Fix reace condition in tw_pahse.c

knizhnik · knizhnik · commit 127dc15dd279 · 2016-12-21T15:13:12.000+03:00
diff --git a/contrib/mmts/arbiter.c b/contrib/mmts/arbiter.c
@@ -371,7 +371,9 @@ static void MtmSendHeartbeat()
 				} else {
 					/* Connectivity mask can be cleared by MtmWatchdog: in this case sockets[i] >= 0 */
 					if (BIT_CHECK(Mtm->connectivityMask, i)) { 
-						MtmDisconnect(i);
+						close(sockets[i]);
+						sockets[i] = -1;
+						MtmReconnectNode(i+1);
 						//MtmOnNodeConnect(i+1);
 					}
 					MTM_LOG4("Send heartbeat to node %d with timestamp %ld", i+1, now);    
diff --git a/contrib/mmts/multimaster.c b/contrib/mmts/multimaster.c
@@ -2021,6 +2021,13 @@ void MtmOnNodeConnect(int nodeId)
 	MtmUnlock();
 }
 
+void MtmReconnectNode(int nodeId)
+{
+	MtmLock(LW_EXCLUSIVE);	
+	BIT_SET(Mtm->reconnectMask, nodeId-1);
+	MtmUnlock();
+}
+
 
 
 /*
@@ -3301,7 +3308,7 @@ bool MtmFilterTransaction(char* record, int size)
 	}
 
 	if (duplicate) {
-		MTM_LOG2("Ignore transaction %s from node %d flags=%x, our restartLSN for node: %lx,restart_lsn = (origin node %d == MtmReplicationNodeId %d) ? end_lsn=%lx, origin_lsn=%lx", 
+		MTM_LOG1("Ignore transaction %s from node %d flags=%x, our restartLSN for node: %lx,restart_lsn = (origin node %d == MtmReplicationNodeId %d) ? end_lsn=%lx, origin_lsn=%lx", 
 				 gid, replication_node, flags, Mtm->nodes[origin_node-1].restartLSN, origin_node, MtmReplicationNodeId, end_lsn, origin_lsn);
 	} else {
 		MTM_LOG2("Apply transaction %s from node %d lsn %lx, flags=%x, origin node %d, original lsn=%lx, current lsn=%lx", 
diff --git a/contrib/mmts/multimaster.h b/contrib/mmts/multimaster.h
@@ -358,6 +358,7 @@ extern void  MtmUnlock(void);
 extern void  MtmLockNode(int nodeId, LWLockMode mode);
 extern void  MtmUnlockNode(int nodeId);
 extern void  MtmDropNode(int nodeId, bool dropSlot);
+extern void  MtmReconnectNode(int nodeId);
 extern void  MtmRecoverNode(int nodeId);
 extern void  MtmOnNodeDisconnect(int nodeId);
 extern void  MtmOnNodeConnect(int nodeId);
diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
@@ -627,9 +627,10 @@ LockGXact(const char *gid, Oid user)
 		before_shmem_exit(AtProcExit_Twophase, 0);
 		twophaseExitRegistered = true;
 	}
-
-	LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
+	MyLockedGxact = NULL;
 	i = string_hash(gid, 0) % max_prepared_xacts;
+  Retry:
+	LWLockAcquire(TwoPhaseStateLock, LW_EXCLUSIVE);
 	for (gxact = TwoPhaseState->hashTable[i]; gxact != NULL; gxact = gxact->next)
 	{
 		if (strcmp(gxact->gid, gid) == 0)
@@ -639,19 +640,25 @@ LockGXact(const char *gid, Oid user)
 			/* Lock gxact. We have to release TwoPhaseStateLock LW-Lock to avoid deadlock */
 
 			LWLockRelease(TwoPhaseStateLock);
-			SpinLockAcquire(&gxact->spinlock);
+
+			if (MyLockedGxact != gxact) {
+				if (MyLockedGxact != NULL) { 
+					SpinLockRelease(&MyLockedGxact->spinlock);
+				}
+				MyLockedGxact = gxact;
+				SpinLockAcquire(&gxact->spinlock);
+				goto Retry;
+			}
 
 			/* Ignore not-yet-valid GIDs */
 			if (!gxact->valid) {
-				SpinLockRelease(&gxact->spinlock);
 				ereport(ERROR,
 						(errcode(ERRCODE_OBJECT_NOT_IN_PREREQUISITE_STATE),
 						 errmsg("prepared transaction with identifier \"%s\" is not valid",
 								gid)));
 			}
 			
 			if (user != gxact->owner && !superuser_arg(user)) {
-				SpinLockRelease(&gxact->spinlock);
 				ereport(ERROR,
 						(errcode(ERRCODE_INSUFFICIENT_PRIVILEGE),
 						 errmsg("permission denied to finish prepared transaction"),
@@ -665,7 +672,6 @@ LockGXact(const char *gid, Oid user)
 			 * someone gets motivated to make it work.
 			 */
 			if (MyDatabaseId != proc->databaseId) {
-				SpinLockRelease(&gxact->spinlock);
 				ereport(ERROR,
 						(errcode(ERRCODE_FEATURE_NOT_SUPPORTED),
 						 errmsg("prepared transaction belongs to another database"),
@@ -676,13 +682,16 @@ LockGXact(const char *gid, Oid user)
 			/* OK for me to lock it */
 			Assert(gxact->locking_pid < 0);
 			gxact->locking_pid = MyProcPid;
-			MyLockedGxact = gxact;
 
 			return gxact;
 		}		
 	}
 
 	LWLockRelease(TwoPhaseStateLock);
+	if (MyLockedGxact != NULL) { 
+		SpinLockRelease(&MyLockedGxact->spinlock);
+		MyLockedGxact = NULL;
+	}
 
 	ereport(ERROR,
 			(errcode(ERRCODE_UNDEFINED_OBJECT),
