Remove process-id checks from asyncio. Asyncio and fork() does not mix.

[kristjanvalur]

Pull Request check-list

• Does $ tox pass with this change (including linting)?
• Do the CI tests pass with this change (enable it first in your forked repo and wait for the github action build to finish)?
• Is the new or changed code fully tested?
• Is a documentation update included (if this change modifies existing APIs, or introduces new ones)?
• Is there an example added to the examples folder (if applicable)?
• Was the change added to CHANGES file?

Description of change

Remove PID checks from asyncio connections, since asyncio and fork() does not mix. See issue #2910

Details:

Python does not support fork() on a process with a running asyncio event loop. Therefore supporting a connection pool of connection from a different process is not necessary.

Similarly, connection pools don't need to deal with different threads either.

We remove the checks for process ids, intended to ignore inherited file descriptors (even though these should probably have been closed on fork anyway).

We remove lenience when returning a connection to a pool, it should always be returned to the pool it originated from. Not doing that would be a resource leak.

We simplify BlockingConnectionPool since it does not need to do any multi-threading. It now just adds a simple wait mechanism on top of the ConnectionPool, reducing code and overhead. The QueueClass argument is now ignored. It always was a weird implementation leak anyway

[codecov-commenter]

Codecov Report

Patch coverage: 60.52% and project coverage change: -24.53% ⚠️

Comparison is base (1596ac6) 91.28% compared to head (d78b85e) 66.75%.

❗ Your organization needs to install the Codecov GitHub app to enable full functionality.

Additional details and impacted files

@@ Coverage Diff @@ ## master #2911 +/- ## =========================================== - Coverage 91.28% 66.75% -24.53%  =========================================== Files 126 126 Lines 32398 32331 -67 =========================================== - Hits 29573 21584 -7989  - Misses 2825 10747 +7922 

Files Changed	Coverage Δ
tests/test_asyncio/conftest.py	49.30% <7.14%> (-35.59%)	⬇️
tests/conftest.py	72.89% <20.00%> (-13.14%)	⬇️
tests/test_asyncio/test_connection_pool.py	66.60% <50.00%> (-25.55%)	⬇️
redis/asyncio/connection.py	79.41% <81.81%> (-3.41%)	⬇️
tests/test_asyncio/test_cluster.py	96.93% <100.00%> (ø)
tests/test_asyncio/test_cwe_404.py	71.42% <100.00%> (-28.58%)	⬇️

... and 76 files with indirect coverage changes

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

[chayim]

@kristjanvalur Just checking in - is this one something you wanted reviewed?

@dvora-h see the test failures. I'm surprised that some of these (cluster) ran.

[kristjanvalur]

This is not reproducible here with individual steps.
It seems to be some fallout from unittests where the connection pool gets polluted with Mock Connections. One of the changes I did was to not allow returning unknown connections to the pool, since that is a programming error / resource leak. But the unit-tests seem to have been relying on that behaviour.

I'll figure it out and let you know when I think I have something which needs can be reviewed

[kristjanvalur]

@chayim Found the issue. Some mocks were not being undone properly which played havoc with the cleanup. Last two commits are test fixes can be applied to main independently, I can create a separate PR for that if you like.
Please review when you have the time.