merlimat opened a new pull request #7065: URL: https://github.com/apache/pulsar/pull/7065
### Motivation When a namespace bundle split operation fails by timeout, we're not cleaning up the current in-memory metadata of the old bundles. This lead to the broker to treat the operation as "non-succeeded" while in fact it might have gone through. In this case, the broker keep thinking that it has to split a particular bundle (based on the old load report information) and that fails with ZK bad-version errors, and again we're not cleaning up the stale metadata. ### Modifications Reverse the logic to first remove the bundle load data and then perform the split. If the split were to fail, its load data will be re-added in any case with the next broker load report. This will ensure that the leader load manager won't be keeping track of a non-existing bundle after a timeout in the split operation. ---------------------------------------------------------------- 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]
