[BUG] UVicorn generates error 404 session id not found when starting multiple workers

[kaka1nices]

Describe the bug
When using uvicorn to start multiple workers, the MCP server will report an error 404 session id Not Found due to session id issues

To Reproduce

uvicorn.run(app='server:app', host="0.0.0.0", port=8000,workers=2)

Error log

INFO:     127.0.0.1:63670 - "POST /mcp/messages/?session_id=5576a7215caf42a38ba8d7c996d07f17 HTTP/1.1" 202 Accepted
INFO:     127.0.0.1:63769 - "GET /mcp HTTP/1.1" 200 OK
INFO:     127.0.0.1:63670 - "POST /mcp/messages/?session_id=495548f6adc54c618f8619d1ae07e51e HTTP/1.1" 404 Not Found

System Info
Python 3.12.5

[tonderflash]

I'll be working on this issue

[surki]

thanks @tonderflash
We are also looking for this, as we run multiple pods and multiple workers, spanning multiple machines.

[tonderflash]

@surki I'm considering implementing a shared storage mechanism for SSE sessions that would work with multiple workers/pods to solve this issue. Do you have any specific preferences or restrictions regarding storage solutions (Redis, databases, etc.) that I should take into account for the use case?

[surki]

@tonderflash Perhaps we should have some kind of common interface for session management. Then we could have few implementations to start with, redis or so (we are looking at leveraging redis since it is already present in our infra).

We should also think about purging stale entries, what happens if the target instance is down etc

[maiyad-sia]

Any updates on this ?