wyndhblb commented on issue #10930: URL: https://github.com/apache/druid/issues/10930#issuecomment-792775848
Yes i don't think this is a druid issue, as the error originates in the kafka client itself. However if the issue is in the producer or brokers, is harder to say. I was looking for a way to "skip" the offending offsets. The issue being that if there are multiple peons (say 8) consuming the stream it only effects one of of the peons, and one of the partitions in that peon group, meaning the partitions it has been assigned get very far behind very fast, while the others move along. Resetting the offsets to "latest" is the only option current, but that will mean 1/8th of the partitions are basically very far behind while the others are not, and of course a slew of data i missed depending on the message rate. I was looking for a way to manually "skip" offsets, i tried the suggestion of `altering the backing offsets in the metadata store` from @JulianJaffePinterest, however there's a lock and a hash associated w/ the values stored, and so resetting them by hand caused even more trouble (all peons crashed and i was forced to move reset everything). So aside from wanting `isolation.level = read_uncommitted` perhaps another thing that's needed is a easy way to properly move offsets. I can see this being a useful thing in general, especially if producers make "other mistakes" (like KF message version that are incorrect, bad encodings, using compression that's unsupported, etc) ---------------------------------------------------------------- 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]
