Radosław Stankiewicz
Radosław Stankiewicz
For stack delta, you look at relevant commit - ontimer
Run Java PreCommit
I've looked at when we acknowledge messages. theory: 1. split was rebalanced 2. checkpoint was rejected, finalization wasn't invoked, msg.ack wasn't invoked 3. split reader started on different worker Solace...
probably related: https://community.solace.com/t/reprocess-messages-which-are-not-ack-back-by-consumer/307
@ppawel , how fast can you learn about data loss? if you restart workers (Cloud compute restart), are you able to see those messages processed?
ok, so previous [comment](https://github.com/apache/beam/issues/36991#issuecomment-3612775061) is invalid if the message that was lost was accepted.
it would be great if you raise the case and add me on cc ([email protected]) , I would love to see the logs where you see SolcaIO statements and missing...
Which runner you are using? Is this same data loss across both dataflow runners? The issue may be with checkpointing logic. During checkpoint creation receivedMessages message references are inserted into...
@ppawel are you able to test with snapshot version? https://github.com/apache/beam/pull/37007 fix is ready for review, once merged you will be able to quickly validate.