Worker handling WS:// gets SIGTERM and causes lost messages

[amalizzio]

Hi all!

I noticed that when my gunicorn worker timeout occurs, SIGTERM kills the proc and gunicorn replaces.

Sometimes, at the time of gunicorn timeout, the replacement worker gets websocket accept, sends a message (or several messages), but the client does not receive.

Strangely the client has the old WS not timed out yet, and the new WS (whcih was accept on server) does not get the new messages.

My question is two-fold:

1 - anyone have an obvious answer and / or seen this before?

2 - if this is non-obvious, at a minimum would it make sense such that uvicorn such that SIGTERM is caught, and any websockets get .close() ?