DanielLeens commented on issue #12492: URL: https://github.com/apache/seatunnel/issues/12492#issuecomment-5855680322
Classification: A / Zeta WAL failure containment. Thanks for tracing the residual path. I checked current dev and the proposed baseline before accepting a direction: #12081 is still open, BLOCKED, and has a failed Build, so its Exception fail-close and timed-wait behavior is not yet a merged dev contract. Please do not start a dependent implementation until that PR is green, merged, and the follow-up is rebased on it. The goal here is valid, but I cannot approve a blanket Exception-to-Throwable change that keeps processing after VirtualMachineError, ThreadDeath, or LinkageError. In particular, continuing a consumer after OutOfMemoryError can make file and process state unsafe. A revised design should instead state the behavior by failure class: fatal VM errors must fail the WAL path loudly and keep it permanently unavailable (with best-effort completion of the affected request) before propagating to the controlled node-failure path; do not silently resume the worker or reopen the writer in process. Any Error made recoverable must be explicitly enumerated and justified. Before a PR, please also define the exact Disruptor exception-handler contract: how it records the permanent failure, what already-queued and later APPENDs observe, and which failures it must rethrow. The regression should inject a safe synthetic Error and prove that callers receive the permanent-failure result rather than only timing out; it must not use a real heap-exhaustion test. Keep worker restart, writer reopen, and changes to #12081 out of scope. No assignment or label change in this pass. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected]
