add linstor-scheduler-extender chart

[WanzenBug]

See https://github.com/piraeusdatastore/piraeus-operator/pull/277

[WanzenBug]

@kvaps could you please give this a try? I tried to make a "standard" chart, with common helm options.

[phoenix-bjoern]

Personally I don’t like storage-dependent declarations in pods. Users forget about them or they get outdated. Instead the scheduler should „automatically detect“ it’s actionable items based on the StorageClass of the PVC a workload is referencing (see https://kubernetes.io/docs/reference/access-authn-authz/admission-controllers/). Based on defaults set during the installation/deployment of the scheduler it can then decide whether it’s logic should be applied for relevant workloads or not. This reduces the effort smart capabilities get available/removed in a cluster.

[kvaps]

Yeah, I had a similar ideas. Maybe some mutation webhook could do that or set the linstor-scheduler as the default one.

[WanzenBug]

I generally agree. Implementation of such a scheme is a different beast however:

• A webhook only works if the PVC exists at Pod creation time, which doesn't need to be true
• We need to update the linstor-extender-plugin to better handle the LINSTOR API being down, otherwise it might block pod creation even for "non-storage" pods.
• If the scheduler is made default, it would need to started as a daemonset

I think a good compromise would be to optionally deploy the extender as stand-alone webhook, and have some instructions on how to integrate with the default scheduler. In any case, this is not something that needs to be solved in this PR. @phoenix-bjoern feel free to open a new issue so that this doesn't get lost.

[WanzenBug]

@kvaps added you to this repo, otherwise I don't think the approval counts.