From 802396d085fbcd76d102178c6a89fad5c3a3a3df Mon Sep 17 00:00:00 2001
From: Victor <87538976+visill@users.noreply.github.com>
Date: Thu, 3 Oct 2024 13:48:12 +0500
Subject: [PATCH] Fix od_rules_group_checker_run (light verion) (#695)

* Fix od_rules_group_checker_run

* Fix memleak when group checker fail to check

* Add test with falling postgres
---
 docker/group/config.conf   |  2 +-
 docker/group/test_group.sh | 19 ++++++++++++++++++-
 sources/rules.c            | 24 +++++++++++++++++++++++-
 3 files changed, 42 insertions(+), 3 deletions(-)

diff --git a/docker/group/config.conf b/docker/group/config.conf
index a656adca2..0caa5daf1 100644
--- a/docker/group/config.conf
+++ b/docker/group/config.conf
@@ -138,5 +138,5 @@ enable_online_restart yes
 bindwith_reuseport yes
 
 stats_interval 60
-
+group_checker_interval 500
 pid_file "/var/run/odyssey.pid"
diff --git a/docker/group/test_group.sh b/docker/group/test_group.sh
index bf3411ef9..253ee68d2 100755
--- a/docker/group/test_group.sh
+++ b/docker/group/test_group.sh
@@ -92,4 +92,21 @@ psql -h ip4-localhost -p 6432 -U group_user7 -c "SELECT 1" group_db >/dev/null 2
 	exit 1
 }
 
-ody-stop
+sudo -u postgres /usr/lib/postgresql/14/bin/pg_ctl -D /var/lib/postgresql/14/repl/ -o '-p 5433' stop
+sudo -u postgres /usr/lib/postgresql/14/bin/pg_ctl -D /var/lib/postgresql/14/main/ stop
+sleep 2
+sudo -u postgres /usr/lib/postgresql/14/bin/pg_ctl -D /var/lib/postgresql/14/main/ start
+sudo -u postgres /usr/lib/postgresql/14/bin/pg_ctl -D /var/lib/postgresql/14/repl/ -o '-p 5433' start
+psql -h ip4-localhost -p 6432 -U group_user7 -c "SELECT 1" group_db >/dev/null 2>&1 && {
+	echo "Break by falling postgres"
+
+	cat /var/log/odyssey.log
+	echo "
+
+	"
+	cat /var/log/postgresql/postgresql-14-main.log
+
+	exit 1
+}
+
+ody-stop
\ No newline at end of file
diff --git a/sources/rules.c b/sources/rules.c
index a8510d392..8ac4fa3bd 100644
--- a/sources/rules.c
+++ b/sources/rules.c
@@ -270,7 +270,7 @@ void od_rules_group_checker_run(void *arg)
 				continue;
 			}
 		}
-
+		/* TODO: remove this loop (always works once)*/
 		for (int retry = 0; retry < group->check_retry; ++retry) {
 			if (od_backend_query_send(
 				    server, "group_checker", group->group_query,
@@ -295,6 +295,11 @@ void od_rules_group_checker_run(void *arg)
 							 "read error: %s",
 							 od_io_error(
 								 &server->io));
+						rc = -1;
+						break;
+					} else {
+						/* If timeout try read again */
+						continue;
 					}
 				}
 
@@ -342,6 +347,23 @@ void od_rules_group_checker_run(void *arg)
 
 			od_router_close(router, group_checker_client);
 
+			if (rc == NOT_OK_RESPONSE) {
+				od_debug(&instance->logger, "group_checker",
+					 group_checker_client, server,
+					 "group check failed");
+
+				od_list_t *it, *n;
+				od_list_foreach_safe(&members, it, n)
+				{
+					member = od_container_of(
+						it, od_group_member_name_item_t,
+						link);
+					if (member)
+						free(member);
+				}
+				break;
+			}
+
 			bool have_default = false;
 			od_list_t *i;
 			int count_group_users = 0;