Correctly handle duplicated PREPARE message

knizhnik · kelvich · commit c94bbba36fae · 2017-11-13T02:17:36.000+03:00
diff --git a/arbiter.c b/arbiter.c
@@ -258,7 +258,7 @@ static int MtmReadSocket(int sd, void* buf, int buf_size)
 {
 	int rc;
 	while ((rc = recv(sd, buf, buf_size, 0)) < 0 && errno == EINTR);			
-	if (rc < 0 && errno == EAGAIN) { 
+	if (rc <= 0 && errno == EAGAIN) { 
 		rc = MtmWaitSocket(sd, false, MtmHeartbeatSendTimeout);
 		if (rc == 1) { 
 			while ((rc = recv(sd, buf, buf_size, 0)) < 0 && errno == EINTR);			
diff --git a/multimaster.c b/multimaster.c
@@ -854,6 +854,7 @@ MtmPrePrepareTransaction(MtmCurrentTrans* x)
 	MtmTransState* ts;
 	MtmTransMap* tm;
 	TransactionId* subxids;
+	bool found;
 	MTM_TXTRACE(x, "PrePrepareTransaction Start");
 
 	if (!x->isDistributed) {
@@ -869,7 +870,7 @@ MtmPrePrepareTransaction(MtmCurrentTrans* x)
 
 	if (!IsBackgroundWorker && Mtm->status != MTM_ONLINE) { 
 		/* Do not take in account bg-workers which are performing recovery */
-		elog(ERROR, "Abort current transaction because this cluster node is in %s status", MtmNodeStatusMnem[Mtm->status]);			
+		elog(ERROR, "Abort transaction %s (%llu)  because this cluster node is in %s status", x->gid, (long64)x->xid, MtmNodeStatusMnem[Mtm->status]);			
 	}
 	if (TransactionIdIsValid(x->gtid.xid) && BIT_CHECK(Mtm->disabledNodeMask, x->gtid.node-1)) {
 		/* Coordinator of transaction is disabled: just abort transaction without any further steps */
@@ -878,6 +879,14 @@ MtmPrePrepareTransaction(MtmCurrentTrans* x)
 
 	MtmLock(LW_EXCLUSIVE);
 
+	Assert(*x->gid != '\0');
+	tm = (MtmTransMap*)hash_search(MtmGid2State, x->gid, HASH_ENTER, &found);
+	if (found && tm->status != TRANSACTION_STATUS_IN_PROGRESS) { 
+		Assert(tm->status == TRANSACTION_STATUS_ABORTED);
+		MtmUnlock();
+		elog(ERROR, "Skip already aborted transaction %s (%llu) from node %d", x->gid, (long64)x->xid, x->gtid.node);			
+	}
+
 	ts = MtmCreateTransState(x);
 	/* 
 	 * Invalid CSN prevent replication of transaction by logical replication 
@@ -898,8 +907,6 @@ MtmPrePrepareTransaction(MtmCurrentTrans* x)
 	x->isPrepared = true;
 	x->csn = ts->csn;
 	
-	Assert(*x->gid != '\0');
-	tm = (MtmTransMap*)hash_search(MtmGid2State, x->gid, HASH_ENTER, NULL);
 	tm->state = ts;
 	tm->status = TRANSACTION_STATUS_IN_PROGRESS;
 	MTM_LOG2("Prepare transaction %s", x->gid);
diff --git a/tests2/test_recovery.py b/tests2/test_recovery.py
@@ -31,6 +31,8 @@ def assertCommits(self, aggs):
         for conn_id, agg in enumerate(aggs):
             commits = commits and 'commit' in agg['transfer']['finish']
         if not commits:
+            print('No commits during aggregation interval')
+            time.sleep(100000)
             raise AssertionError('No commits during aggregation interval')
 
     def assertNoCommits(self, aggs):

Original file line number	Diff line number	Diff line change
`@@ -258,7 +258,7 @@ static int MtmReadSocket(int sd, void* buf, int buf_size)`
`258`	`258`	`{`
`259`	`259`	`int rc;`
`260`	`260`	`while ((rc = recv(sd, buf, buf_size, 0)) < 0 && errno == EINTR);`
`261`		`- if (rc < 0 && errno == EAGAIN) {`
	`261`	`+ if (rc <= 0 && errno == EAGAIN) {`
`262`	`262`	`rc = MtmWaitSocket(sd, false, MtmHeartbeatSendTimeout);`
`263`	`263`	`if (rc == 1) {`
`264`	`264`	`while ((rc = recv(sd, buf, buf_size, 0)) < 0 && errno == EINTR);`