Hi Jakub, as you say, this was supposed to work according to
documentation and KIP-778. Instead, the "unsafe" override is not yet
implemented. For this reason, I logged a bug and proposed a
documentation fix.

https://issues.apache.org/jira/browse/KAFKA-15537

Thanks for raising.

Br
Fede

On 2023/10/01 21:19:59 Jakub Scholz wrote:
> Hi,
>
> I have a Kraft cluster running Kafka 3.5 and using the metadata.version 3.5
> as well:
>
> $ bin/kafka-features.sh --bootstrap-server localhost:9092 describe
> Feature: metadata.version       SupportedMinVersion: 3.0-IV1
>  SupportedMaxVersion: 3.5-IV2    FinalizedVersionLevel: 3.5-IV2  Epoch: 140
>
> The upgrade docs <https://kafka.apache.org/documentation/#upgrade_350_kraft>
> suggest that I should be able to downgrade the metadata version as long as
> it is not to 3.0-3.2 versions. However, whatever version I try to downgrade
> it to and whether it is safe or unsafe downgrade, it always seems to fail:
>
> $ ./bin/kafka-features.sh --bootstrap-server localhost:9092 downgrade
> --metadata 3.4
> Could not downgrade metadata.version to 8. Invalid metadata.version 8.
> Refusing to perform the requested downgrade because it might delete
> metadata information. Retry using UNSAFE_DOWNGRADE if you want to force the
> downgrade to proceed.
> 1 out of 1 operation(s) failed.
> $ ./bin/kafka-features.sh --bootstrap-server localhost:9092 downgrade
> --metadata 3.4 --unsafe
> Could not downgrade metadata.version to 8. Invalid metadata.version 8.
> Unsafe metadata downgrade is not supported in this version.
> 1 out of 1 operation(s) failed.
>
> And it does not really matter whether I try to downgrade to 3.4, 3.4-IV0,
> 3.5-IV1. In the logs, I see the following INFO message while running the
> downgrade command:
>
> 2023-10-01 20:59:59,910 INFO Unable to get feature level information for
> controller(s): 4, 5 (org.apache.kafka.controller.QuorumFeatures)
> [quorum-controller-3-event-handler]
>
> Not sure if that is related. The cluster seems to work fine apart from the
> metadata downgrade. (The cluster I was trying it on has 3 controllers with
> node IDs 3, 4, and 5)
>
> Anyone has an idea what I'm doing wrong? Or is this something that is not
> supported? Or some bug?
>
> Thanks & Regards
> Jakub
>

Reply via email to