global logging breaks when uvicorn.run(reload=True)

[Lx]

I have a script that sets up logging via logging.basicConfig, then starts Uvicorn using uvicorn.run.

If I don't use reload=True in the uvicorn.run call, logging works as expected.

If I do use reload=True in the uvicorn.run call:

• the root logger (belonging to logging) gets obliterated (its id even changes)
• all custom logging configuration is lost
• all existing loggers revert to printing just their messages to stderr.

This problem exists in the latest released version at the time of this discussion being opened: 0.17.6.

Am I okay to raise this as a "real" issue?

---

One further detail in case it helps: the problem exists with or without installation of watchgod.

[Kludex]

Do you have a repository/gist so I can reproduce that? 🙏

[Kludex]

This #328 ?

[Lx]

Probably the minimal useful example (which I appreciate is still not very useful) would be:

logging.basicConfig(
    level="NOTSET",
    format="[%(name)s] %(message)s",
)
uvicorn.run(
    app="thingo.server:app",
    host="0.0.0.0",  # nosec
    reload=True,  # TODO: work out why "reload=True" breaks logging
)

In my scenario, and in case it's relevant, things.server:app is a FastAPI app. I can't offer a repo URL as it's a private project.

[Lx]

It doesn't seem to be the same problem as #328, because no exception is thrown and I'm using Python 3.10.4 for this project.