KKcorps commented on PR #20161: URL: https://github.com/apache/pinot/pull/20161#issuecomment-6680080992
**The model says this clears the restart it targets. A server upgraded from the old snapshot writer no longer restores a stale validDocIds bitmap as out of TTL, so a key stops showing twice. It found nothing new.** I replayed the model's known bad orders with and without the change, and only one that an earlier fix already closed changes. A 61 minute search over 43 table shapes found nothing new. In 104 more searches that began right after the upgrade, today's code showed the key twice and the change never did. I have one question from reading the code, which the model cannot judge. The snapshot check runs when a consumer starts, right after the previous segment's offload releases the permit inside its ONLINE transition, so ZK may still show that segment as CONSUMING. On a full upsert table, whose first consumer does not wait for loading, could every later check miss and keep snapshots and TTL cleanup off for good? The model covers one partition and judged the duplicate on partial upsert only. It sees an ERROR segment hold snapshots and cleanup, but not that the hold spans every partition. <!-- upsert-model-check head=507d55ca7240b1cc7db6b6fb6e3dd87e656e96d3 --> -- 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]
