Jose Luis López created HADOOP-19981:
----------------------------------------
Summary: Re-enable GHA-excluded tests whose failure modes have
been fixed
Key: HADOOP-19981
URL: https://issues.apache.org/jira/browse/HADOOP-19981
Project: Hadoop Common
Issue Type: Task
Components: common, hdfs, test
Reporter: Jose Luis López
h3. Background
{\{.github/gha-tests/exclude-tests.txt}} lists the test classes Surefire skips
in
the GitHub Actions build, wired in via \{{-Dsurefire.excludesFile}} in
{\{.github/workflows/tmpl_build_and_test.yml}}. It holds 143 entries at the time
of filing.
{\{.github/gha-tests/README.md}} states the policy in both directions: add a
test
when it fails or aborts, repeat until five consecutive successes; and
"Contributors are encouraged to diagnose and improve the excluded tests, and
remove them from the excluded list once they are stable... when deleting tests
from exclude-tests.txt, the GHA workflow was successfully executed 5 times
consecutively."
h3. Problem
Several classes excluded for hangs, races and unrealistic per-method timeouts
have since
been fixed by HDFS-17957, HDFS-17972 and HADOOP-19979, yet remain excluded.
h3. Scope
Work the list down in small, reviewable batches. One subtask per batch. Each
subtask must:
# Name the specific failure mode and the JIRA that fixed it.
# Carry any residual fix the batch still needs - a timeout left behind, a leak
the original fix did not cover.
# Remove only the entries it can justify.
# Be validated by five consecutive successful GHA runs before commit, per the
README.
h3. Planned Removals
* TestDFSClientRetries and TestBalancerWithHANameNodes (hadoop-hdfs), enabled
by HDFS-17972 and HDFS-17957.
* TestLogAggregationService (hadoop-yarn-server-nodemanager), enabled
by HADOOP-19979.
* TestFsVolumeList (hadoop-hdfs) and TestRouterNamenodeHeartbeat
(hadoop-hdfs-rbf), both enabled by HDFS-17957, deliberately placed in
different matrix entries.
* TestMultipleArchiveFiles (hadoop-streaming), enabled by
HADOOP-19963. Best preceded by moving its MiniDFSCluster and MiniMRCluster
construction out of the constructor into \{{@BeforeEach}}, so a constructor
failure cannot bypass teardown.
--
This message was sent by Atlassian Jira
(v8.20.10#820010)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
