alanlau28 opened a new pull request, #23285: URL: https://github.com/apache/kafka/pull/23285
Jira: https://issues.apache.org/jira/browse/KAFKA-20940 The windowed changelog restore optimization added in KAFKA-13499, which seeks past already-expired records instead of restoring them, only ran for partitions with no checkpointed offset; whenever a checkpoint existed the reader sought straight to it and restored every record after it, replaying a prefix of data that retention had already expired. This extends the optimization to checkpointed active windowed changelogs whose checkpoint is far enough behind the end offset to be worth it, running them through the same probe but floored at the checkpoint so restoration never rewinds past records that have already been applied. When the probe cannot place a partition, or the checkpoint sits below the log start or the log is empty, restore falls back to the checkpoint rather than the beginning and leaves the existing truncation and corruption handling unchanged. -- 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]
