hubcio opened a new pull request, #2958: URL: https://github.com/apache/iggy/pull/2958
During journal.commit() -> async persist, new appends make the journal non-empty while the in-flight buffer holds old committed data. The polling router only checked in-flight when the journal was empty (Case 0), so committed-but-unpersisted messages were invisible to consumers in Cases 1-3. With auto-commit, Case 3 returned only journal messages (higher offsets), skipping the in-flight range permanently - the "jump to end" symptom reported by users with ~50M records. Replace the case-based routing with a three-phase priority lookup (disk -> in-flight -> journal) that checks all tiers regardless of journal state. Also fix the (count-1).max(1) off-by-one that inflated end_offset when count=1, and add in-flight awareness to load_segment_messages. Fixes #2715 and part of same problem reported in #2924. -- 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]
