JulianJaffePinterest commented on issue #10930: URL: https://github.com/apache/druid/issues/10930#issuecomment-792398881
I've caused similar bugs in the past while testing failure handling with Flink and similar systems. The default Kafka consumer, which Druid and most other projects use under the hood, intentionally blocks forever when it encounters invalid records. I've worked around this in the past by just blindly advancing offsets until we re-encountered parseable offsets, but that was in a context without strict guarantees. Druid could implement a similar solution (probably in `SeekableStreamIndexTaskRunner`, judging by the stack trace, but I haven't looked at the code). The tricky part will be supporting multiple different delivery semantics. Hopefully your intuition is right and standardizing the isolation level works for you, but until then manually advancing the offsets (either by resetting the supervisor to the latest offsets and thus dropping data or by altering the backing offsets in the metadata store) is probably your only option. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
