Skip to contentSkip to conversation

rgmanager: Don't duplicate configuration change events

PATCH New Ryan McCabe 2 participants 2 messages Feb 26, 2013 – Feb 27, 2013

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 rgmanager: Don't duplicate configuration change events

If a cluster configuration change is detected while there are events on the event queue that take some time to process, a large number of duplicate configuration update events are put onto the event queue. With a large number of services and a small status poll interval, thousands of duplicate events can be put onto the queue.

This patch causes rgmanager to queue only one config event per detected configuration change.

This applies only to the RHEL5 branch.

---
 rgmanager/src/daemons/main.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
rgmanager/src/daemons/main.c +7 -1
index 1c7f746..59150c3 100644
@@ -73,6 +73,7 @@ static int port = RG_PORT;
 static char *rgmanager_lsname = "rgmanager"; /* XXX default */
 static int status_poll_interval = DEFAULT_CHECK_INTERVAL;
 static int stops_queued = 0;
+static int lastqver = 0;
 
 int next_node_id(cluster_member_list_t *membership, int me);
 
@@ -750,7 +751,7 @@ dump_internal_state(int fd)
 int
 event_loop(msgctx_t *localctx, msgctx_t *clusterctx)
 {
-	int n = 0, max, ret, oldver, newver;
+	int n = 0, max, ret, oldver = 0, newver = 0;
 	fd_set rfds;
 	msgctx_t *newctx;
 	struct timeval tv;
@@ -826,9 +827,14 @@ event_loop(msgctx_t *localctx, msgctx_t *clusterctx)
 		return 0;
 
 	if (need_reconfigure || check_config_update(&oldver, &newver)) {
+		/* Only queue configuration events once per detected config update */
+		if (!need_reconfigure && newver <= lastqver)
+			return 0;
+
 		need_reconfigure = 0;
 		configure_rgmanager(-1, 0, NULL);
 		config_event_q(oldver, newver);
+		lastqver = newver;
 		return 0;
 	}
 
Signature · 1 line
1.8.1.2
#1 lore raw

PATCH 1/1 rgmanager: Don't duplicate configuration change events

Quoted 8 lines · Ryan McCabe

If a cluster configuration change is detected while there are events on the event queue that take some time to process, a large number of duplicate configuration update events are put onto the event queue. With a large number of services and a small status poll interval, thousands of duplicate events can be put onto the queue.

This patch causes rgmanager to queue only one config event per detected configuration change.

Good catch there.

Ack /

-- Lon

Quoted 45 lines

This applies only to the RHEL5 branch.

---
 rgmanager/src/daemons/main.c | 8 +++++++-
 1 file changed, 7 insertions(+), 1 deletion(-)
rgmanager/src/daemons/main.c +7 -1
index 1c7f746..59150c3 100644
@@ -73,6 +73,7 @@ static int port = RG_PORT;
 static char *rgmanager_lsname = "rgmanager"; /* XXX default */
 static int status_poll_interval = DEFAULT_CHECK_INTERVAL;
 static int stops_queued = 0;
+static int lastqver = 0;
 
 int next_node_id(cluster_member_list_t *membership, int me);
 
@@ -750,7 +751,7 @@ dump_internal_state(int fd)
 int
 event_loop(msgctx_t *localctx, msgctx_t *clusterctx)
 {
-	int n = 0, max, ret, oldver, newver;
+	int n = 0, max, ret, oldver = 0, newver = 0;
 	fd_set rfds;
 	msgctx_t *newctx;
 	struct timeval tv;
@@ -826,9 +827,14 @@ event_loop(msgctx_t *localctx, msgctx_t *clusterctx)
 		return 0;
 
 	if (need_reconfigure || check_config_update(&oldver, &newver)) {
+		/* Only queue configuration events once per detected config update */
+		if (!need_reconfigure && newver <= lastqver)
+			return 0;
+
 		need_reconfigure = 0;
 		configure_rgmanager(-1, 0, NULL);
 		config_event_q(oldver, newver);
+		lastqver = newver;
 		return 0;
 	}