Persitent Volume "devworkspace" has node selector on multi-node cluster

[granicayv]

Description

When the first workspace was created via Dashboard, Persistent Volume with claim "claim-devworkspace" was created by "devworkspace-operator". Persistent Volume with claim "claim-devworkspace" has node selection:

apiVersion: v1
kind: PersistentVolume
...
spec:
  accessModes:
  - ReadWriteOnce
...
  claimRef:
    apiVersion: v1
    kind: PersistentVolumeClaim
    name: claim-devworkspace
    namespace: admin-che
...
  nodeAffinity:
    required:
      nodeSelectorTerms:
      - matchExpressions:
        - key: kubernetes.io/hostname
          operator: In
          values:
          - node-3

All created workspaces use this Volume, so all of them were created on only one node despite the cluster having several nodes.

How To Reproduce

Steps to reproduce the behavior:

1. Create a multi-node microk8s cluster, for example, with 3 control plane nodes.
2. Install Che on this multi-node cluster.
3. Create several workspace pods
4. Make sure all of them were created on only one node

Also, you can create some simple deployment yaml file, add this volume to spec and apply this file. The created pod was located in the same node as all workspaces despite other nodes being less loaded.

Expected behavior

Expected that no one Kubernetes object has a node selector. It is necessary to make microk8s load nodes by itself.

Questions

1. Why this Persistent Volume is needed for workspaces?
2. Can we change it somehow to remove the node selector? Is it safety for Che's behavior?

[amisevsk]

The DevWorkspace Operator does not create Persistent Volumes -- PVs are being created here by the cluster in response to the claim-devworkspace persistent volume claim created by the DevWorkspace Operator. How PVs are created is up to the cluster configuration in most cases, e.g. see here for the docs on configuring NFS volumes for microk8s. Likely the node selector is provisioned based on the type of storage backing the PVCs requested by the DevWorkspace Operator.

If you would prefer to provision one PVC per workspace rather than one PVC per namespace (shared by all workspaces in that namespace), you can configure the storage-type of your workspaces by applying the attribute controller.devfile.io/storage-type: per-workspace to your Devfile/DevWorkspace (see docs). If you are using Eclipse Che, this can also be configured there.

In general, storage shared by multiple pods needs to run on the same node. Some storage types support transferring the persistent volume between nodes for load balancing purposes, but it appears the default storage available on your cluster does not.

Persistent storage is required by workspaces in order to store changes made inside workspaces between restarts -- otherwise every time your workspace shuts down, all changes you made to any cloned repositories would be lost. You can enable this behavior by setting the attribute controller.devfile.io/storage-type: ephemeral.

[granicayv]

Thank you for clarifying, closing the issue