JWT Refresh not working reliably.

[adamhf]

I'm running LearningHouse in docker, invoked with the following command

docker run --name learninghouse --rm -v brains:/learninghouse/brains -p 5000:5000 -e "LEARNINGHOUSE_HOST=0.0.0.0" -e "LEARNINGHOUSE_BASE_URL=http://helios" -e "LEARNINGHOUSE_LOGGING_LEVEL=DEBUG" -e LEARNINGHOUSE_JWT_SECRET="supersecret" -e "LEARNINGHOUSE_JWT_EXPIRE_MINUTES=600" ghcr.io/deroetzi/learninghouse:main

The JWT refresh seems unreliable. Sometimes it will refresh expected, others it will fail as seen in the logs below:

2024-04-27 14:58:42.956 | INFO     | learninghouse.services.auth:create_token:60 - Admin user logged in sucessfully
2024-04-27 14:58:42.957 | INFO     | uvicorn.protocols.http.httptools_impl:send:496 - 10.10.10.199:52153 - "POST /api/auth/token HTTP/1.1" 200
2024-04-27 14:58:42.977 | INFO     | uvicorn.protocols.http.httptools_impl:send:496 - 10.10.10.199:52153 - "GET /api/brains/info HTTP/1.1" 200
2024-04-27 14:58:49.998 | INFO     | learninghouse.services.auth:refresh_token:72 - Admin token refreshed
2024-04-27 14:58:49.999 | INFO     | uvicorn.protocols.http.httptools_impl:send:496 - 10.10.10.199:52165 - "PUT /api/auth/token HTTP/1.1" 200
2024-04-27 14:58:58.358 | ERROR    | learninghouse.services.auth:verify_jwt:269 - No valid refresh token
2024-04-27 14:58:58.358 | WARNING  | learninghouse.api.errors:learninghouse_exception_handler:166 - Could not validate credentials
2024-04-27 14:58:58.359 | INFO     | uvicorn.protocols.http.httptools_impl:send:496 - 10.10.10.199:52172 - "PUT /api/auth/token HTTP/1.1" 401

The scenario above was:

• Log in as the admin user,
• wait ~10s
• Click the refresh token button, the token refresh succeeds
• wait ~10s
• Click the button again, the token refresh fails

This scenario is quite contrived, who sits there and and refreshes a token? The bug becomes an issue when the refresh fails and you get logged out every 60 seconds.

[DerOetzi]

Please recheck with newest main package i have replaced the underlying JWT libraries

[DerOetzi]

The problem is in production mode with several workers after task #373 is implemented it should work