Skip to content

RAGFlow is an agent-native database: a Retrieval-Augmented Generation (RAG) engine that grew into context infrastructure.

Fix: SIGTERM not handle well in ragflow server chat channel (#17828)

Wang Qif057544e313c
diff --git a/api/channels/bootstrap.py b/api/channels/bootstrap.py
index a607d40383..5139cd11e5 100644
--- a/api/channels/bootstrap.py
+++ b/api/channels/bootstrap.py
@@ -219,13 +219,17 @@ async def _start_channel(running: dict, account_id: str, channel: str, credentia
return True
-async def _reconcile(running: dict, failed: dict) -> None:
+async def _reconcile(running: dict, failed: dict, stop_event: threading.Event) -> None:
"""Diff desired (DB) vs running channels and apply start/stop/restart.
``failed`` remembers configs that could not be started so they are not
retried (and re-logged) every tick until their credentials change.
"""
+ if stop_event.is_set():
+ return
desired = await asyncio.to_thread(_desired_channels)
+ if stop_event.is_set():
+ return
# Stop channels that were removed or whose credentials/type changed.
for account_id in list(running.keys()):
@@ -266,7 +270,12 @@ async def run_channels(stop_event: threading.Event) -> None:
try:
while not stop_event.is_set():
try:
- await _reconcile(running, failed)
+ await _reconcile(running, failed, stop_event)
+ except RuntimeError as ex:
+ if stop_event.is_set():
+ LOGGER.info("chat channel reconcile stopped")
+ break
+ LOGGER.error("chat channel reconcile failed: %s", ex)
except Exception as ex:
LOGGER.error("chat channel reconcile failed: %s", ex)
diff --git a/api/ragflow_server.py b/api/ragflow_server.py
index 5f580075d6..f9c9d62c08 100644
--- a/api/ragflow_server.py
+++ b/api/ragflow_server.py
@@ -48,6 +48,7 @@ from agent.plugin import GlobalPluginManager
from rag.utils.redis_conn import RedisDistributedLock
stop_event = threading.Event()
+chat_channel_thread = None
RAGFLOW_DEBUGPY_LISTEN = int(os.environ.get("RAGFLOW_DEBUGPY_LISTEN", "0"))
@@ -71,11 +72,16 @@ def update_progress():
stop_event.wait(6)
+def stop_background_services():
+ stop_event.set()
+ if chat_channel_thread and chat_channel_thread.is_alive() and chat_channel_thread is not threading.current_thread():
+ chat_channel_thread.join(timeout=5)
+
+
def signal_handler(sig, frame):
logging.info("Received interrupt signal, shutting down...")
shutdown_all_mcp_sessions()
- stop_event.set()
- stop_event.wait(1)
+ stop_background_services()
sys.exit(0)
@@ -137,17 +143,18 @@ if __name__ == "__main__":
t.start()
def start_chat_channels():
+ global chat_channel_thread
try:
from api.channels.bootstrap import start_channel_server
logging.info("Starting chat channel server thread")
- t = threading.Thread(
+ chat_channel_thread = threading.Thread(
target=start_channel_server,
args=(stop_event,),
daemon=True,
name="chat-channels",
)
- t.start()
+ chat_channel_thread.start()
except Exception:
logging.exception("Failed to start chat channel server")
@@ -165,6 +172,7 @@ if __name__ == "__main__":
app.run(host=settings.HOST_IP, port=settings.HOST_PORT, use_reloader=RuntimeConfig.DEBUG, debug=False)
except Exception as e:
logging.exception(f"Unhandled exception: {e}")
- stop_event.set()
- stop_event.wait(1)
+ stop_background_services()
os.kill(os.getpid(), signal.SIGKILL)
+ finally:
+ stop_background_services()