docs: Update default middleware setup example

[sondrelg]

It makes sense to want the default setup to cover 500-range responses.

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 96.59%. Comparing base (974653d) to head (69e05d2).

Additional details and impacted files

@@           Coverage Diff           @@
##             main     #109   +/-   ##
=======================================
  Coverage   96.59%   96.59%           
=======================================
  Files          11       11           
  Lines         440      440           
=======================================
  Hits          425      425           
  Misses         15       15           

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[sebastian-correa]

My point about the type [ref] is that if we use this pattern, the app stops being a FastAPI instance.

This is ultimately user choice, since it's only a documentation change, but it may introduce lots of squigglies in code.

You can cast it with typing.cast, but that's a bit hacky.

Maybe we need to move from a dataclass to a normal class and use some typing magic, or add a from_app classmethod whose typehint is still FasAPI, or look into using the post_init to auto register the handler for errors, or something.

[microwavenby]

Maybe we need to move from a dataclass to a normal class and use some typing magic, or add a from_app classmethod whose typehint is still FasAPI, or look into using the post_init to auto register the handler for errors, or something.

It definitely results in squigglies when it comes to trying to access the test_client in pytest --

AttributeError: 'CorrelationIdMiddleware' object has no attribute 'test_client'

[sondrelg]

I both think:

• Changing the type of app will break a lot of tooling and seems close to a hard blocker
• Returning request IDs for 500s is pretty much always desirable

Do you reckon there's a way to satisfy both of these, or is it a hard trade-off?

[microwavenby]

FastAPI's docs include support for an exception handler:
https://fastapi.tiangolo.com/tutorial/handling-errors/#install-custom-exception-handlers

And, it's possible to override the default exception handlers
https://fastapi.tiangolo.com/tutorial/handling-errors/#override-the-default-exception-handlers

I'm guessing the right way to go about this is to also add an exception handler that generates or repeats the correlation-id from the request headers into the logs and/or response?

[sebastian-correa]

Sorry, I'm quite late on this. GitHub notifications don't work well for me.

I proposed an idea in #114. I think it ticks both boxes @sondrelg proposed (which I think should be what we aim for).