tillrohrmann commented on pull request #15019: URL: https://github.com/apache/flink/pull/15019#issuecomment-786772737
The semantics of the execution attempts is a very good question. I think I would keep it as you've implemented it in the PR: We keep monotonously increasing attempt counters for all ever seen `Executions`. * In case of a scale up, new `Executions` with an attempt number `0` will be inserted * In case of a scale down, some of the `Executions` won't be restarted and, thus, their attempt counter stays as is * Increasing the attempt counter for those `Executions` which will be restarted. Concerning the implementation, have you considered not introducing the counter to the `ExecutionGraph` but rather maintaining it outside? What this would require is the following: 1) A way to give an immutable map of attempts to the EG when it is created in order to initialize the `Executions`. 2) A way to retrieve the set of last attempt counters when the EG is terminated. This value could be used to update the external attempt counter. ---------------------------------------------------------------- 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. For queries about this service, please contact Infrastructure at: [email protected]
