Martijn Visser created FLINK-41072:
--------------------------------------
Summary: Test configuration randomization on GitHub Actions always
uses the same seed
Key: FLINK-41072
URL: https://issues.apache.org/jira/browse/FLINK-41072
Project: Flink
Issue Type: Bug
Components: Build System / CI
Affects Versions: 2.4.0
Reporter: Martijn Visser
Assignee: Martijn Visser
On GitHub Actions the test configuration randomization never finds a seed.
PseudoRandomValueSelector seeds from test.randomization.seed, else from git
rev-parse HEAD, else from the commit id in EnvironmentInformation. job_init
moves only *, .scalafmt.conf and .mvn into the container, so /root/flink has no
.git and the compile in the same container leaves the commit id unknown as
well. Every GHA fork logs
{code}
WARN org.apache.flink.runtime.testutils.PseudoRandomValueSelector [] - Test
randomization was enabled but neither test.randomization.seed was configured
nor could the commit hash be retrieved from git or the EnvironmentInformation.
Please set the test.randomization.seed property manually to make the build
reproducible.
{code}
and falls back to the constant seed, so every GHA run on every branch draws the
same values for every randomized option, while Azure draws per commit. The
option combinations the randomization exists to cover never run on GHA.
We should either move .git along in job_init or pass
-Dtest.randomization.seed=${{ github.sha }} to the test run, and fail the run
when the warning fires.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
