Let MtmRefreshClusterStatus enable concrete node

knizhnik · kelvich · commit 62697ed8ff3f · 2017-11-13T02:17:31.000+03:00
diff --git a/arbiter.c b/arbiter.c
@@ -1030,7 +1030,7 @@ static void MtmTransReceiver(Datum arg)
 			}
 			if (n == 0 && Mtm->disabledNodeMask != 0) { 
 				/* If timeout is expired and there are disabled nodes, then recheck cluster's state */
-				MtmRefreshClusterStatus(false);
+				MtmRefreshClusterStatus(false, 0);
 			}
 		}
 	}
diff --git a/multimaster.c b/multimaster.c
@@ -801,7 +801,7 @@ MtmPrePrepareTransaction(MtmCurrentTrans* x)
 		timestamp_t now = MtmGetSystemTime();		
 		if (Mtm->lastClusterStatusUpdate + MSEC_TO_USEC(MtmHeartbeatRecvTimeout) < now) { 
 			Mtm->lastClusterStatusUpdate = now;
-			MtmRefreshClusterStatus(true);
+			MtmRefreshClusterStatus(true, 0);
 		}
 		if (!IsBackgroundWorker && Mtm->status != MTM_ONLINE) { 
 			/* Do not take in account bg-workers which are performing recovery */
@@ -1461,9 +1461,9 @@ MtmBuildConnectivityMatrix(nodemask_t* matrix, bool nowait)
  * Build connectivity graph, find clique in it and extend disabledNodeMask by nodes not included in clique.
  * This function returns false if current node is excluded from cluster, true otherwise
  */
-bool MtmRefreshClusterStatus(bool nowait)
+bool MtmRefreshClusterStatus(bool nowait, int testNodeId)
 {
-	nodemask_t mask, clique, disabled;
+	nodemask_t mask, clique, disabled, enabled;
 	nodemask_t matrix[MAX_NODES];
 	MtmTransState *ts;
 	int clique_size;
@@ -1496,17 +1496,13 @@ bool MtmRefreshClusterStatus(bool nowait)
 				MtmDisableNode(i+1);
 			}
 		}		
-#if 0	/* Do  not enable nodes here: them will be enabled after completion of recovery */
 		enabled = clique & Mtm->disabledNodeMask; /* new enabled nodes mask */		
-		for (i = 0, mask = enabled; mask != 0; i++, mask >>= 1) {
-			if (mask & 1) { 
-				MtmEnableNode(i+1);
-			}
+		if (testNodeId != 0 && BIT_CHECK(enabled, testNodeId-1)) { 
+			MtmEnableNode(testNodeId);
 		}
-		Mtm->reconnectMask |= clique & Mtm->disabledNodeMask; /* new enabled nodes mask */		
-#endif
+//		Mtm->reconnectMask |= clique & Mtm->disabledNodeMask; /* new enabled nodes mask */		
 
-		if (disabled) { 
+		if (disabled|enabled) { 
 			MtmCheckQuorum();
 		}
 		/* Interrupt voting for active transaction and abort them */
@@ -1622,7 +1618,7 @@ void MtmOnNodeDisconnect(int nodeId)
 		}
 		MtmUnlock();
 	} else { 
-		MtmRefreshClusterStatus(false);
+		MtmRefreshClusterStatus(false, 0);
     }
 }
 
@@ -2595,7 +2591,7 @@ MtmReplicationStartupHook(struct PGLogicalStartupHookArgs* args)
 			MtmCheckQuorum();
 		} else {
 			MtmUnlock();
-			MtmRefreshClusterStatus(true);
+			MtmRefreshClusterStatus(true, MtmReplicationNodeId);
 			MtmLock(LW_SHARED);
 			if (BIT_CHECK(Mtm->disabledNodeMask, MtmReplicationNodeId-1)) {
 				MtmUnlock();
diff --git a/multimaster.h b/multimaster.h
@@ -273,7 +273,7 @@ extern TransactionId MtmGetCurrentTransactionId(void);
 extern XidStatus MtmGetCurrentTransactionStatus(void);
 extern XidStatus MtmExchangeGlobalTransactionStatus(char const* gid, XidStatus status);
 extern bool  MtmIsRecoveredNode(int nodeId);
-extern bool  MtmRefreshClusterStatus(bool nowait);
+extern bool  MtmRefreshClusterStatus(bool nowait, int testNodeId);
 extern void  MtmSwitchClusterMode(MtmNodeStatus mode);
 extern void  MtmUpdateNodeConnectionInfo(MtmConnectionInfo* conn, char const* connStr);
 extern void  MtmSetupReplicationHooks(struct PGLogicalHooks* hooks);

Original file line number	Diff line number	Diff line change
`@@ -1030,7 +1030,7 @@ static void MtmTransReceiver(Datum arg)`
`1030`	`1030`	`}`
`1031`	`1031`	`if (n == 0 && Mtm->disabledNodeMask != 0) {`
`1032`	`1032`	`/* If timeout is expired and there are disabled nodes, then recheck cluster's state */`
`1033`		`- MtmRefreshClusterStatus(false);`
	`1033`	`+ MtmRefreshClusterStatus(false, 0);`
`1034`	`1034`	`}`
`1035`	`1035`	`}`
`1036`	`1036`	`}`