Is it possible to use two external linstor clusters at the same time?

[jonathon2nd]

Hello!

We are planning some new servers, and we are evaluating having two clusters for HA.
Is it possible to have two apiVersion: piraeus.io/v1 kind: LinstorCluster k8s objects for our external linstor clusters, and have some way to have k8s workers 1 2 and 3 use cluster1, and workers 4 5 and 6 use cluster2?

I noticed in apiVersion: piraeus.io/v1 kind: LinstorSatellite the following in the yaml

spec:
  clusterRef:
    externalController:
      url: >-
        http://10.2.4.10:3370/,http://10.2.4.11:3370/,http://10.2.4.12:3370/,http://10.2.4.13:3370/
    name: linstorcluster

So I think it is possible, I have checked linstor doc and not seeing anything.

Our idea would be to have one k8s cluster span these two host clusters (as opposed to using something like liqo)

[WanzenBug]

It's an interesting idea. It's not something we have anticipated, and if you did not use a an external cluster, the answer would be clearly no.

Right now, I do see a few issues:

• Some of the resources deployed by the operator would have a name collision, such as the CSI Controller Deployment and CSI Node daemonset: those are not "inherited" from the LinstorCluster resource, so they would always end up named linstor-csi-controller.
• The LinstorSatelliteConfiguration would always apply to both cluster. Solvable by proper node selectors.
• You have to use some strict topology settings in your StorageClass, otherwise Kubernetes might schedule a Pod to a node that is not connected to the "right" LINSTOR cluster.

Generally, I don't see any advantage regarding HA, as any volume on one partition of the cluster could only ever exist on those nodes: the other partition simply does not know about the LINSTOR cluster and its volumes.

[jonathon2nd]

The idea is to treat each cluster as a zone, to tolerate external linstor cluster level failure.

We have already tolerated an external linstor satellite failure, but a persistent controller failure is what this would be mitigating ideally.

[WanzenBug]

Is you external controller already set up for HA?