kamalcph commented on PR #21361:
URL: https://github.com/apache/kafka/pull/21361#issuecomment-3809041483

   Thanks for the PR, @jiafu1115. 
   
   1. We cannot delete the segments in-middle when the `maxTimestamp` of the 
candidate segment is less than the `retention.ms` and move the 
log-start-offset. This will break the deletion breach by retention time logic. 
In 
[UnifiedLog#deleteRetentionMsBreachedSegments](https://sourcegraph.com/github.com/apache/kafka/-/blob/storage/src/main/java/org/apache/kafka/storage/internals/log/UnifiedLog.java?L1973:17-1973:50),
 the segments are being scanned from the beginning of the log. We cannot expect 
that the maxTimestamp in the segments will increase monotonically.
   
   2. This optimization can be applied only to the deletion breached by size 
logic and can be moved to UnifiedLog. Similar to the `DELETE_RECORDS` API,  
that moves the log-start-offset. The segment deletion logic due to breach by 
size can increment the log-start-offset instead of local-log-start-offset when 
the `local-log-size > complete-retention` time. See 
[UnifiedLog](https://sourcegraph.com/github.com/apache/kafka/-/blob/storage/src/main/java/org/apache/kafka/storage/internals/log/UnifiedLog.java?L1927)
 for more details. 


-- 
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]

Reply via email to