Gluesys FlexA CSI Driver for Kubernetes

The official Container Storage Interface driver for Gluesys FlexA storage (FlexStor, ExaStor).

Container Images & Kubernetes Compatibility

Driver Name: csi.flexa.com

Driver Version	Image	Supported K8s Version
1.0.0	ghcr.io/gluesys/flexa-csi:1.0.0	1.21+

The Gluesys FlexA CSI driver supports:

• Access modes (CSI): SINGLE_NODE_WRITER, MULTI_NODE_MULTI_WRITER (typical Kubernetes mapping: RWO and RWX-style multi-writer NFS where applicable).
• Controller: Create/Delete volume, List volumes, Get capacity.
• Node: Stage/Unstage (no-op for the current NFS flow), Publish/Unpublish, Get volume stats.

Not supported at this time: snapshots, cloning. Volume expansion is supported (expand-only; shrink is not supported).

Prerequisites

• Kubernetes 1.21 or above
• Gluesys FlexA storage (FlexStor, ExaStor) 1.4.2 or above, with at least one storage pool for ZFS-backed volumes, or Lustre cluster configuration as required by your deployment
• Go 1.21+ recommended when building from source

Before you install

1. For ZFS workflows, create and initialize at least one storage pool on Gluesys FlexA storage.
2. Deploy the driver using Helm (recommended) or raw manifests (deploy/kubernetes) as described in Installation.

Installation

From the repository root (directory that contains charts/ and config/).

Helm (recommended)

Chart: charts/flexa-csi.

Minimal install (uses charts/flexa-csi/values.yaml for clientInfo.content and other defaults):

helm upgrade --install flexa-csi ./charts/flexa-csi \
  --namespace flexa-csi

Create the namespace first if it does not exist (kubectl create namespace flexa-csi), or add --create-namespace to the helm line so Helm creates it. The chart itself does not install a Namespace resource.

• Proxy / VIP config: Edit clientInfo.content in values.yaml, or use -f my-values.yaml, or --set-file clientInfo.content=/absolute/path/to/client-info.yml (paths are relative to your shell cwd). You do not need a separate config/client-info.yml unless you choose the file-based workflow.
• Secret managed by chart: With clientInfo.create: true (default), the chart creates the Secret from clientInfo.content. To use a Secret you created yourself: kubectl create secret ... then install with --set clientInfo.create=false (Secret name must match clientInfo.secretName, default client-info-secret).
• Optional: Tune image, sidecars, host paths, and storageClass entries under values.yaml (templates/storageclass.yaml renders enabled entries). Run helm lint ./charts/flexa-csi before applying.

Raw manifests

1. git clone https://github.com/gluesys/flexa-csi.git and cd flexa-csi

2. cp config/client-info-template.yml config/client-info.yml and edit config/client-info.yml (proxy endpoints and optional mountIP; see client-info / Secret and StorageClass parameters).

3. Create the client-info Secret in the driver namespace (see Creating the Secret manually).

4. Install:

   cd deploy/kubernetes
   sh install.sh

5. Verify: kubectl get pods -n flexa-csi

CSI Driver Configuration

You need a Secret (client-info.yml) and StorageClasses. Volume snapshots are not implemented; do not rely on VolumeSnapshotClass until support exists.

Contents: client-info / Secret · StorageClasses · PVC annotations · Pod annotations · VolumeContext

client-info / Secret

Helm: Configure proxy endpoints in charts/flexa-csi/values.yaml (clientInfo.content) or use -f / --set-file as described in Installation.

Raw manifests: Prepare config/client-info.yml with profiles (proxyIP, proxyPort, optional mountIP for VIP resolve). See the template at config/client-info-template.yml.

Creating the Secret manually

Use when not using Helm, or when you pre-create the Secret for Helm (clientInfo.create=false):

1. Example config/client-info.yml:

   profiles:
     zfs:
       proxyIP: 10.0.0.11
       proxyPort: 9001
       mountIP: "192.168.0.0/18"
     lustre:
       proxyIP: 10.0.0.12
       proxyPort: 9001
       mountIP: "192.168.0.0/18"

2. Create the Secret so the data key is client-info.yml (required by the driver):

   kubectl create secret generic client-info-secret -n flexa-csi \
     --from-file=client-info.yml=./config/client-info.yml

   Replace the namespace or file path as needed. If you rename the Secret, update the same name in deploy/kubernetes/controller.yml and node.yml (or Helm clientInfo.secretName).

The driver watches client-info-secret in the driver namespace and applies updates at runtime.

Creating Storage Classes

Parameter names are case-sensitive (poolName, not poolname).

Helm: Define entries under storageClass in charts/flexa-csi/values.yaml. Each key can enable a class; ZFS entries set poolName, Lustre entries set clusterName. The chart template is charts/flexa-csi/templates/storageclass.yaml.

Raw YAML examples: deploy/kubernetes/storage-class-zfs.yml, deploy/kubernetes/storage-class-lustre.yml

ZFS example:

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: flexa-sc-zfs
provisioner: csi.flexa.com
parameters:
  fs: "zfs"
  poolName: "kubernetes"
  protocol: "nfs"
  proxyProfile: "zfs"
reclaimPolicy: Delete
allowVolumeExpansion: true

Lustre example:

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: flexa-sc-lustre
provisioner: csi.flexa.com
parameters:
  fs: "lustre"
  clusterName: "cvol"
  protocol: "nfs"
  proxyProfile: "lustre"
reclaimPolicy: Delete
allowVolumeExpansion: true

Parameters

Name	Required	Description
fs	Yes	zfs or lustre.
poolName	For fs=zfs	ZFS pool name on Gluesys FlexA storage.
clusterName	For fs=lustre	Lustre cluster name.
protocol	Typically set	e.g. nfs.
proxyProfile	Optional	Profile name under profiles in client-info. When set, the driver uses that profile’s proxyIP, proxyPort, and mountIP.

Volume expansion is expand-only. Shrink (reducing requested size) is not supported.

Apply raw YAML:

kubectl apply -f <storageclass_yaml>

PVC annotations

The controller reads PersistentVolumeClaim annotations with the flexa.io/ prefix at volume creation. Examples: deploy/kubernetes/pvc_zfs.yaml, deploy/kubernetes/pvc_lustre.yaml.

Annotation	Purpose (summary)
flexa.io/optionISS	Instant secure sync (ZFS).
flexa.io/optionSVS	Secure volume service (ZFS).
flexa.io/optionComp	Compression (ZFS).
flexa.io/optionDedup	Deduplication (ZFS).
flexa.io/secureAddress	Secure network address for export.
flexa.io/secureSubnet	Secure network mask.
flexa.io/nfsAccess	NFS access mode (e.g. RW).
flexa.io/nfsNoRootSquashing	NFS root squashing.
flexa.io/nfsInsecure	NFS insecure port.

Use values such as "on" / "off" or addresses as in the samples; follow Gluesys FlexA documentation for your environment.

Pod annotations

Used at NodePublishVolume (NFS mount):

Annotation	Purpose
flexa.io/mountOptions	Comma-separated NFS options (e.g. vers=4,ro,async,timeo=666). See deploy/kubernetes/pod_zfs.yaml.
flexa.io/serviceVIP	If set, NFS server address for the mount; otherwise the controller-provisioned vip from VolumeContext is used. One of these must be available for publish to succeed.

VolumeContext (advanced)

On provision, the PV carries metadata such as vip, baseDir, poolName, fs, clusterName, protocol, pvcName, pvcNS, proxyProfile, proxyIP, proxyPort, mountIP. Do not edit these under normal operation; they support delete and node publish. For troubleshooting, check vip, baseDir, and pod flexa.io/serviceVIP.

Building and publishing images

• Build binary: make flexa-csi-driver
• Build image: make docker-build
• Publish to GHCR: docker login ghcr.io then make docker-publish (publishes ghcr.io/gluesys/flexa-csi:1.0.0 and :latest)

By default the cluster pulls ghcr.io/gluesys/flexa-csi from GHCR. For a locally built image, set imagePullPolicy: IfNotPresent and the image reference in Helm values.yaml or in deploy/kubernetes/controller.yml / node.yml.

Installing a build is covered in Installation; override image.repository / image.tag in Helm or edit the manifests before kubectl apply.

Uninstall

• Helm: helm uninstall flexa-csi -n flexa-csi (use helm list -n flexa-csi if the release name differs).
• Raw manifests: From deploy/kubernetes, run sh cleanup.sh (review the script if you customized StorageClass names).

Ensure no workloads still depend on Gluesys FlexA volumes. PVCs, PVs, and backend volumes may remain depending on reclaimPolicy and storage behavior; delete or retain them according to your operations policy.