[refer #PGPRO-4197] Add request ID

knizhnik · knizhnik · commit 9ad9d93516fe · 2020-12-01T14:44:11.000+03:00
diff --git a/pg_query_state.c b/pg_query_state.c
@@ -952,6 +952,7 @@ GetRemoteBackendQueryStates(PGPROC *leader,
 	shm_mq_result	 mq_receive_result;
 	shm_mq_msg		*msg;
 	Size			 len;
+	static int       reqid = 0;
 
 	Assert(QueryStatePollReason != INVALID_PROCSIGNAL);
 	Assert(mq);
@@ -963,6 +964,7 @@ GetRemoteBackendQueryStates(PGPROC *leader,
 	params->buffers = buffers;
 	params->triggers = triggers;
 	params->format = format;
+	params->reqid = ++reqid;
 	pg_write_barrier();
 
 	/* initialize message queue that will transfer query states */
@@ -1002,9 +1004,13 @@ GetRemoteBackendQueryStates(PGPROC *leader,
 
 	/* extract query state from leader process */
 	mqh = shm_mq_attach(mq, NULL, NULL);
+	elog(DEBUG1, "Wait response from leader %d", leader->pid);
 	mq_receive_result = shm_mq_receive(mqh, &len, (void **) &msg, false);
 	if (mq_receive_result != SHM_MQ_SUCCESS)
 		goto mq_error;
+	if (msg->reqid != reqid)
+		goto mq_error;
+
 	Assert(len == msg->length);
 	result = lappend(result, copy_msg(msg));
 #if PG_VERSION_NUM < 100000
@@ -1021,6 +1027,7 @@ GetRemoteBackendQueryStates(PGPROC *leader,
 		PGPROC 			*proc = (PGPROC *) lfirst(iter);
 
 		/* prepare message queue to transfer data */
+		elog(DEBUG1, "Wait response from worker %d", proc->pid);
 		mq = shm_mq_create(mq, QUEUE_SIZE);
 		shm_mq_set_sender(mq, proc);
 		shm_mq_set_receiver(mq, MyProc);	/* this function notifies the
@@ -1034,9 +1041,12 @@ GetRemoteBackendQueryStates(PGPROC *leader,
 														(void **) &msg,
 														MAX_RCV_TIMEOUT);
 		if (mq_receive_result != SHM_MQ_SUCCESS)
+		{
 			/* counterpart is died, not consider it */
-			continue;
-
+			goto mq_error;
+		}
+		if (msg->reqid != reqid)
+			goto mq_error;
 		Assert(len == msg->length);
 
 		/* aggregate result data */
@@ -1054,6 +1064,11 @@ GetRemoteBackendQueryStates(PGPROC *leader,
 	ereport(ERROR, (errcode(ERRCODE_INTERNAL_ERROR),
 				errmsg("invalid send signal")));
 mq_error:
+#if PG_VERSION_NUM < 100000
+	shm_mq_detach(mq);
+#else
+	shm_mq_detach(mqh);
+#endif
 	ereport(ERROR, (errcode(ERRCODE_INTERNAL_ERROR),
 				errmsg("error in message queue data transmitting")));
 }
diff --git a/pg_query_state.h b/pg_query_state.h
@@ -42,6 +42,7 @@ typedef enum
  */
 typedef struct
 {
+	int     reqid;
 	int		length;							/* size of message record, for sanity check */
 	PGPROC	*proc;
 	PG_QS_RequestResult	result_code;
@@ -56,6 +57,7 @@ typedef struct
 /* pg_query_state arguments */
 typedef struct
 {
+	int     reqid;
 	bool 	verbose;
 	bool	costs;
 	bool	timing;
diff --git a/signal_handler.c b/signal_handler.c
@@ -160,6 +160,7 @@ SendQueryState(void)
 	instr_time	start_time;
 	instr_time	cur_time;
 	int64 		delay = MAX_SND_TIMEOUT;
+	int         reqid = params->reqid;
 
 	INSTR_TIME_SET_CURRENT(start_time);
 
@@ -189,21 +190,21 @@ SendQueryState(void)
 		CHECK_FOR_INTERRUPTS();
 		ResetLatch(MyLatch);
 	}
-
+	elog(DEBUG1, "Worker %d receives pg_query_state request from %d", shm_mq_get_sender(mq)->pid, shm_mq_get_receiver(mq)->pid);
 	mqh = shm_mq_attach(mq, NULL, NULL);
 
 	/* check if module is enabled */
 	if (!pg_qs_enable)
 	{
-		shm_mq_msg msg = { BASE_SIZEOF_SHM_MQ_MSG, MyProc, STAT_DISABLED };
+		shm_mq_msg msg = { reqid, BASE_SIZEOF_SHM_MQ_MSG, MyProc, STAT_DISABLED };
 
 		shm_mq_send(mqh, msg.length, &msg, false);
 	}
 
 	/* check if backend doesn't execute any query */
 	else if (list_length(QueryDescStack) == 0)
 	{
-		shm_mq_msg msg = { BASE_SIZEOF_SHM_MQ_MSG, MyProc, QUERY_NOT_RUNNING };
+		shm_mq_msg msg = { reqid, BASE_SIZEOF_SHM_MQ_MSG, MyProc, QUERY_NOT_RUNNING };
 
 		shm_mq_send(mqh, msg.length, &msg, false);
 	}
@@ -215,6 +216,7 @@ SendQueryState(void)
 		int				msglen = sizeof(shm_mq_msg) + serialized_stack_length(qs_stack);
 		shm_mq_msg		*msg = palloc(msglen);
 
+		msg->reqid = reqid;
 		msg->length = msglen;
 		msg->proc = MyProc;
 		msg->result_code = QS_RETURNED;
@@ -229,5 +231,6 @@ SendQueryState(void)
 		serialize_stack(msg->stack, qs_stack);
 		shm_mq_send(mqh, msglen, msg, false);
 	}
+	elog(DEBUG1, "Worker %d sends response for pg_query_state to %d", shm_mq_get_sender(mq)->pid, shm_mq_get_receiver(mq)->pid);
 	DetachPeer();
 }

Original file line number	Diff line number	Diff line change
`@@ -160,6 +160,7 @@ SendQueryState(void)`
`160`	`160`	`instr_time start_time;`
`161`	`161`	`instr_time cur_time;`
`162`	`162`	`int64 delay = MAX_SND_TIMEOUT;`
	`163`	`+ int reqid = params->reqid;`
`163`	`164`
`164`	`165`	`INSTR_TIME_SET_CURRENT(start_time);`
`165`	`166`
`@@ -189,21 +190,21 @@ SendQueryState(void)`
`189`	`190`	`CHECK_FOR_INTERRUPTS();`
`190`	`191`	`ResetLatch(MyLatch);`
`191`	`192`	`}`
`192`		`-`
	`193`	`+ elog(DEBUG1, "Worker %d receives pg_query_state request from %d", shm_mq_get_sender(mq)->pid, shm_mq_get_receiver(mq)->pid);`
`193`	`194`	`mqh = shm_mq_attach(mq, NULL, NULL);`
`194`	`195`
`195`	`196`	`/* check if module is enabled */`
`196`	`197`	`if (!pg_qs_enable)`
`197`	`198`	`{`
`198`		`- shm_mq_msg msg = { BASE_SIZEOF_SHM_MQ_MSG, MyProc, STAT_DISABLED };`
	`199`	`+ shm_mq_msg msg = { reqid, BASE_SIZEOF_SHM_MQ_MSG, MyProc, STAT_DISABLED };`
`199`	`200`
`200`	`201`	`shm_mq_send(mqh, msg.length, &msg, false);`
`201`	`202`	`}`
`202`	`203`
`203`	`204`	`/* check if backend doesn't execute any query */`
`204`	`205`	`else if (list_length(QueryDescStack) == 0)`
`205`	`206`	`{`
`206`		`- shm_mq_msg msg = { BASE_SIZEOF_SHM_MQ_MSG, MyProc, QUERY_NOT_RUNNING };`
	`207`	`+ shm_mq_msg msg = { reqid, BASE_SIZEOF_SHM_MQ_MSG, MyProc, QUERY_NOT_RUNNING };`
`207`	`208`
`208`	`209`	`shm_mq_send(mqh, msg.length, &msg, false);`
`209`	`210`	`}`
`@@ -215,6 +216,7 @@ SendQueryState(void)`
`215`	`216`	`int msglen = sizeof(shm_mq_msg) + serialized_stack_length(qs_stack);`
`216`	`217`	`shm_mq_msg *msg = palloc(msglen);`
`217`	`218`
	`219`	`+ msg->reqid = reqid;`
`218`	`220`	`msg->length = msglen;`
`219`	`221`	`msg->proc = MyProc;`
`220`	`222`	`msg->result_code = QS_RETURNED;`
`@@ -229,5 +231,6 @@ SendQueryState(void)`
`229`	`231`	`serialize_stack(msg->stack, qs_stack);`
`230`	`232`	`shm_mq_send(mqh, msglen, msg, false);`
`231`	`233`	`}`
	`234`	`+ elog(DEBUG1, "Worker %d sends response for pg_query_state to %d", shm_mq_get_sender(mq)->pid, shm_mq_get_receiver(mq)->pid);`
`232`	`235`	`DetachPeer();`
`233`	`236`	`}`