zebra: Temporarily block the execution of the rib_process function wh…

…ile the thread t_dplane is waiting to be scheduled. This allows t_dplane to prioritize freeing up the cache structures of zebra_dplane_ctx. This addresses the issue where a large number of zebra_dplane_ctx nodes are cached on the rib_dplane_q when a surge of routes is inserted in a short period of time, leading to the consumption of a significant amount of temporary memory and potentially causing system memory overload and kernel OOM (Out of Memory) problems. Signed-off-by: hanyu.zly <[email protected]>
FRRouting · Dec 25, 2023 · 3b8c4f1 · 3b8c4f1
1 parent 080299f
commit 3b8c4f1
Showing 1 changed file with 2 additions and 1 deletion.
diff --git a/zebra/zebra_rib.c b/zebra/zebra_rib.c
@@ -3228,7 +3228,7 @@ static wq_item_status meta_queue_process(struct work_queue *dummy, void *data)
 	/* Ensure there's room for more dataplane updates */
 	queue_limit = dplane_get_in_queue_limit();
 	queue_len = dplane_get_in_queue_len();
-	if (queue_len > queue_limit) {
+	if (queue_len > queue_limit || t_dplane) {
 		if (IS_ZEBRA_DEBUG_RIB_DETAILED)
 			zlog_debug(
 				"rib queue: dplane queue len %u, limit %u, retrying",
@@ -4974,6 +4974,7 @@ static void rib_process_dplane_results(struct event *thread)
 		}
 
 	} while (1);
+	t_dplane = NULL;
 
 #ifdef HAVE_SCRIPTING
 	if (fs)