Repository navigation
The PostCommit Java PVR Flink Batch job is flaky #38934
Description
The PostCommit Java PVR Flink Batch is failing over 50% of the time.
Please visit https://github.com/apache/beam/actions/workflows/beam_PostCommit_Java_PVR_Flink_Batch.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=5&var-Workflow=PostCommit%20Java%20PVR%20Flink%20Batch
Activity
Kriti-dev07 commentedon Jun 13, 2026
I investigated the flaky failure and found it is consistently associated with
ParDoTest$StateTests.testSideInputNotReadyTimer, which was recently introduced as part of the side-inputs-in-timers work. The test relies on a delayed side input (Thread.sleep) and has already been excluded from Spark runners because side inputs in timers are not supported there.I'm currently checking whether this is a Flink runner limitation, a runner bug, or test flakiness before proposing a fix.
Amar3tto commentedon Jun 18, 2026
Cause: #38363
aIbrahiim commentedon Oct 16, 2026
#40324 set beam.test.parallelThreads=8 and Flink portable VR picked it also, so we now run up to 8 tests in one JVM with a 4g heap so each test starts a job server and loads jars which can OOM or hang for 20 mins so ViewTest is probably just the one that was running when that happened and spark already limits this to 1 thread @Abacn @reuvenlax