Local CSI Driver installation is missing from Helm installation guide

[rzetelskik]

What happened?

Following the Helm installation guide (https://operator.docs.scylladb.com/stable/installation/helm.html) results in a broken ScyllaDB Manager cluster because of unbound PVCs. This is due to ScyllaDB Manager's ScyllaCluster requiring scylladb-local-xfs StorageClass provisioned by Local CSI Driver. The same thing applies to ScyllaCluster created by scylla chart.

/kind documentation

What did you expect to happen?

Helm installation guide should involve installing Local CSI Driver and transitively, a NodeConfig, by either specifying it as a prerequisite or mentioning it as a step in the procedure.

How can we reproduce it (as minimally and precisely as possible)?

Follow Helm installation guide https://operator.docs.scylladb.com/stable/installation/helm.html

Scylla Operator version

master

Kubernetes platform name and version

n/a

Please attach the must-gather archive.

n/a

Anything else we need to know?

No response

[mflendrich]

My hunch is that the messaging there should be along the lines of

ScyllaDB Operator requires a CSI StorageClass in your Kubernetes cluster to configure data volumes.

We strongly recommend that ScyllaDB data volumes be locally attached to Kubernetes nodes, and use the XFS filesystem. While Scylla should work with any storage provisioner, it's strongly discouraged for production use because ScyllaDB heavily relies on asynchronous IO (AIO) that comes with XFS.

Option 1 - install Local CSI driver (and set it up for an XFS file system)
... instructions ...

Option 2 - use the CSI StorageClass that comes with your cloud provider (not recommended for production use)
... instructions ...

Do we want a stricter message here?
@tzach @ylebi

[rzetelskik]

@mflendrich we already have the storage options explained to an extent: https://operator.docs.scylladb.com/stable/architecture/storage/overview.html. IMO the Helm guide shouldn't go into details and just provide steps for the supported setup(s) (and maybe reference the other section going into details), mirroring the GitOps guide (https://operator.docs.scylladb.com/stable/installation/gitops.html#setting-up-local-storage-on-nodes-and-enabling-tuning and https://operator.docs.scylladb.com/stable/installation/gitops.html#local-csi-driver)

[ylebi]

@mflendrich we already have the storage options explained to an extent: https://operator.docs.scylladb.com/stable/architecture/storage/overview.html. IMO the Helm guide shouldn't go into details and just provide steps for the supported setup(s) (and maybe reference the other section going into details), mirroring the GitOps guide (https://operator.docs.scylladb.com/stable/installation/gitops.html#setting-up-local-storage-on-nodes-and-enabling-tuning and https://operator.docs.scylladb.com/stable/installation/gitops.html#local-csi-driver)

+1

My hunch is that the messaging there should be along the lines of

ScyllaDB Operator requires a CSI StorageClass in your Kubernetes cluster to configure data volumes.
We strongly recommend that ScyllaDB data volumes be locally attached to Kubernetes nodes, and use the XFS filesystem. While Scylla should work with any storage provisioner, it's strongly discouraged for production use because ScyllaDB heavily relies on asynchronous IO (AIO) that comes with XFS.
Option 1 - install Local CSI driver (and set it up for an XFS file system)
... instructions ...
Option 2 - use the CSI StorageClass that comes with your cloud provider (not recommended for production use)
... instructions ...

Do we want a stricter message here? @tzach @ylebi

@mflendrich in the instructions, we should just reference to the documentation.