Do not enable node without recovery

Author: knizhnik

contrib/mmts/arbiter.c

@@ -779,7 +779,7 @@ static void MtmMonitor(Datum arg)
 		if (rc & WL_POSTMASTER_DEATH) { 
 			break;
 		}
-		MtmRefreshClusterStatus(true, 0);
+		MtmRefreshClusterStatus(true);
 	}
 }
 
@@ -1086,7 +1086,7 @@ static void MtmReceiver(Datum arg)
 			}
 			if (n == 0 && Mtm->disabledNodeMask != 0) { 
 				/* If timeout is expired and there are disabled nodes, then recheck cluster's state */
-				MtmRefreshClusterStatus(false, 0);
+				MtmRefreshClusterStatus(false);
 			}
 		}
 	}

contrib/mmts/multimaster.c

@@ -1628,8 +1628,8 @@ bool MtmRecoveryCaughtUp(int nodeId, XLogRecPtr slotLSN)
 				BIT_CLEAR(Mtm->nodeLockerMask, nodeId-1);
 				Mtm->nLockers -= 1;
 			} else { 
-				MTM_LOG1("%d: node %d is caugth-up without locking cluster", MyProcPid, nodeId);	
-				/* We are lucky: caugth-up without locking cluster! */
+				MTM_LOG1("%d: node %d is caught-up without locking cluster", MyProcPid, nodeId);	
+				/* We are lucky: caught-up without locking cluster! */
 			}
 			MtmEnableNode(nodeId);
 			Mtm->nConfigChanges += 1;
@@ -1779,9 +1779,9 @@ 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, int testNodeId)
+bool MtmRefreshClusterStatus(bool nowait)
 {
-	nodemask_t mask, clique, disabled, enabled;
+	nodemask_t mask, clique, disabled;
 	nodemask_t matrix[MAX_NODES];
 	int clique_size;
 	int i;
@@ -1822,13 +1822,6 @@ bool MtmRefreshClusterStatus(bool nowait, int testNodeId)
 					}
 				}
 			}
-		}		
-		enabled = clique & Mtm->disabledNodeMask; /* new enabled nodes mask */		
-		if (testNodeId != 0 && BIT_CHECK(enabled, testNodeId-1)) { 
-			MtmEnableNode(testNodeId);
-		}
-
-		if (disabled|enabled) {
 			MtmCheckQuorum();
 		}			
 #if 0
@@ -1908,7 +1901,7 @@ void MtmOnNodeDisconnect(int nodeId)
 	}
 
 	MtmSleep(MSEC_TO_USEC(MtmHeartbeatSendTimeout));
-	MtmRefreshClusterStatus(false, 0);
+	MtmRefreshClusterStatus(false);
 }
 
 void MtmOnNodeConnect(int nodeId)
@@ -3077,12 +3070,7 @@ MtmReplicationStartupHook(struct PGLogicalStartupHookArgs* args)
 			MtmCheckQuorum();
 		} else {
 			MtmUnlock();
-			MtmRefreshClusterStatus(true, MtmReplicationNodeId);
-			MtmLock(LW_SHARED);
-			if (BIT_CHECK(Mtm->disabledNodeMask, MtmReplicationNodeId-1)) {
-				MtmUnlock();
-				elog(ERROR, "Disabled node %d tries to reconnect without recovery", MtmReplicationNodeId); 
-			}
+			elog(ERROR, "Disabled node %d tries to reconnect without recovery", MtmReplicationNodeId); 
 		}
 	} else {
 		MTM_LOG1("Node %d start logical replication to node %d in normal mode", MtmNodeId, MtmReplicationNodeId); 

contrib/mmts/multimaster.h

@@ -350,7 +350,7 @@ 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, int testNodeId);
+extern bool  MtmRefreshClusterStatus(bool nowait);
 extern void  MtmSwitchClusterMode(MtmNodeStatus mode);
 extern void  MtmUpdateNodeConnectionInfo(MtmConnectionInfo* conn, char const* connStr);
 extern void  MtmSetupReplicationHooks(struct PGLogicalHooks* hooks);