fix: log unexpected errors server-side, return generic message to client (#991)

[SAY-5]

Misc PR

Type of PR

• Bug Fix
• New Feature
• Documentation
• Other

Description

• Link to Issue: fixes: m serve: generic except-Exception handler masks errors and leaks message to client #991

The catch-all except Exception in cli/serve/app.py had two problems:

1. No server-side log trail. Operators saw a 500 response with no stack trace in the application log, the exact case where a visible traceback matters.
2. Exception message leaked verbatim to clients. f"Internal server error: {e!s}" returned arbitrary str(e) content (file paths, internal module names, schema fragments) across the OpenAI-compatible API boundary.

Replaces the handler with logger.exception("Unhandled error in chat-completion handler") (full traceback, named logger) plus a generic client message. The explicit except ValueError 400-mapping above it stays untouched, that path is intentional validation feedback.

Testing

• Tests added to the respective file if code was changed
• New code has 100% coverage if code as added
• Ensure existing tests and github automation passes (a maintainer will kick off the github automation when the rest of the PR is populated)

Attribution

• AI coding assistants used

[github-actions]

The PR description has been updated. Please fill out the template for your PR to be reviewed.

[akihikokuroda]

Thanks for your contribution. I updated the title and descriptions to make the CI works.

[psschwei]

I believe this line being here is what's causing the CI failure. Can you move this down to after all the imports?

[SAY-5]

Good catch, I'll move that import below the others and push.

[psschwei]

@SAY-5 looks like there's a test error

[SAY-5]

Good catch. The two failing assertions were checking that the raw exception text leaks into the client response, which is exactly what this PR removes. I flipped both checks to assert the opposite (raw exception text must NOT appear) and kept the generic message check. Pushed a50ea10 + 42f1988.

[SAY-5]

The logger = logging.getLogger(__name__) line is already after all imports (line 46, after the .utils import). The actual CI failure was the test assertions still expecting the raw exception text to appear in the response body, which I fixed in 42f1988 by updating both assertions to check that the raw text does NOT leak.

[psschwei]

We probably should use MelleaLogger instead of the python one

[SAY-5]

Switched to MelleaLogger.get_logger in d2080c9. The 19 serve tests still pass locally.

[psschwei]

Thanks @SAY-5 !