More correct rollback of hanged replicated transactions

knizhnik · kelvich · commit e4ceec5df673 · 2017-11-13T00:17:10.000+03:00
diff --git a/arbiter.c b/arbiter.c
@@ -347,7 +347,7 @@ static void MtmSendHeartbeat()
 		{ 
 			size_t rc = send(sockets[i], &msg, sizeof(msg), 0);
 			if ((size_t)rc != sizeof(msg)) { 
-				elog(LOG, "Failed to send heartbeat to node %d: %d", i, errno);
+				elog(LOG, "Failed to send heartbeat to node %d: %d", i+1, errno);
 			}
 		}
 	}
diff --git a/multimaster.c b/multimaster.c
@@ -1063,7 +1063,7 @@ void MtmWakeUpBackend(MtmTransState* ts)
 void MtmAbortTransaction(MtmTransState* ts)
 {	
 	if (ts->status != TRANSACTION_STATUS_ABORTED) { 
-		MTM_LOG1("Rollback active transaction %d:%d", ts->gtid.node, ts->gtid.xid);
+		MTM_LOG1("Rollback active transaction %d:%d (local xid %d)", ts->gtid.node, ts->gtid.xid, ts->xid);
 		ts->status = TRANSACTION_STATUS_ABORTED;
 		MtmAdjustSubtransactions(ts);
 		Mtm->nActiveTransactions -= 1;
@@ -1387,8 +1387,8 @@ bool MtmRefreshClusterStatus(bool nowait)
 					MtmAbortTransaction(ts);
 					MtmWakeUpBackend(ts);
 				}
-			} else if (BIT_CHECK(disabled, ts->gtid.node-1)) { // coordinator of transaction is on disabled node
-				if (ts->status != TRANSACTION_STATUS_ABORTED) {
+			} else if (TransactionIdIsValid(ts->gtid.xid) && BIT_CHECK(disabled, ts->gtid.node-1)) { // coordinator of transaction is on disabled node
+				if (ts->gid[0] && ts->status != TRANSACTION_STATUS_ABORTED) {
 					MtmAbortTransaction(ts);
 					FinishPreparedTransaction(ts->gid, false);
 				}
@@ -1457,8 +1457,8 @@ void MtmOnNodeDisconnect(int nodeId)
 						MtmAbortTransaction(ts);
 						MtmWakeUpBackend(ts);
 					}
-				} else if (ts->gtid.node == nodeId) { //coordinator of transaction is on disabled node
-					if (ts->status != TRANSACTION_STATUS_ABORTED) {
+				} else if (TransactionIdIsValid(ts->gtid.xid) && ts->gtid.node == nodeId) { //coordinator of transaction is on disabled node
+					if (ts->gid[0] && ts->status != TRANSACTION_STATUS_ABORTED) {
 						MtmAbortTransaction(ts);
 						FinishPreparedTransaction(ts->gid, false);
 					}

Original file line number	Diff line number	Diff line change
`@@ -347,7 +347,7 @@ static void MtmSendHeartbeat()`
`347`	`347`	`{`
`348`	`348`	`size_t rc = send(sockets[i], &msg, sizeof(msg), 0);`
`349`	`349`	`if ((size_t)rc != sizeof(msg)) {`
`350`		`- elog(LOG, "Failed to send heartbeat to node %d: %d", i, errno);`
	`350`	`+ elog(LOG, "Failed to send heartbeat to node %d: %d", i+1, errno);`
`351`	`351`	`}`
`352`	`352`	`}`
`353`	`353`	`}`
Original file line number	Diff line number	Diff line change
`@@ -1063,7 +1063,7 @@ void MtmWakeUpBackend(MtmTransState* ts)`
`1063`	`1063`	`void MtmAbortTransaction(MtmTransState* ts)`
`1064`	`1064`	`{`
`1065`	`1065`	`if (ts->status != TRANSACTION_STATUS_ABORTED) {`
`1066`		`- MTM_LOG1("Rollback active transaction %d:%d", ts->gtid.node, ts->gtid.xid);`
	`1066`	`+ MTM_LOG1("Rollback active transaction %d:%d (local xid %d)", ts->gtid.node, ts->gtid.xid, ts->xid);`
`1067`	`1067`	`ts->status = TRANSACTION_STATUS_ABORTED;`
`1068`	`1068`	`MtmAdjustSubtransactions(ts);`
`1069`	`1069`	`Mtm->nActiveTransactions -= 1;`
`@@ -1387,8 +1387,8 @@ bool MtmRefreshClusterStatus(bool nowait)`
`1387`	`1387`	`MtmAbortTransaction(ts);`
`1388`	`1388`	`MtmWakeUpBackend(ts);`
`1389`	`1389`	`}`
`1390`		`- } else if (BIT_CHECK(disabled, ts->gtid.node-1)) { // coordinator of transaction is on disabled node`
`1391`		`- if (ts->status != TRANSACTION_STATUS_ABORTED) {`
	`1390`	`+ } else if (TransactionIdIsValid(ts->gtid.xid) && BIT_CHECK(disabled, ts->gtid.node-1)) { // coordinator of transaction is on disabled node`
	`1391`	`+ if (ts->gid[0] && ts->status != TRANSACTION_STATUS_ABORTED) {`
`1392`	`1392`	`MtmAbortTransaction(ts);`
`1393`	`1393`	`FinishPreparedTransaction(ts->gid, false);`
`1394`	`1394`	`}`
`@@ -1457,8 +1457,8 @@ void MtmOnNodeDisconnect(int nodeId)`
`1457`	`1457`	`MtmAbortTransaction(ts);`
`1458`	`1458`	`MtmWakeUpBackend(ts);`
`1459`	`1459`	`}`
`1460`		`- } else if (ts->gtid.node == nodeId) { //coordinator of transaction is on disabled node`
`1461`		`- if (ts->status != TRANSACTION_STATUS_ABORTED) {`
	`1460`	`+ } else if (TransactionIdIsValid(ts->gtid.xid) && ts->gtid.node == nodeId) { //coordinator of transaction is on disabled node`
	`1461`	`+ if (ts->gid[0] && ts->status != TRANSACTION_STATUS_ABORTED) {`
`1462`	`1462`	`MtmAbortTransaction(ts);`
`1463`	`1463`	`FinishPreparedTransaction(ts->gid, false);`
`1464`	`1464`	`}`