The GitHub Actions job "Flink CDC CI" on flink-cdc.git/FLINK-40040 has failed.
Run started by GitHub user Zhile (triggered by leonardBang).

Head commit for run:
9031b6b4bd8f0320eee49d55f15a0880caec5a97 / Jiyong Wang <[email protected]>
[FLINK-40040][tests] Add pipeline E2E case for coordinator failure with 
JobManager-only exception class

Reproduces the production classpath asymmetry: a jar containing
JobManagerOnlyException is installed into the JobManager container's
/opt/flink/lib only, never shipped to the TaskManager container or
submitted with the pipeline jars. The values sink gains an
error.on.schema.change.exception.class option so its metadata applier
raises that class (instantiated reflectively on the coordinator) as the
cause of a rejected schema change.

Verified against the pre-fix runtime: without the SerializedThrowable
wrapping in SchemaRegistry.failJob, the exceptionally completed
coordination response cannot be deserialized by the TaskManager's RPC
layer, which tears down the JobManager <-> TaskManager Pekko association
("Association with remote system ... has failed, address is now gated.
Reason: [org.apache.flink.cdc...Exception]"). Cancellation then stalls
for a full RPC timeout cycle before the job reaches FAILED, and the
original cause never surfaces on the TaskManager. The test asserts the
job fails cleanly and that the TaskManager's RPC layer never chokes on
the coordinator failure; it fails on the pre-fix runtime and passes
after the fix (19s vs 79s runtime).

Report URL: https://github.com/apache/flink-cdc/actions/runs/28916700495

With regards,
GitHub Actions via GitBox

Reply via email to