MM2 Switchover and Failover process

[lsoica]

I'm configuring Kafka replication using MirrorMaker 2 (MM2) across two Kubernetes clusters, A and B. Each cluster runs the Strimzi operator and a Kafka server.

I begin by deploying MM2 on cluster B to replicate data from A to B.

For a switchover:

1. Stop consumers on A.
2. Stop MM2 on B.
3. Deploy MM2 on A to replicate from B to A, using auto.offset.reset: latest.
4. Move consumers to B (my understanding here is that consumers will take it over from where they left of in A).

Later, I want to switch back:

1. Stop consumers on B.
2. Stop MM2 on A.
3. Deploy MM2 on B to replicate from A to B again.

At this point, MM2's internal topics on B may be outdated. Should I first remove the MM2 internal topics on B, or is there something I'm missing in my understanding?

[katheris]

Hi @lsoica, yes you are correct that if you've stopped MM2 on cluster B and later want to start it again you should delete the MM2 internal topics on B so as to start fresh. However the MirrorSourceConnector also stores some data in an offset-syncs topic, which by default is in the source cluster, so it would be useful to also clean that up.

[romankydybets]

As i understand correctly you could go with this to manage topics.
https://strimzi.io/blog/2024/11/04/managing-connector-offsets/
Also, if we consumer.auto.offset.reset: latest will it starts from the latest message in topic, or internal topic of MM2 will keep the offset?

[katheris]

As i understand correctly you could go with this to manage topics.

Yes, that is the simplest approach

If you have reset the offsets to be empty then it will start from the latest, otherwise if there is an offset in the internal MM2 topic then it will use it.