windows

CSI Driver on Windows

Note

Only use root share for one SMB server in one cluster and use subPath in deployment, if there is already \\smb-server\share\test1 mounted, would get error when mounting volume \\smb-server\share\test2 after Windows node reboot. Workaround is only use \\smb-server\share as source, details here.

Feature Status: GA

Prerequisite

• Install CSI-Proxy on Windows Node
• install csi-proxy on k8s 1.23+ Windows node using host process daemonset directly

kubectl apply -f https://raw.githubusercontent.com/kubernetes-csi/csi-driver-smb/master/deploy/example/windows/csi-proxy.yaml

Deploy a Windows pod with PVC mount

Create a Windows deployment

kubectl apply -f https://raw.githubusercontent.com/kubernetes-csi/csi-driver-smb/master/deploy/example/windows/statefulset.yaml

Enter pod container to verify

$ kubectl exec -it busybox-smb-0  -- bash
C:/ $ ls mnt/smb

In the above example, there is a c:\mnt\smb directory mounted as SMB filesystem.

Name resolution

It is important to note that if you have defined your Storage class based on the name of the service, then the Windows NODE must be able to resolve that name (ie. outside Kubernetes).

e.g.

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: smb
provisioner: smb.csi.k8s.io
allowVolumeExpansion: true
parameters:
  source: //smb-server.default.svc.cluster.local/share
...

and the error you get in logs (kubectl logs -n kube-system ds/csi-proxy) is the famous The network path was not found:

I0706 02:11:47.703698    4460 server.go:56] calling NewSmbGlobalMapping with remote path "\\\\smb-server.default.svc.cluster.local\\share\\pvc-1a6c607b-df88-4a61-8628-4d9d066dc158"
I0706 02:11:48.524376    4460 server.go:98] Remote \\smb-server.default.svc.cluster.local\share not mapped. Mapping now!
E0706 02:11:49.319012    4460 server.go:101] failed NewSmbGlobalMapping NewSmbGlobalMapping failed. output: "New-SmbGlobalMapping : The network path was not found. \r\nAt line:1 char:190\r\n+ ... , $PWord;New-SmbGlobalMapping -RemotePath $Env:smbremotepath -Cred ...\r\n+                 ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~\r\n    + CategoryInfo          : NotSpecified: (MSFT_SmbGlobalMapping:ROOT/Microsoft/...mbGlobalMapping) [New-SmbGlobalMa \r\n   pping], CimException\r\n    + FullyQualifiedErrorId : Windows System Error 53,New-SmbGlobalMapping\r\n \r\n", err: exit status 1

For most cloud providers that offer Windows nodes, this is not an issue. Those nodes will be part of a domain where name resolution includes the Kubernetes dns resolution.

For self-hosted Kubernetes clusters there is no default name resolution to solve this issue.

Name resolution solution

A simple hack workaround is to use a fixed ClusterIP for the smb service, and use the IP in the storageClass.

It is VERY VERY much not recommended to use fixed IPs inside Kubernetes, but in this very specific case it make for a simple workaround.

The IP needs to be in the service CIDR range, e.g. if it is 10.96.0.0/12, you could pick 10.111.254.254.

apiVersion: v1
kind: Service
metadata:
...
spec:
  type: ClusterIP
  clusterIP: 10.111.254.254
...

and

apiVersion: storage.k8s.io/v1
kind: StorageClass
metadata:
  name: smb
provisioner: smb.csi.k8s.io
allowVolumeExpansion: true
parameters:
  source: //10.111.254.254/share
...

If kubectl logs -n kube-system ds/csi-proxy continues to use names after this fix, it might be necessary to reboot the node.