fix: control the metric set frequency (#458)

ModelTC · Jul 8, 2024 · b8155f0 · b8155f0
1 parent e7e9490
commit b8155f0
Showing 1 changed file with 5 additions and 4 deletions.
diff --git a/lightllm/server/router/manager.py b/lightllm/server/router/manager.py
@@ -190,10 +190,11 @@ async def loop_for_fwd(
             else:
                 self.shared_token_load.set_dynamic_max_load(0.0)
                 self.shared_token_load.set_current_load(0.0)
-                self.metric_client.gauge_set("lightllm_batch_current_size", 0.0)
-                self.metric_client.gauge_set("lightllm_batch_pause_size", 0.0)
-                self.metric_client.gauge_set("lightllm_queue_size", 0.0)
-                self.metric_client.gauge_set("lightllm_batch_current_max_tokens", 0.0)
+                if counter_count % 100 == 0:
+                    self.metric_client.gauge_set("lightllm_batch_current_size", 0.0)
+                    self.metric_client.gauge_set("lightllm_batch_pause_size", 0.0)
+                    self.metric_client.gauge_set("lightllm_queue_size", 0.0)
+                    self.metric_client.gauge_set("lightllm_batch_current_max_tokens", 0.0)
 
             if self.running_batch is None:
                 await asyncio.sleep(0.01)  # 10ms