obmp_query中重试整个SQL之前，可能需要调用本接口来刷新Location，以避免总是发给了错误的服务器
 int ObResultSet::refresh_location_cache(bool is_nonblock)
 {
   return ObTaskExecutorCtxUtil::refresh_location_cache(get_exec_context().get_task_exec_ctx(), is_nonblock);