gh-82054: allow test runner to split test_asyncio to execute in parallel by sharding. #103927
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
All commit authors signed the Contributor License Agreement. |
|
@zitterbewegung The commits are created using the wrong e-mail. Could you |
…asyncio in parallel. By sharding the individual tests in test_multiprocessing and test_asyncio we are able to reduce considerably the time it takes to run the whole test suite.
zitterbewegung
force-pushed
the
gh-82054
branch
from
b366f9f
to
557e6b7
Compare
@arhadthedev Okay recreated and force pushed. |
gpshead
reviewed
Apr 29, 2023
There was a problem hiding this comment.
I think we've still got a while before this meaningfully reduces most CI system (presumed 2-4 thread) or developer (presumed 8-32 thread) whole test suite latency. With this PR on a fast system:
10 slowest tests:
- test_concurrent_futures: 2 min 14 sec
- test_multiprocessing_spawn: 1 min 25 sec
- test_multiprocessing_forkserver: 1 min 9 sec
- test_signal: 1 min 7 sec
- test_tools: 1 min 3 sec
- test_multiprocessing_fork: 1 min 3 sec
- test_socket: 37.6 sec
- test_imaplib: 30.6 sec
- test_io: 29.7 sec
- test.test_asyncio.test_tasks: 28.8 sec
Until Lib/test_concurrent_futures.py is broken up into a package of parallelizable items we've still got a pretty long tail.
The same really should be done with Lib/test/_test_multiprocessing.py as well, but it's a more complicated beast as it is a single test "template" that gets invoked with three different start method settings configured by the Lib/test/test_multiprocessing_{fork,forkserver,spawn}.py files.
Regardless this PR is still a win even with our longest test, test_asyncio, which would otherwise be 3 minutes above, split up. That reduces the total run time latency 10% on a CI-like -j 4 run.
zitterbewegung
changed the title
…ode cleanup. For the new test arguments they were made keyword only and also test_multiprocessing was removed from the SPLITTESTDIRS. The argument splittestdirs was renamed to split_test_dirs for findtests.
gpshead
approved these changes
Apr 29, 2023
carljm
added a commit
to carljm/cpython
that referenced
this pull request
May 1, 2023
* main: (26 commits) pythongh-104028: Reduce object creation while calling callback function from gc (pythongh-104030) pythongh-104036: Fix direct invocation of test_typing (python#104037) pythongh-102213: Optimize the performance of `__getattr__` (pythonGH-103761) pythongh-103895: Improve how invalid `Exception.__notes__` are displayed (python#103897) Adjust expression from `==` to `!=` in alignment with the meaning of the paragraph. (pythonGH-104021) pythongh-88496: Fix IDLE test hang on macOS (python#104025) Improve int test coverage (python#104024) pythongh-88773: Added teleport method to Turtle library (python#103974) pythongh-104015: Fix direct invocation of `test_dataclasses` (python#104017) pythongh-104012: Ensure test_calendar.CalendarTestCase.test_deprecation_warning consistently passes (python#104014) pythongh-103977: compile re expressions in platform.py only if required (python#103981) pythongh-98003: Inline call frames for CALL_FUNCTION_EX (pythonGH-98004) Replace Netlify with Read the Docs build previews (python#103843) Update name in acknowledgements and add mailmap (python#103696) pythongh-82054: allow test runner to split test_asyncio to execute in parallel by sharding. (python#103927) Remove non-existing tools from Sundry skiplist (python#103991) pythongh-103793: Defer formatting task name (python#103767) pythongh-87092: change assembler to use instruction sequence instead of CFG (python#103933) pythongh-103636: issue warning for deprecated calendar constants (python#103833) Various small fixes to dis docs (python#103923) ...
vstinner
pushed a commit
to vstinner/cpython
that referenced
this pull request
Sep 2, 2023
… parallel by sharding. (python#103927) This runs test_asyncio sub-tests in parallel using sharding from Cinder. This suite is typically the longest-pole in runs because it is a test package with a lot of further sub-tests otherwise run serially. By breaking out the sub-tests as independent modules we can run a lot more in parallel. After porting we can see the direct impact on a multicore system. Without this change: Running make test is 5 min 26 seconds With this change: Running make test takes 3 min 39 seconds That'll vary based on system and parallelism. On a `-j 4` run similar to what CI and buildbot systems often do, it reduced the overall test suite completion latency by 10%. The drawbacks are that this implementation is hacky and due to the sorting of the tests it obscures when the asyncio tests occur and involves changing CPython test infrastructure but, the wall time saved it is worth it, especially in low-core count CI runs as it pulls a long tail. The win for productivity and reserved CI resource usage is significant. Future tests that deserve to be refactored into split up suites to benefit from are test_concurrent_futures and the way the _test_multiprocessing suite gets run for all start methods. As exposed by passing the -o flag to python -m test to get a list of the 10 longest running tests. --------- Co-authored-by: Carl Meyer <[email protected]> Co-authored-by: Gregory P. Smith <[email protected]> [Google, LLC] (cherry picked from commit 9e011e7)
vstinner
added a commit
that referenced
this pull request
Sep 3, 2023
…108820) * Revert "[3.11] gh-101634: regrtest reports decoding error as failed test (#106169) (#106175)" This reverts commit d5418e9. * Revert "[3.11] bpo-46523: fix tests rerun when `setUp[Class|Module]` fails (GH-30895) (GH-103342)" This reverts commit ecb09a8. * Revert "gh-95027: Fix regrtest stdout encoding on Windows (GH-98492)" This reverts commit b2aa28e. * Revert "[3.11] gh-94026: Buffer regrtest worker stdout in temporary file (GH-94253) (GH-94408)" This reverts commit 0122ab2. * Revert "Run Tools/scripts/reindent.py (GH-94225)" This reverts commit f0f3a42. * Revert "gh-94052: Don't re-run failed tests with --python option (GH-94054)" This reverts commit 1347607. * Revert "[3.11] gh-84461: Fix Emscripten umask and permission issues (GH-94002) (GH-94006)" This reverts commit 1073184. * gh-93353: regrtest checks for leaked temporary files (#93776) When running tests with -jN, create a temporary directory per process and mark a test as "environment changed" if a test leaks a temporary file or directory. (cherry picked from commit e566ce5) * gh-93353: Fix regrtest for -jN with N >= 2 (GH-93813) (cherry picked from commit 36934a1) * gh-93353: regrtest supports checking tmp files with -j2 (#93909) regrtest now also implements checking for leaked temporary files and directories when using -jN for N >= 2. Use tempfile.mkdtemp() to create the temporary directory. Skip this check on WASI. (cherry picked from commit 4f85cec) * gh-84461: Fix Emscripten umask and permission issues (GH-94002) - Emscripten's default umask is too strict, see emscripten-core/emscripten#17269 - getuid/getgid and geteuid/getegid are stubs that always return 0 (root). Disable effective uid/gid syscalls and fix tests that use chmod() current user. - Cannot drop X bit from directory. (cherry picked from commit 2702e40) * gh-94052: Don't re-run failed tests with --python option (#94054) (cherry picked from commit 0ff7b99) * Run Tools/scripts/reindent.py (#94225) Reindent files which were not properly formatted (PEP 8: 4 spaces). Remove also some trailing spaces. (cherry picked from commit e87ada4) * gh-94026: Buffer regrtest worker stdout in temporary file (GH-94253) Co-authored-by: Victor Stinner <[email protected]> (cherry picked from commit 199ba23) * gh-96465: Clear fractions hash lru_cache under refleak testing (GH-96689) Automerge-Triggered-By: GH:zware (cherry picked from commit 9c8f379) * gh-95027: Fix regrtest stdout encoding on Windows (#98492) On Windows, when the Python test suite is run with the -jN option, the ANSI code page is now used as the encoding for the stdout temporary file, rather than using UTF-8 which can lead to decoding errors. (cherry picked from commit ec1f6f5) * gh-98903: Test suite fails with exit code 4 if no tests ran (#98904) The Python test suite now fails wit exit code 4 if no tests ran. It should help detecting typos in test names and test methods. * Add "EXITCODE_" constants to Lib/test/libregrtest/main.py. * Fix a typo: "NO TEST RUN" becomes "NO TESTS RAN" (cherry picked from commit c76db37) * gh-100086: Add build info to test.libregrtest (#100093) The Python test runner (libregrtest) now logs Python build information like "debug" vs "release" build, or LTO and PGO optimizations. (cherry picked from commit 3c89202) * bpo-46523: fix tests rerun when `setUp[Class|Module]` fails (#30895) Co-authored-by: Jelle Zijlstra <[email protected]> Co-authored-by: Łukasz Langa <[email protected]> (cherry picked from commit 9953860) * gh-82054: allow test runner to split test_asyncio to execute in parallel by sharding. (#103927) This runs test_asyncio sub-tests in parallel using sharding from Cinder. This suite is typically the longest-pole in runs because it is a test package with a lot of further sub-tests otherwise run serially. By breaking out the sub-tests as independent modules we can run a lot more in parallel. After porting we can see the direct impact on a multicore system. Without this change: Running make test is 5 min 26 seconds With this change: Running make test takes 3 min 39 seconds That'll vary based on system and parallelism. On a `-j 4` run similar to what CI and buildbot systems often do, it reduced the overall test suite completion latency by 10%. The drawbacks are that this implementation is hacky and due to the sorting of the tests it obscures when the asyncio tests occur and involves changing CPython test infrastructure but, the wall time saved it is worth it, especially in low-core count CI runs as it pulls a long tail. The win for productivity and reserved CI resource usage is significant. Future tests that deserve to be refactored into split up suites to benefit from are test_concurrent_futures and the way the _test_multiprocessing suite gets run for all start methods. As exposed by passing the -o flag to python -m test to get a list of the 10 longest running tests. --------- Co-authored-by: Carl Meyer <[email protected]> Co-authored-by: Gregory P. Smith <[email protected]> [Google, LLC] (cherry picked from commit 9e011e7) * Display the sanitizer config in the regrtest header. (#105301) Display the sanitizers present in libregrtest. Having this in the CI output for tests with the relevant environment variable displayed will help make it easier to do what we need to create an equivalent local test run. (cherry picked from commit 852348a) * gh-101634: regrtest reports decoding error as failed test (#106169) When running the Python test suite with -jN option, if a worker stdout cannot be decoded from the locale encoding report a failed testn so the exitcode is non-zero. (cherry picked from commit 2ac3eec) * gh-108223: test.pythoninfo and libregrtest log Py_NOGIL (#108238) Enable with --disable-gil --without-pydebug: $ make pythoninfo|grep NOGIL sysconfig[Py_NOGIL]: 1 $ ./python -m test ... == Python build: nogil debug ... (cherry picked from commit 5afe0c1) * gh-90791: test.pythoninfo logs ASAN_OPTIONS env var (#108289) * Cleanup libregrtest code logging ASAN_OPTIONS. * Fix a typo on "ASAN_OPTIONS" vs "MSAN_OPTIONS". (cherry picked from commit 3a1ac87) * gh-108388: regrtest splits test_asyncio package (#108393) Currently, test_asyncio package is only splitted into sub-tests when using command "./python -m test". With this change, it's also splitted when passing it on the command line: "./python -m test test_asyncio". Remove the concept of "STDTESTS". Python is now mature enough to not have to bother with that anymore. Removing STDTESTS simplify the code. (cherry picked from commit 174e9da) * regrtest computes statistics (#108793) test_netrc, test_pep646_syntax and test_xml_etree now return results in the test_main() function. Changes: * Rewrite TestResult as a dataclass with a new State class. * Add test.support.TestStats class and Regrtest.stats_dict attribute. * libregrtest.runtest functions now modify a TestResult instance in-place. * libregrtest summary lists the number of run tests and skipped tests, and denied resources. * Add TestResult.has_meaningful_duration() method. * Compute TestResult duration in the upper function. * Use time.perf_counter() instead of time.monotonic(). * Regrtest: rename 'resource_denieds' attribute to 'resource_denied'. * Rename CHILD_ERROR to MULTIPROCESSING_ERROR. * Use match/case syntadx to have different code depending on the test state. Co-authored-by: Alex Waygood <[email protected]> (cherry picked from commit d4e534c) * gh-108822: Add Changelog entry for regrtest statistics (#108821) --------- Co-authored-by: Christian Heimes <[email protected]> Co-authored-by: Zachary Ware <[email protected]> Co-authored-by: Nikita Sobolev <[email protected]> Co-authored-by: Joshua Herman <[email protected]> Co-authored-by: Gregory P. Smith <[email protected]>
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Summary:
This runs test_asyncio sub-tests in parallel using sharding from Cinder. This suite is typically the longest-pole in runs because it is a test package with a lot of further sub-tests otherwise run serially. By breaking out the sub-tests as independent modules we can run a lot more in parallel.
After porting we can see the direct impact on a multicore system.
The drawbacks are that this implementation is hacky and due to the sorting of the tests it obscures when the asyncio tests occur and involves changing CPython test infrastructure but, the wall time saved it is worth it, especially in low-core count CI runs as it pulls a long tail. The win for productivity and reserved CI resource usage is significant.
Future tests that deserve to be refactored into split up suites to benefit from are
test_concurrent_futuresand the way the_test_multiprocessingsuite gets run for all start methods. As exposed by passing the-oflag topython -m testto get a list of the 10 longest running tests.