More correct rollback of hanged replicated transactions

knizhnik · knizhnik · commit ed87b14e1b58 · 2016-06-27T19:48:20.000+03:00
diff --git a/contrib/mmts/arbiter.c b/contrib/mmts/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/contrib/mmts/multimaster.c b/contrib/mmts/multimaster.c
@@ -1064,7 +1064,7 @@ 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;
@@ -1388,9 +1388,8 @@ 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) {
-					MTM_LOG1("1) Rollback active transaction %d:%d:%d", ts->gtid.node, ts->gtid.xid, ts->xid);
+			} 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);
 				}
@@ -1459,9 +1458,8 @@ 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) {
-						MTM_LOG1("2) Rollback active transaction %d:%d", ts->gtid.node, ts->gtid.xid);
+				} 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`	`}`