gh-102024: Reduced _idle_semaphore.release calls

[jackcvr]

_idle_semaphore.release() is called if only work_queue is empty

Benchmarks:
submit_only:

./python -m timeit -n 100000 -r 10 -s 'W = 1; from concurrent.futures import ThreadPoolExecutor; tpe = ThreadPoolExecutor(W)' 'tpe.submit(lambda: None)'

before:
W=1: 100000 loops, best of 10: 17.6 usec per loop
W=2: 100000 loops, best of 10: 10.5 usec per loop
W=4: 100000 loops, best of 10: 10.5 usec per loop
after:
W=1: 100000 loops, best of 10: 7.42 usec per loop
W=2: 100000 loops, best of 10: 7.5 usec per loop
W=4: 100000 loops, best of 10: 7.5 usec per loop

submit_then_wait:

./python -m timeit -n 10 -r 10 -s 'W = 1; from concurrent.futures import ThreadPoolExecutor, wait; tpe = ThreadPoolExecutor(W)' 'fs=[tpe.submit(lambda: None) for _ in range(10_000)]; wait(fs)'

before:
W=1: 10 loops, best of 10: 110 msec per loop
W=2: 10 loops, best of 10: 100 msec per loop
W=4: 10 loops, best of 10: 114 msec per loop
after:
W=1: 10 loops, best of 10: 85.2 msec per loop
W=2: 10 loops, best of 10: 83.9 msec per loop
W=4: 10 loops, best of 10: 86.5 msec per loop

• Issue: Reduce _idle_semaphore calls in ThreadPoolExecutor #102024

[bedevere-bot]

Most changes to Python require a NEWS entry.

Please add it using the blurb_it web app or the blurb command-line tool.

[cpython-cla-bot]

All commit authors signed the Contributor License Agreement.

[bedevere-bot]

Most changes to Python require a NEWS entry.

Please add it using the blurb_it web app or the blurb command-line tool.

[bedevere-bot]

Most changes to Python require a NEWS entry.

Please add it using the blurb_it web app or the blurb command-line tool.

[sobolevn]

Thanks! One more important thing is missing: benchmarks.
Can you please provide some small code sample that will show us that your solution is actually faster now?

[arhadthedev]

@sobolevn BTW is there any newcomer tutorial (edit: how-to) dedicated to pyperf one-liners? https://pyperf.readthedocs.io/en/latest/index.html seems to not serve this purpose.

[jackcvr]

Added some benchmarks. In my tests there is noticeable boost for submits to one thread.

[sobolevn]

@arhadthedev I don't know :(
I never needed one, since I've used pyperf before and quite a lot.

[hauntsaninja]

Thanks, this looks good to me. I have this exact same change in some custom thread pool code I use :-)

I'm a little confused by your benchmarking. I'm not sure why you'd expect the only submit case to speed up (and I can't repro a speed up on that case locally), without shutting down the executor. Of course, if you add the shutdown, I can measure a win (though ofc the size of win varies based on task count, task size, thread count).

[bedevere-bot]

🤖 New build scheduled with the buildbot fleet by @hauntsaninja for commit afbdad6 🤖

If you want to schedule another build, you need to add the 🔨 test-with-buildbots label again.

[miss-islington]

Thanks @jackcvr for the PR, and @gpshead for merging it 🌮🎉.. I'm working now to backport this PR to: 3.12.
🐍🍒⛏🤖

[bedevere-bot]

GH-104959 is a backport of this pull request to the 3.12 branch.