Abort transaction ni case of cluster configuration change

knizhnik · kelvich · commit bf168d647507 · 2017-11-13T00:15:36.000+03:00
diff --git a/multimaster.c b/multimaster.c
@@ -273,8 +273,8 @@ void MtmSleep(timestamp_t interval)
 {
     struct timespec ts;
     struct timespec rem;
-    ts.tv_sec = interval/1000000;
-    ts.tv_nsec = interval%1000000*1000;
+    ts.tv_sec = interval/USECS_PER_SEC;
+    ts.tv_nsec = interval%USECS_PER_SEC*1000;
 
     while (nanosleep(&ts, &rem) < 0) { 
         Assert(errno == EINTR);
@@ -751,16 +751,20 @@ MtmPostPrepareTransaction(MtmCurrentTrans* x)
 	} else { 
 		time_t timeout = Max(Mtm2PCMinTimeout, (ts->csn - ts->snapshot)*Mtm2PCPrepareRatio/100000); /* usec->msec and percents */ 
 		int result = 0;
+		int nConfigChanges = Mtm->nConfigChanges;
 		/* wait votes from all nodes */
 		while (!ts->votingCompleted && !(result & WL_TIMEOUT)) {
 			MtmUnlock();
 			result = WaitLatch(&MyProc->procLatch, WL_LATCH_SET|WL_TIMEOUT, timeout);
 			ResetLatch(&MyProc->procLatch);			
 			MtmLock(LW_SHARED);
 		}
-		if (!ts->votingCompleted) {
+		if (!ts->votingCompleted) { 
 			ts->status = TRANSACTION_STATUS_ABORTED;
-			elog(WARNING, "Transaction is aborted because of %d msec timeout expiration, prepare time %d msec", (int)timeout, (int)((ts->csn - x->snapshot)/1000));
+			elog(WARNING, "Transaction is aborted because of %d msec timeout expiration, prepare time %d msec", (int)timeout, (int)USEC_TO_MSEC(ts->csn - x->snapshot));
+		} else if (nConfigChanges != Mtm->nConfigChanges) {
+			ts->status = TRANSACTION_STATUS_ABORTED;
+			elog(WARNING, "Transaction is aborted because cluster configuration is changed during commit");
 		}
 		x->status = ts->status;
 		MTM_LOG3("%d: Result of vote: %d", MyProcPid, ts->status);
@@ -1088,6 +1092,7 @@ bool MtmRecoveryCaughtUp(int nodeId, XLogRecPtr slotLSN)
 			BIT_CLEAR(Mtm->disabledNodeMask, nodeId-1);
 			Mtm->nodes[nodeId-1].lastStatusChangeTime = MtmGetSystemTime();
 			Mtm->nLiveNodes += 1;
+			Mtm->nConfigChanges += 1;
 			caughtUp = true;
 		} else if (!BIT_CHECK(Mtm->nodeLockerMask, nodeId-1)
 				   && slotLSN + MtmMinRecoveryLag > walLSN) 
@@ -1262,6 +1267,7 @@ bool MtmRefreshClusterStatus(bool nowait)
 
 void MtmCheckQuorum(void)
 {
+	Mtm->nConfigChanges += 1;
 	if (Mtm->nLiveNodes < Mtm->nAllNodes/2+1) {
 		if (Mtm->status == MTM_ONLINE) { /* out of quorum */
 			elog(WARNING, "Node is in minority: disabled mask %lx", (long) Mtm->disabledNodeMask);
@@ -1459,6 +1465,7 @@ static void MtmInitialize()
         Mtm->nReceivers = 0;
 		Mtm->timeShift = 0;
 		Mtm->transCount = 0;
+		Mtm->nConfigChanges = 0;
 		Mtm->localTablesHashLoaded = false;
 		for (i = 0; i < MtmNodes; i++) {
 			Mtm->nodes[i].oldestSnapshot = 0;
diff --git a/multimaster.h b/multimaster.h
@@ -169,7 +169,8 @@ typedef struct
     int    nReceivers;                 /* Number of initialized logical receivers (used to determine moment when Mtm intialization is completed */
 	int    nLockers;                   /* Number of lockers */
 	int    nActiveTransactions;        /* Nunmber of active 2PC transactions */
-	long   timeShift;                  /* Local time correction */
+	int    nConfigChanges;             /* Number of cluster configuration changes */
+	int64  timeShift;                  /* Local time correction */
 	csn_t  csn;                        /* Last obtained CSN: used to provide unique acending CSNs based on system time */
 	MtmTransState* votingTransactions; /* L1-list of replicated transactions sendings notifications to coordinator.
 									 	 This list is used to pass information to mtm-sender BGW */
diff --git a/tests/dtmacid.cpp b/tests/dtmacid.cpp
@@ -144,7 +144,6 @@ void* reader(void* arg)
 				}
 			}
 		}
-        t.transactions += 2;
         t.selects += 2;
         txn1.commit();
         txn2.commit();

Original file line number	Diff line number	Diff line change
`@@ -144,7 +144,6 @@ void* reader(void* arg)`
`144`	`144`	`}`
`145`	`145`	`}`
`146`	`146`	`}`
`147`		`- t.transactions += 2;`
`148`	`147`	`t.selects += 2;`
`149`	`148`	`txn1.commit();`
`150`	`149`	`txn2.commit();`