[Frontend][Core] Add shutdown timeout - allowing in-flight requests to finish (#34730)

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Co-authored-by: Claude Sonnet 4.5 <noreply@anthropic.com>

vllm/v1/engine/utils.py

@@ -3,6 +3,7 @@
 
 import contextlib
 import os
+import threading
 import weakref
 from collections.abc import Callable, Iterator
 from dataclasses import dataclass
@@ -151,11 +152,12 @@ class CoreEngineProcManager:
         finally:
             # Kill other procs if not all are running.
             if self.finished_procs():
-                self.close()
+                self.shutdown()
 
-    def close(self):
-        """Shutdown all procs."""
-        self._finalizer()
+    def shutdown(self, timeout: float | None = None) -> None:
+        """Shutdown engine core processes with configurable timeout."""
+        if self._finalizer.detach() is not None:
+            shutdown(self.processes, timeout=timeout)
 
     def join_first(self):
         """Wait for any process to exit."""
@@ -173,6 +175,33 @@ class CoreEngineProcManager:
         }
 
 
+class SignalCallback:
+    """Safely trigger a callback from signal handler context via a dedicated thread."""
+
+    def __init__(self, callback: Callable[[], None]):
+        self._callback = callback
+        self._event = threading.Event()
+        self._stopped = False
+        self._thread = threading.Thread(
+            target=self._run,
+            daemon=True,
+            name="signal-callback",
+        )
+        self._thread.start()
+
+    def _run(self):
+        self._event.wait()
+        if not self._stopped:
+            self._callback()
+
+    def trigger(self):
+        self._event.set()
+
+    def stop(self):
+        self._stopped = True
+        self._event.set()
+
+
 @contextlib.contextmanager
 def set_device_control_env_var(
     vllm_config: VllmConfig, local_dp_rank: int
@@ -768,7 +797,7 @@ class CoreEngineActorManager:
     def get_run_refs(self):
         return self.run_refs
 
-    def close(self):
+    def shutdown(self, timeout: float | None = None) -> None:
         import ray
 
         for actor in self.local_engine_actors + self.remote_engine_actors: