Repository navigation
The PostCommit Python job is flaky #38691
Description
The PostCommit Python is failing over 50% of the time.
Please visit https://github.com/apache/beam/actions/workflows/beam_PostCommit_Python.yml?query=is%3Afailure+branch%3Amaster to see all failed workflow runs.
See also Grafana statistics: http://metrics.beam.apache.org/d/CTYdoxP4z/ga-post-commits-status?orgId=1&viewPanel=2&var-Workflow=PostCommit%20Python
Activity
shunping commentedon May 26, 2026
I think this is caused by https://github.com/apache/beam/pull/38611/changes#diff-cc1f3d7f808c692a6102847bec78809f2e4350c5ee34278100ce0f55d8c23d68
A fix has been proposed: #38688
aIbrahiim commentedon Jun 6, 2026
This PR #38839 fixes the Python 3.10 postcommit failure as the test failed because gemini-2.0-flash-001 was retired on Vertex AI and I updated the example to use gemini-2.5-flash which is still supported
shunping commentedon Jun 6, 2026
The segv on python 3.14 seems related to pandas version. Is there any version update on that?
I also see similar errors in #38838
aIbrahiim commentedon Jun 6, 2026
Thanks for flagging the segv and I had a look at pandas too but we’re still on pandas 2.2.3 for py3.14 (no py3.14-specific bump yet), and Beam currently caps pandas at <2.3. Since pandas 2.3.3 is the first version with official py3.14 support, a segfault on 3.14 could definitely be a pandas/version issue, separately, I was looking at another py3.14 failure in dataflow postCommitIT where pytest exits with code 5 and collects 0 tests xdist workers crash, and logs show warnings about @pytest.mark.it_postcommit on unittest.TestCase methods returning MarkDecorator that looks like a different problem from a segv but for #38838 (Xlang IO Direct), Im not sure yet which bucket it falls into @shunping
aIbrahiim commentedon Jun 8, 2026
3.14 passed for the fix in this pr #38844, @shunping
shunping commentedon Jun 8, 2026
Sounds promising. Let's see how other tests go.
github-actions commentedon Jun 10, 2026
Reopening since the workflow is still flaky
aIbrahiim commentedon Jun 10, 2026
it starts to fail probably because of #38798 as uv in installGcpTest uses a global cache, and parallel PostCommit Python jobs step on each other but #38862 fixes that with a per venv cache
github-actions commentedon Jun 24, 2026
Reopening since the workflow is still flaky