Fix wrong quorum calculation when walsender restarts before we can detect failure with heartbeats

kelvich · kelvich · commit 574ee5617281 · 2018-08-17T15:56:01.000+03:00
diff --git a/state.c b/state.c
@@ -247,9 +247,9 @@ MtmStateProcessNeighborEvent(int node_id, MtmNeighborEvent ev) // XXXX camelcase
 			 */
 			if (!BIT_CHECK(Mtm->disabledNodeMask, node_id-1))
 			{
-				BIT_SET(Mtm->disabledNodeMask, node_id-1);
+				MTM_LOG1("[WARN] node %d started recovery, but it wasn't disabled", node_id);
+				MtmOnNodeDisconnect(node_id-1);
 			}
-			// BIT_SET(Mtm->pglogicalSenderMask, node_id - 1);
 			break;
 
 		case MTM_NEIGHBOR_WAL_SENDER_START_RECOVERED:

Original file line number	Diff line number	Diff line change
`@@ -247,9 +247,9 @@ MtmStateProcessNeighborEvent(int node_id, MtmNeighborEvent ev) // XXXX camelcase`
`247`	`247`	`*/`
`248`	`248`	`if (!BIT_CHECK(Mtm->disabledNodeMask, node_id-1))`
`249`	`249`	`{`
`250`		`- BIT_SET(Mtm->disabledNodeMask, node_id-1);`
	`250`	`+ MTM_LOG1("[WARN] node %d started recovery, but it wasn't disabled", node_id);`
	`251`	`+ MtmOnNodeDisconnect(node_id-1);`
`251`	`252`	`}`
`252`		`- // BIT_SET(Mtm->pglogicalSenderMask, node_id - 1);`
`253`	`253`	`break;`
`254`	`254`
`255`	`255`	`case MTM_NEIGHBOR_WAL_SENDER_START_RECOVERED:`