From 73486625b436fc1befe1e4dac5b086085b0adace Mon Sep 17 00:00:00 2001
From: Olga Andreeva <124622579+oandreeva-nv@users.noreply.github.com>
Date: Fri, 20 Dec 2024 15:10:58 -0800
Subject: [PATCH] Setting shutdown asyncio event in a thread-safe manner (#78)

---
 ci/L0_backend_vllm/metrics_test/test.sh | 4 ++--
 src/model.py                            | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)

diff --git a/ci/L0_backend_vllm/metrics_test/test.sh b/ci/L0_backend_vllm/metrics_test/test.sh
index 884be62..a9a4db9 100755
--- a/ci/L0_backend_vllm/metrics_test/test.sh
+++ b/ci/L0_backend_vllm/metrics_test/test.sh
@@ -75,11 +75,11 @@ run_test() {
         fi
     fi
 
+    set -e
+
     # TODO: Non-graceful shutdown when metrics are enabled.
     kill $SERVER_PID
     wait $SERVER_PID
-
-    set -e
 }
 
 RET=0
diff --git a/src/model.py b/src/model.py
index 46c35a2..4c351f1 100644
--- a/src/model.py
+++ b/src/model.py
@@ -800,7 +800,7 @@ def _check_health(self, requests):
 
     def finalize(self):
         self.logger.log_info("[vllm] Issuing finalize to vllm backend")
-        self._llm_engine_shutdown_event.set()
+        self._event_loop.call_soon_threadsafe(self._llm_engine_shutdown_event.set)
 
         # Shutdown the event thread.
         if self._event_thread is not None: