Repository navigation
sys.settrace dramatic slowdown in 3.12 #107674
Description
Bug report
Checklist
- I am confident this is a bug in CPython, not a bug in a third-party project
- I have searched the CPython issue tracker, and am confident this bug has not been reported before
A clear and concise description of the bug
A bug report in coverage (coveragepy/coveragepy#1665) is reduced here to a pure do-nothing trace function for sys.settrace.
Reproduction:
% git clone https://github.com/nedbat/ndindex
% cd ndindex
% python3.11 -VV
Python 3.11.4 (main, Jun 7 2023, 08:42:37) [Clang 14.0.3 (clang-1403.0.22.14.1)]
% python3.11 -m venv v311
% ./v311/bin/pip install numpy
% python3.12 -VV
Python 3.12.0b4 (main, Jul 11 2023, 20:38:26) [Clang 14.0.3 (clang-1403.0.22.14.1)]
% python3.12 -m venv v312
% ./v312/bin/pip install --pre --extra-index https://pypi.anaconda.org/scientific-python-nightly-wheels/simple numpy
% # Run the code without the trace function.
% for x in 1 2 3; time ./v311/bin/python -m ndindex.tests.justdoit notrace
./v311/bin/python -m ndindex.tests.justdoit notrace 7.05s user 0.37s system 108% cpu 6.813 total
./v311/bin/python -m ndindex.tests.justdoit notrace 7.04s user 0.34s system 109% cpu 6.724 total
./v311/bin/python -m ndindex.tests.justdoit notrace 7.00s user 0.35s system 109% cpu 6.696 total
% # 3.12 is slightly faster without the trace function.
% for x in 1 2 3; time ./v312/bin/python -m ndindex.tests.justdoit notrace
./v312/bin/python -m ndindex.tests.justdoit notrace 6.56s user 0.30s system 106% cpu 6.422 total
./v312/bin/python -m ndindex.tests.justdoit notrace 6.48s user 0.31s system 106% cpu 6.359 total
./v312/bin/python -m ndindex.tests.justdoit notrace 6.38s user 0.28s system 107% cpu 6.217 total
% # 3.11 with tracing is about 3x slower.
% for x in 1 2 3; time ./v311/bin/python -m ndindex.tests.justdoit trace
./v311/bin/python -m ndindex.tests.justdoit trace 22.64s user 0.51s system 101% cpu 22.772 total
./v311/bin/python -m ndindex.tests.justdoit trace 21.92s user 0.46s system 101% cpu 21.979 total
./v311/bin/python -m ndindex.tests.justdoit trace 21.55s user 0.35s system 102% cpu 21.379 total
% # 3.12 with tracing is 7x slower.
% for x in 1 2 3; time ./v312/bin/python -m ndindex.tests.justdoit trace
./v312/bin/python -m ndindex.tests.justdoit trace 49.47s user 0.40s system 100% cpu 49.676 total
./v312/bin/python -m ndindex.tests.justdoit trace 49.53s user 0.39s system 100% cpu 49.784 total
./v312/bin/python -m ndindex.tests.justdoit trace 50.44s user 0.38s system 100% cpu 50.739 total
I don't know if the unusual numpy build has something to do with this.
Linked PRs
- GH-107674: Avoid allocating boxed ints for
sys.settraceline events #107780 - [3.12] GH-107674: Avoid allocating boxed ints for
sys.settraceline events (GH-107780) #107841 - gh-107674: Improve performance of
sys.settrace#114986 - gh-107674: Improve performance of
sys.settrace#117133 - gh-107674: Remove some unnecessary code in instrumentation code #117393
- gh-107674: Lazy load line number to improve performance of tracing #118127
Activity
neonene commentedon Aug 6, 2023
This issue can be seen starting from the commit 411b169 with usual numpy-1.25.2.
AlexWaygood commentedon Aug 7, 2023
cc. @markshannon
GH-107674: Avoid allocating boxed ints for `sys.settrace` line events (…
pythonGH-107674: Avoid allocating boxed ints for `sys.settrace` line …
nedbat commentedon Oct 24, 2023
@markshannon Is there any news about this?
nedbat commentedon Oct 30, 2023
@iritkatriel perhaps you have some idea?
iritkatriel commentedon Oct 31, 2023
Did #107780 help?
gaogaotiantian commentedon Feb 4, 2024
Hi @nedbat , I circled back to this issue. If you are still interested, could you try the patch of #114986 and see if the result is improved?
nedbat commentedon Feb 15, 2024
@gaogaotiantian Thanks. I tried #114986, and it didn't run faster. I used the https://github.com/Quansight-Labs/ndindex repo test suite:
3.11: 160s
3.12: 255s
3.13 main: 248s
pr114986: 245s
I wish I had better news.
gaogaotiantian commentedon Feb 15, 2024
Oh, did you test it with an empty trace function?
nedbat commentedon Feb 15, 2024
I will do more tests, but the ultimate goal is to perform better in real-world scenarios. Here is more data from a subset of the test suite (
-k test_chunking), which looks much more promising. It's hard to get consistent numbers, but this looks good. I would merge this change.gaogaotiantian commentedon Feb 15, 2024
Thank you for the test. Yes the PR only improves the performance of the actual trace function - which is normally the major cost for tracing tools. Good to know it helps with the more real-life benchmark. Unfortunately, due to the change of the tracing mechanism, the overhead for
sys.settracewill always be larger than the old way - I did not see a quick solution to that.45 remaining items
gaogaotiantian commentedon Sep 23, 2024
Thanks @nedbat . I'm a bit surprised that on some cases 3.13 is slower than 3.12 - I did not think of anything that we did to slow down the tracing. Glad to see we had some significant improvements on certain cases.
nedbat commentedon Feb 12, 2025
I continue to get people reporting serious performance problems on 3.12. Is there any chance of an improvement?
junkmd commentedon Feb 14, 2025
Considering the issue observed in enthought/comtypes#775, where execution time increased nearly tenfold, there is a performance "regression" in Python 3.12.
Only one bugfix release remains for Python 3.12:
https://pythonpeps.python.org/pep-0693/
If #107841 is not merged by then, I believe the Python community will miss an opportunity to reconsider this issue.
If "the fix patch" is merged before Python 3.12 transitions to security-fix only status, I would accept enabling coverage measurement in the
comtypesCI tests for all Python versions (and waiting until 3.12.10 is released). Otherwise, only Python 3.12 will be tested without coverage measurement in thecomtypesCI pipeline.As discussed in #107841, I understand that this issue doesn't involve "broken" programs but rather a "performance improvement", and that the changes are complex to accept during the bugfix phase. However, fixing this could potentially benefit projects and the wider ecosystem running on Python 3.12.
I would like to, once again, hear the opinions of the core developers and the community on this matter.
gaogaotiantian commentedon Feb 14, 2025
Ah, @markshannon merged #128350 in main but did not backport it to 3.12 and 3.13 due to conflicts. #107841 is not the problem fixer. I believe backporting #128350 should fix the issue in 3.12 and 3.13.
poc: python/cpython#107674
poc:
junkmd commentedon Feb 18, 2025
I tested whether #107841 is the problem fixer.
I built Python from
v3.12.8and also from the3.12branch with 37d8b90 cherry-picked.And I compared the execution time of
comtypestest cases.The build with 37d8b90 changes is about 153s faster than
v3.12.8.However, this performance improvement is still not significant.
I haven't tested backporting #128350 yet because I'm not sure how to resolve conflicts.
python 3.12 coverage regression workaround
python 3.12 coverage regression workaround
[3.12] GH-107674: Avoid allocating boxed ints for `sys.settrace` line…
gaogaotiantian commentedon Apr 25, 2025
I think we can close this issue now. The fix did not make 3.12 as it missed the last bug fix release. It was backported to 3.13 though. I don't think we have more tasks for this issue anymore.
jack-mcivor commentedon Jun 17, 2025
Is this true? I found this issue from the changelog of 3.12.10, which seems to suggest the fix made it in?
improve performance of tests in Python>=3.12 when used with coverage
improve performance of tests in Python>=3.12 when used with coverage
corneliusroemer commentedon Jul 20, 2025
Yep, closed as not planned/won't fix. The 3.13 performance looks as good as 3.11, even better in my own benchmarks against https://github.com/nextstrain/augur with pytest on an M4 Max using conda-forge binaries.
Python 3.13.5
Python 3.12.11
Python 3.11.13