[GC Debugger] Simply and improve GC Debugger Utils (#29029)

Signed-off-by: Jialin Ouyang <Jialin.Ouyang@gmail.com>

vllm/utils/gc_utils.py

@@ -68,9 +68,10 @@ class GCDebugger:
             # Before GC started, record GC start time
             # and top collected objects
             self.start_time_ns = time.monotonic_ns()
-            self.gc_top_collected_objects = _compute_top_gc_collected_objects(
-                gc.get_objects(generation), self.config.top_objects
-            )
+            if (top_objects := self.config.top_objects) > 0:
+                self.gc_top_collected_objects = _compute_top_gc_collected_objects(
+                    gc.get_objects(generation), top_objects
+                )
         elif phase == "stop":
             # After GC finished, Record GC elapsed time and
             # optionally top collected objects

vllm/v1/engine/core.py

@@ -206,6 +206,8 @@ class EngineCore:
         # Mark the startup heap as static so that it's ignored by GC.
         # Reduces pause times of oldest generation collections.
         freeze_gc_heap()
+        # If enable, attach GC debugger after static variable freeze.
+        maybe_attach_gc_debug_callback()
 
     def _initialize_kv_caches(
         self, vllm_config: VllmConfig
@@ -645,9 +647,6 @@ class EngineCoreProc(EngineCore):
                 assert addresses.coordinator_input is not None
                 logger.info("Waiting for READY message from DP Coordinator...")
 
-        # If enable, attach GC debugger after static variable freeze.
-        maybe_attach_gc_debug_callback()
-
         # Enable environment variable cache (e.g. assume no more
         # environment variable overrides after this point)
         enable_envs_cache()