Fix missing self.servers exception

[Archmonger]

If another webserver is currently parked on uvicorn's port, then this exception will occur:

Traceback (most recent call last):
  File "C:\Users\Markg\Documents\Repositories\idom-docs\.venv\lib\site-packages\idom\backend\default.py", line 38, in serve_development_app
    return await _default_implementation().serve_development_app(
  File "C:\Users\Markg\Documents\Repositories\idom-docs\.venv\lib\site-packages\idom\backend\starlette.py", line 76, in serve_development_app
    await serve_development_asgi(app, host, port, started)
  File "C:\Users\Markg\Documents\Repositories\idom-docs\.venv\lib\site-packages\idom\backend\_common.py", line 53, in serve_development_asgi
    await asyncio.wait_for(server.shutdown(), timeout=3)
  File "C:\Users\Markg\AppData\Local\Programs\Python\Python310\lib\asyncio\tasks.py", line 445, in wait_for
    return fut.result()
  File "C:\Users\Markg\Documents\Repositories\idom-docs\.venv\lib\site-packages\uvicorn\server.py", line 258, in shutdown
    for server in self.servers:
AttributeError: 'Server' object has no attribute 'servers'

This PR ensures self.servers is always initialized so the error can be handled gracefully.

[Kludex]

What's your goal?

[Archmonger]

To prevent an unexpected exception from occuring if something is taking the requested Uvicorn port during startup.

Now, uvicorn will now properly log that the port is busy during the self-activated shutdown.

[Kludex]

Do you have a snippet that demonstrates what you mean?

[Archmonger]

Not exactly a snippet, but the process I've done to create this exception is:

1. Run another webserver on a specific port (ex. running mkdocs serve on port 8000)
2. Run uvicorn on an arbitrary ASGI application using that specific port. In my case, I was utilizing IDOM to run uvicorn

[Kludex]

I'd appreciate an MRE. 👀

[Archmonger]

Sure I'll draft one in a bit

[Kludex]

Sure I'll draft one in a bit

Ok 👀

[Archmonger]

Oops, thanks for the reminder.

MRE generated. See the readme for steps to reproduce.

https://github.com/Archmonger/uvicorn-server-exception

---

This issue can only be reproduced when uvicorn is run within asyncio.run(serve_development_asgi) in idom.

With the changes in this PR, everything seems to be handled gracefully.

[Kludex]

So it cannot be reproduced with uvicorn standalone?

[Archmonger]

Not from my one quick test.

serve_development_asgi seems to be a basic asyncio runner though.

Not entirely sure how that's causing this problem, but it does look like uvicorn's Server.shutdown() is making some unsafe assumptions that self.servers always exists. But self.servers will only exist if at least one successful Server.startup() has taken place.

[Kludex]

This is the first time I'm seeing this issue, and it doesn't look like is reproducible against uvicorn standalone... Even if it's one line, I need a proof that the issue is here, and I can only have that if it's possible to reproduce with only uvicorn.

[Archmonger]

No problem. I'm actually seeing that this issue is ultimately caused by this LOC in IDOM.

Although the variable safety issue still exists within Server.shutdown, it at least doesn't seem to be breaking things yet.

I will refactor the server running logic in IDOM.