fix shutdown failure, reduce log garbage

Vladimir Ershov · Vladimir Ershov · commit 6ad88b93ada9 · 2016-11-15T11:19:57.000+03:00
diff --git a/src/pgpro_scheduler.c b/src/pgpro_scheduler.c
@@ -80,14 +80,12 @@ void reload_db_role_config(char *dbname)
 	StartTransactionCommand();
 	databaseid = get_database_oid(dbname, false);
 
-	elog(LOG, "start read");
 	relsetting = heap_open(DbRoleSettingRelationId, AccessShareLock);
 	snapshot = RegisterSnapshot(GetCatalogSnapshot(DbRoleSettingRelationId));
 	ApplySetting(snapshot, databaseid, InvalidOid, relsetting, PGC_S_DATABASE);
 	UnregisterSnapshot(snapshot);
 	heap_close(relsetting, AccessShareLock);
 	CommitTransactionCommand();
-	elog(LOG, "got read");
 }
 
 TimestampTz timestamp_add_seconds(TimestampTz to, int add)
diff --git a/src/sched_manager_poll.c b/src/sched_manager_poll.c
@@ -114,6 +114,8 @@ int stopAllManagers(schd_managers_poll_t *poll)
 	PGPROC *child;
 	schd_manager_share_t *shared;
 
+	if(poll->n == 0) return 0;
+
 	elog(LOG, "Scheduler stops all managers");
 
 	for(i=0; i < poll->n; i++)
@@ -136,6 +138,7 @@ int stopAllManagers(schd_managers_poll_t *poll)
 	{
 		destroyManagerRecord(poll->workers[i]);
 	}
+
 	pfree(poll->workers);
 	poll->workers = NULL;
 	poll->n = 0;
diff --git a/src/scheduler_executor.c b/src/scheduler_executor.c
@@ -147,12 +147,14 @@ void executor_worker_main(Datum arg)
 			status = SchdExecutorError;
 			if(error)
 			{
-				push_executor_error(&EE, "error on %d: %s", i+1, error);
+				push_executor_error(&EE, "error in command #%d: %s",
+															i+1, error);
 				pfree(error);
 			}
 			else
 			{
-				push_executor_error(&EE, "error on %d: code: %d", i+1, ret);
+				push_executor_error(&EE, "error in command #%d: code: %d",
+															i+1, ret);
 			}
 			ABORT_SPI_SNAP();
 			SetConfigOption("schedule.transaction_state", "failure", PGC_INTERNAL, PGC_S_SESSION);
diff --git a/src/scheduler_manager.c b/src/scheduler_manager.c
@@ -152,10 +152,9 @@ int refresh_scheduler_manager_context(scheduler_manager_ctx_t *ctx)
 	scheduler_manager_slot_t **old;
 
 	N = get_scheduler_maxworkers();
-	elog(LOG, "max-workers: %d", N);
 	if(N != ctx->slots_len)
 	{
-		elog(LOG, "Change available workers number %d => %d", ctx->slots_len, N);
+		elog(LOG, "Scheduler Manager %s: Change available workers number %d => %d", ctx->database, ctx->slots_len, N);
 	} 
 
 	if(N > ctx->slots_len)
@@ -315,8 +314,8 @@ scheduler_task_t *scheduler_get_active_tasks(scheduler_manager_ctx_t *ctx, int *
 	}
 	else if(ret != SPI_OK_SELECT)
 	{
-		elog(LOG, "Scheduler manager: %s: cannot get \"at\" tasks: error code %d",
-				MyBgworkerEntry->bgw_name, ret);
+		elog(LOG, "Scheduler manager %s: cannot get \"at\" tasks: error code %d",
+				ctx->database, ret);
 
         scheduler_manager_stop(ctx);
 		return NULL;
@@ -1048,8 +1047,8 @@ int scheduler_vanish_expired_jobs(scheduler_manager_ctx_t *ctx)
 			move_ret  = move_job_to_log(&expired[i], 0);
 			if(move_ret < 0)
 			{
-				elog(LOG, "Scheduler manager: cannot move job %d@%s:00 to log",
-								expired[i].cron_id, ts);
+				elog(LOG, "Scheduler manager %s: cannot move job %d@%s:00 to log",
+								ctx->database, expired[i].cron_id, ts);
 				ret--;
 			}
 			pfree(ts);
@@ -1235,7 +1234,7 @@ bool check_parent_stop_signal(scheduler_manager_ctx_t *ctx)
 		shared->setbyparent = false;
 		if(shared->status == SchdManagerStop)
 		{
-			elog(LOG, "Scheduler manager: receive stop signal from supervisor");
+			elog(LOG, "Scheduler manager %s: receive stop signal from supervisor", ctx->database);
 			return true;
 		}
 	}
@@ -1323,12 +1322,9 @@ void manager_worker_main(Datum arg)
 			if(rc & WL_POSTMASTER_DEATH) proc_exit(1);
 			if(got_sighup)
 			{
-			elog(LOG, "manager got sighup");
 				got_sighup = false;
 				ProcessConfigFile(PGC_SIGHUP);
-				elog(LOG, "DONE CONFIG");
 				reload_db_role_config(database);
-				elog(LOG, "DONE role");
 				refresh_scheduler_manager_context(ctx);
 				set_slots_stat_report(ctx);
 			}

Original file line number	Diff line number	Diff line change
`@@ -114,6 +114,8 @@ int stopAllManagers(schd_managers_poll_t *poll)`
`114`	`114`	`PGPROC *child;`
`115`	`115`	`schd_manager_share_t *shared;`
`116`	`116`
	`117`	`+ if(poll->n == 0) return 0;`
	`118`	`+`
`117`	`119`	`elog(LOG, "Scheduler stops all managers");`
`118`	`120`
`119`	`121`	`for(i=0; i < poll->n; i++)`
`@@ -136,6 +138,7 @@ int stopAllManagers(schd_managers_poll_t *poll)`
`136`	`138`	`{`
`137`	`139`	`destroyManagerRecord(poll->workers[i]);`
`138`	`140`	`}`
	`141`	`+`
`139`	`142`	`pfree(poll->workers);`
`140`	`143`	`poll->workers = NULL;`
`141`	`144`	`poll->n = 0;`
Original file line number	Diff line number	Diff line change
`@@ -147,12 +147,14 @@ void executor_worker_main(Datum arg)`
`147`	`147`	`status = SchdExecutorError;`
`148`	`148`	`if(error)`
`149`	`149`	`{`
`150`		`- push_executor_error(&EE, "error on %d: %s", i+1, error);`
	`150`	`+ push_executor_error(&EE, "error in command #%d: %s",`
	`151`	`+ i+1, error);`
`151`	`152`	`pfree(error);`
`152`	`153`	`}`
`153`	`154`	`else`
`154`	`155`	`{`
`155`		`- push_executor_error(&EE, "error on %d: code: %d", i+1, ret);`
	`156`	`+ push_executor_error(&EE, "error in command #%d: code: %d",`
	`157`	`+ i+1, ret);`
`156`	`158`	`}`
`157`	`159`	`ABORT_SPI_SNAP();`
`158`	`160`	`SetConfigOption("schedule.transaction_state", "failure", PGC_INTERNAL, PGC_S_SESSION);`
Original file line number	Diff line number	Diff line change
`@@ -152,10 +152,9 @@ int refresh_scheduler_manager_context(scheduler_manager_ctx_t *ctx)`
`152`	`152`	`scheduler_manager_slot_t **old;`
`153`	`153`
`154`	`154`	`N = get_scheduler_maxworkers();`
`155`		`- elog(LOG, "max-workers: %d", N);`
`156`	`155`	`if(N != ctx->slots_len)`
`157`	`156`	`{`
`158`		`- elog(LOG, "Change available workers number %d => %d", ctx->slots_len, N);`
	`157`	`+ elog(LOG, "Scheduler Manager %s: Change available workers number %d => %d", ctx->database, ctx->slots_len, N);`
`159`	`158`	`}`
`160`	`159`
`161`	`160`	`if(N > ctx->slots_len)`
`@@ -315,8 +314,8 @@ scheduler_task_t scheduler_get_active_tasks(scheduler_manager_ctx_t ctx, int *`
`315`	`314`	`}`
`316`	`315`	`else if(ret != SPI_OK_SELECT)`
`317`	`316`	`{`
`318`		`- elog(LOG, "Scheduler manager: %s: cannot get \"at\" tasks: error code %d",`
`319`		`- MyBgworkerEntry->bgw_name, ret);`
	`317`	`+ elog(LOG, "Scheduler manager %s: cannot get \"at\" tasks: error code %d",`
	`318`	`+ ctx->database, ret);`
`320`	`319`
`321`	`320`	`scheduler_manager_stop(ctx);`
`322`	`321`	`return NULL;`
`@@ -1048,8 +1047,8 @@ int scheduler_vanish_expired_jobs(scheduler_manager_ctx_t *ctx)`
`1048`	`1047`	`move_ret = move_job_to_log(&expired[i], 0);`
`1049`	`1048`	`if(move_ret < 0)`
`1050`	`1049`	`{`
`1051`		`- elog(LOG, "Scheduler manager: cannot move job %d@%s:00 to log",`
`1052`		`- expired[i].cron_id, ts);`
	`1050`	`+ elog(LOG, "Scheduler manager %s: cannot move job %d@%s:00 to log",`
	`1051`	`+ ctx->database, expired[i].cron_id, ts);`
`1053`	`1052`	`ret--;`
`1054`	`1053`	`}`
`1055`	`1054`	`pfree(ts);`
`@@ -1235,7 +1234,7 @@ bool check_parent_stop_signal(scheduler_manager_ctx_t *ctx)`
`1235`	`1234`	`shared->setbyparent = false;`
`1236`	`1235`	`if(shared->status == SchdManagerStop)`
`1237`	`1236`	`{`
`1238`		`- elog(LOG, "Scheduler manager: receive stop signal from supervisor");`
	`1237`	`+ elog(LOG, "Scheduler manager %s: receive stop signal from supervisor", ctx->database);`
`1239`	`1238`	`return true;`
`1240`	`1239`	`}`
`1241`	`1240`	`}`
`@@ -1323,12 +1322,9 @@ void manager_worker_main(Datum arg)`
`1323`	`1322`	`if(rc & WL_POSTMASTER_DEATH) proc_exit(1);`
`1324`	`1323`	`if(got_sighup)`
`1325`	`1324`	`{`
`1326`		`- elog(LOG, "manager got sighup");`
`1327`	`1325`	`got_sighup = false;`
`1328`	`1326`	`ProcessConfigFile(PGC_SIGHUP);`
`1329`		`- elog(LOG, "DONE CONFIG");`
`1330`	`1327`	`reload_db_role_config(database);`
`1331`		`- elog(LOG, "DONE role");`
`1332`	`1328`	`refresh_scheduler_manager_context(ctx);`
`1333`	`1329`	`set_slots_stat_report(ctx);`
`1334`	`1330`	`}`