Close unclosed MemoryObjectReceiveStream in TestClient

[Moataz-E]

Summary

TestClient isn't closing stream_receive: StreamObjectStream correctly during shutdown. This has been confirmed by multiple people as part of discussion #2603. @markwaddle has identified that the issue is due to stream_receive not being disposed of correctly in the shutdown function. Adding stream_receive to the shutdown function's context manager resolved the issue as confirmed by myself and others in the discussion.

Checklist

• I understand that this PR may be closed in case there was no previous discussion. (This doesn't apply to typos!)
• I've added a test for each change that was introduced, and I tried as much as possible to make a single atomic change.
• I've updated the documentation accordingly.

[adriangb]

Looks good to me!

[Kludex]

We have an ignore of resource warning on the pyproject. Is this PR enough to remove that?

[abdullahsych]

@Kludex Any idea when this will be merged? We are currently working around this issue by pinning an older version of anyio.

[Kludex]

It would be cool to get my question replied...

[Moataz-E]

It would be cool to get my question replied...

I removed the ignore filter and re-ran the tests and can see 7 tests still fail because of this warning. So this only prevents the issue from propagating to people who are using TestClient.

@Kludex If you would like these issues resolved first before the PR is merged, let me know and I can take a look this evening.

[Kludex]

It would be cool to get my question replied...

I removed the ignore filter and re-ran the tests and can see 7 tests still fail because of this warning. So this only prevents the issue from propagating to people who are using TestClient.

@Kludex If you would like these issues resolved first before the PR is merged, let me know and I can take a look this evening.

Ideally, it would be cool to solve everything that was introduced by the bump of anyio on the project before merging this.

But... It's not a blocker...

[Moataz-E]

It would be cool to get my question replied...

I removed the ignore filter and re-ran the tests and can see 7 tests still fail because of this warning. So this only prevents the issue from propagating to people who are using TestClient.
@Kludex If you would like these issues resolved first before the PR is merged, let me know and I can take a look this evening.

Ideally, it would be cool to solve everything that was introduced by the bump of anyio on the project before merging this.

But... It's not a blocker...

No worries, I will take a look when I get back on my computer.

[Kludex]

Thanks.

[Kludex]

I'll make a release with this PR, but I think this will lower the motivation to find out why we have more unclosed memory objects... 😢

[alvindera97]

I had more warnings by a factor of about 4X...

Edit:
My mistake. I had to update starlette manually to the latest version (as of time of writing) 0.39.2 and now I don't have any warnings anymore.

[Moataz-E]

I had more warnings by a factor of about 4X

Which version were you running before and after? Can you provide more details about the warning numbers.

[alvindera97]

I had more warnings by a factor of about 4X

Which version were you running before and after? Can you provide more details about the warning numbers.

Updating anyio to the latest version (as of writing) 4.6.0 was what increased the warning by a factor of 4. I was previously on anyio version 4.4.0.

So updating to the latest version starlette got rid of all the warnings.

[Moataz-E]

@alvindera97 if the issue is related to anyio and not this change then please create a separate issue / discussion.

[juliangilbey]

I'm still seeing this error when testing fastapi using the following package versions:

• fastapi version 0.115.5
• starlette version 0.41.2
• anyio version 4.6.2
  So there still seem to be cases where this PR is insufficient to close the streams, unfortunately. But I have no idea how to track down the cause.

[Kludex]

If you read this PR, I mention this in the beginning.