The kafka main nodes are split between buster and stretch:
- kafka-main00[1-3] are on stretch
- kafka-main00 are on buster
The  nodes were added as part of T225005 and they got Buster straight away, the other ones are still todo. In theory we could do the following:
- Create partman re-use recipes for kafka main, to keep the /srv partition while reimaging
- Reimage one node at the time to Buster
From the past experience of Kafka Jumbo it seems fine to keep the list of brokers that clients use in their config intact, since they should be able to react to brokers not being reachable without issues. I'd avoid Bullseye for the moment since we have the puppet code already working for Buster, and it should be quick and easy to upgrade the remaining nodes.
I can take care of it if you like the idea :)