Description
Context
If Gunicron is configured with keepalive != 0, then base_async.py will handle it with keepalive loop.
# keepalive loop
proxy_protocol_info = {}
while True:
req = None
with self.timeout_ctx():
req = next(parser)
if not req:
break
if req.proxy_protocol_info:
proxy_protocol_info = req.proxy_protocol_info
else:
req.proxy_protocol_info = proxy_protocol_info
self.handle_request(listener_name, req, client, addr)
Here when using worker type as: worker_class = "gevent", then gevent.py defined as:
def timeout_ctx(self):
return gevent.Timeout(self.cfg.keepalive, False)
This logic will throw out gevent.Timeout type when timeout and our logic cannot handle that in all different kinds of exception logic until getting to the final branch: https://github.com/benoitc/gunicorn/blob/master/gunicorn/workers/base_async.py#L87
Because:
class Timeout(BaseException):
Impact
This will be a problem as a connection between the gateway (eg: Nginx) and Gunicorn will be suddenly closed without response and Nginx will expect the service was died and return back clients with error 502.
From Nginx you will see something that is not reasonable like:
2023/01/20 13:46:02 [error] 1027#1027: *209 upstream prematurely closed connection while reading response header from upstream, client: 127.0.0.1, server: flask_server.com, request: "GET /sleep?time=1 HTTP/1.1", upstream: "http://127.0.0.1:8000/sleep?time=1", host: "flask_server.com"
The error can also be triggered when throwing out any other exception that is not inherited from Exception but BaseException from user logic in Flask app.