FrankChen021 commented on PR #20382:
URL: https://github.com/apache/druid/pull/20382#issuecomment-5770275619

   > > Finding: When druid.storage.zip=false, pushNoZip uploads each file 
directly under the final segment prefix and only then lists and deletes stale 
objects. If a historical or another pusher reads the same path during a retry 
or same-path rewrite, it can observe missing new files, old chunks, or a 
mixture of two segment versions; the puller can then materialize a corrupt or 
incomplete segment even though the push later succeeds.
   > 
   > I don't think this is really a problem, is it? Historicals won't ever be 
pulling a segment that didn't publish successfully (load spec doesn't exist in 
metadata until pushes all suceed), ingest time replicas I don't believe will 
publish at the same time (and they should be producing identical contents), and 
the only real replace case I'm aware of where content could be different is 
like when streaming failed to publish and has to retry and while starting at 
the same offsets, might end at different offsets, but that runs with 
useUniquePath=true so avoid the problem.
   
   You're right. Sometimes AI does not correctly under the business flow.


-- 
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]


---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]

Reply via email to