Hi David,

Thanks for the feedback. See my comments below.

On Mon, Sep 29, 2025 at 2:25 PM David Arthur <[email protected]> wrote:
> If we automatically rewrite the bootstrap checkpoint into the zero
> checkpoint, does this limit how far back we can downgrade a controller's
> binary?

If the cluster has been activated, the bootstrap records in
bootstrap.checkpoint or 0-0.checkpoint have been replicated to the
cluster metadata partition and a checkpoint has been generated, the
user can downgrade to any Kafka release they want that supports the
records in the cluster metadata partition.

One small clarification. The KIP doesn't propose rewriting the records
from the bootstrap.checkpoint to the 0-0.checkpoint. The KIP will
commit bootstrap records just like it does today by writing the
bootstrap records to the cluster metadata partition log. The main
capability added by this KIP is that, during activation, the bootstrap
metadata records can come from either the bootstrap.checkpoint or the
0-0.checkpoint.

> Prior to KIP-853, I assume we don't have a zero checkpoint?

Correct. KIP-853 introduced the 0-0.checkpoint.

Thanks,
-- 
-José

Reply via email to