pgpool2.git / summary / log / commit / refs

commit    3dd1cd3f15287ee6bb8b09f0642f99db98e9776a
Author:   Muhammad Usama <m.usama@gmail.com>
Date:     Wed May 15 21:36:35 2019 +0000

    Fix for [pgpool-hackers: 3295] duplicate failover request ...
    
    Pgpool should keep the backend health check running on quarantined nodes so
    that when the connectivity resumes, they should automatically get removed
    from the quarantine. Otherwise the temporary network glitch could send the node
    into permanent quarantine state.


 src/main/health_check.c | 28 ++++++++++++++++++++++++----
 1 file changed, 24 insertions(+), 4 deletions(-)

diff --git a/src/main/health_check.c b/src/main/health_check.c
index ecf9b348b..6cb13d5c4 100644
--- a/src/main/health_check.c
+++ b/src/main/health_check.c
@@ -178,6 +178,7 @@ do_health_check_child(int *node_id)
 		else if (pool_config->health_check_params[*node_id].health_check_period > 0)
 		{
 			bool		result;
+			BackendInfo *bkinfo = pool_get_node_info(*node_id);
 
 			result = establish_persistent_connection(*node_id);
 
@@ -200,11 +201,27 @@ do_health_check_child(int *node_id)
 					ereport(LOG, (errmsg("health check failed on node %d (timeout:%d)",
 										 *node_id, health_check_timer_expired)));
 
-					/* trigger failover */
-					partial = health_check_timer_expired ? false : true;
-					degenerate_backend_set(node_id, 1, partial ? REQ_DETAIL_SWITCHOVER : 0);
+					if (bkinfo->backend_status == CON_DOWN && bkinfo->quarantine == true)
+					{
+						ereport(LOG, (errmsg("health check failed on quarantine node %d (timeout:%d)",
+											 *node_id, health_check_timer_expired),
+									  errdetail("ignoring..")));
+					}
+					else
+					{
+						/* trigger failover */
+						partial = health_check_timer_expired ? false : true;
+						degenerate_backend_set(node_id, 1, partial ? REQ_DETAIL_SWITCHOVER : 0);
+					}
 				}
 			}
+			else if (bkinfo->backend_status == CON_DOWN && bkinfo->quarantine == true)
+			{
+				/* The node has become reachable again. Reset
+				 * the quarantine state
+				 */
+				send_failback_request(*node_id, false, REQ_DETAIL_UPDATE | REQ_DETAIL_WATCHDOG);
+			}
 
 			/* Discard persistent connections */
 			discard_persistent_connection(*node_id);
@@ -228,9 +245,12 @@ establish_persistent_connection(int node)
 
 	/*
 	 * If the node is already in down status or unused, do nothing.
+	 * except when the node state is down because of quarantine operation
+	 * since we want to detect when the node cames back to life again to
+	 * remove it from the quarantine state
 	 */
 	if (bkinfo->backend_status == CON_UNUSED ||
-		bkinfo->backend_status == CON_DOWN)
+		(bkinfo->backend_status == CON_DOWN && bkinfo->quarantine == false))
 		return false;
 
 	/*


[parent: a45298c4d206]