After update, hard refresh required or chats hang forever

[jimbo-p]

I’m having issues when updating Open WebUI where chats will hang indefinitely. A hard refresh seems to fix the issue.

In going from 0.6.2x to 0.6.30, all users were required to hard refresh the Open WebUI page or chats would hang.
In going from 0.6.30 to 0.6.36, the same issue occurred.

This was reproducible:

1. Start in 0.6.30, test it
2. Upgrade to 0.6.36
3. Type a message, it hangs (‘thinking dot’ appears and doesn’t go away).
4. Refresh page, response appears. So the model did respond, it just didn’t display.
5. Open a new browser session and start after at (2).
6. Hard refresh and then type a message, everything works and continues to work from there on out.

I’m running this on 4 instances of ECS with redis and postgres. Routing is handled by an ALB / sticky sessions, conditional routing. Requests go to litellm proxy - doesn’t seem to be issues there. I’ve tried removing redis and scaling down to 1 running instance and the issue persisted. I could not find any errors in Open WebUI backend or in Dev Tools on the front end.

Anyone else have this issue or have ideas on how to fix?

[jimbo-p]

Deployed nginx as a reverse proxy sidecar that enforces cache-control policies: critical files (HTML, service workers, manifests) are never cached while static assets with hashed filenames cache aggressively. This seems to solve the stale version problem where browser-cached service workers would continue serving old application code after updates.