[Question] How many controllers do I need to run? And what happens during a migration?

[JohnKiller]

Hi, I didn't find this answer on the documentation, sorry if it's already been answered somewhere.

I'm using Nomad as the orchestrator, and I would like to know how many controllers do I need to run to maintain HA.

If only 1 is needed, what happens when the orchestrator moves the controller to another host? Any downtime?

Thanks

[Madhu-1]

Am not exactly sure how Nomad works with storage, but CephCSI alone cannot be run in HA in the CO. it needs to be run with some sidecar, for example, external-provisioner, external-snapshotter/external-resizer which are the sidecars in the Kubernetes which works with the leader election, in Kubernetes, you can have active/passive leaders. if one leader pod goes down another leader will take over.

[JohnKiller]

In case of a node/host failure, Nomad will reschedule the controller on another node, but it will be started from scratch. Does the controller need to persist its state somewhere? Do I need to mount a volume to it and migrate that too?

[Madhu-1]

In case of a node/host failure, Nomad will reschedule the controller on another node, but it will be started from scratch.

This is expected and fine as cephcsi is stateless.

Does the controller need to persist its state somewhere?

I dont think it's required. It will serve the request as before and recover if something is broken during the failure.

Do I need to mount a volume to it and migrate that too?

What do you mean by this

[JohnKiller]

Thanks for your answer. Since you said that it's stateless, the question I had about the volume it's not relevant anymore. I'll do some testing in my lab.

Thanks again

[Madhu-1]

Feel free to open issues if you find any.