Merge pull request #12140 from opensourcerouting/fix/watchfrr_dont_givup

watchfrr: Delay restart_kill if still reading configuration

1 files changed, 24 insertions, 0 deletions

diff --git a/watchfrr/watchfrr.c b/watchfrr/watchfrr.c
index 423f25faa2..4a3575ae75 100644
--- a/watchfrr/watchfrr.c
+++ b/watchfrr/watchfrr.c
@@ -112,6 +112,7 @@ static struct global_state {
 	long period;
 	long timeout;
 	long restart_timeout;
+	bool reading_configuration;
 	long min_restart_interval;
 	long max_restart_interval;
 	long operational_timeout;
@@ -365,6 +366,16 @@ static void restart_kill(struct thread *t_kill)
 	struct timeval delay;
 
 	time_elapsed(&delay, &restart->time);
+
+	if (gs.reading_configuration) {
+		zlog_err(
+			"%s %s child process appears to still be reading configuration, delaying for another %lu time",
+			restart->what, restart->name, gs.restart_timeout);
+		thread_add_timer(master, restart_kill, restart,
+				 gs.restart_timeout, &restart->t_kill);
+		return;
+	}
+
 	zlog_warn(
 		"%s %s child process %d still running after %ld seconds, sending signal %d",
 		restart->what, restart->name, (int)restart->pid,
@@ -1059,6 +1070,8 @@ void watchfrr_status(struct vty *vty)
 	vty_out(vty, " Min Restart Interval: %ld\n", gs.min_restart_interval);
 	vty_out(vty, " Max Restart Interval: %ld\n", gs.max_restart_interval);
 	vty_out(vty, " Restart Timeout: %ld\n", gs.restart_timeout);
+	vty_out(vty, " Reading Configuration: %s\n",
+		gs.reading_configuration ? "yes" : "no");
 	if (gs.restart.pid)
 		vty_out(vty, "    global restart running, pid %ld\n",
 			(long)gs.restart.pid);
@@ -1264,6 +1277,16 @@ static void netns_setup(const char *nsname)
 }
 #endif
 
+static void watchfrr_start_config(void)
+{
+	gs.reading_configuration = true;
+}
+
+static void watchfrr_end_config(void)
+{
+	gs.reading_configuration = false;
+}
+
 static void watchfrr_init(int argc, char **argv)
 {
 	const char *special = "zebra";
@@ -1558,6 +1581,7 @@ int main(int argc, char **argv)
 	master = frr_init();
 	watchfrr_error_init();
 	watchfrr_init(argc, argv);
+	cmd_init_config_callbacks(watchfrr_start_config, watchfrr_end_config);
 	watchfrr_vty_init();
 
 	frr_config_fork();