-
Notifications
You must be signed in to change notification settings - Fork 3.2k
New issue
Have a question about this project? # for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “#”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? # to your account
Not switching to a leader without leader epochs while doing a cluster roll to upgrade the cluster #4796
Labels
Comments
To prevent this and improve availability when doing a cluster roll to upgrade to a version with KIP 320 we can make a change so validations are skipped when new leader doesn't support them, even if a previous leader did. |
7 tasks
We're having a similar problem. Is there any way to make consumer client keep work during the cluster rolling upgrade? |
# for free
to join this conversation on GitHub.
Already have an account?
# to comment
Description:
After upgrading Kafka from version 2.2.1 to 3.5.1, we observed issues related to leader switching. The specific issues and reproduction steps are as follows:
Process:
Expected Behavior:
In such cases, the metadata should update correctly, allowing proper leader switching and load balancing.
Actual Behavior:
The metadata fails to update correctly, causing all requests to be incorrectly routed to a specific node, forming an error loop.
Additional Information:
Relevant Kafka configuration and librdkafka version are as follows:
partial logger
code
The text was updated successfully, but these errors were encountered: