automountServiceAccountToken

[jinxingwang]

Anyone want me to support this automountServiceAccountToken feature into spark operator?

[hopper-signifyd]

@jinxingwang That would be great to have!

[sparsh-95]

Can any maintainer share the update on when this will be prioritized? TIA :)

[Aransh]

Any updates on this issue? we need this for compliance with Azure Government cloud. @jinxingwang @sparsh-95

CC @EliranTurgeman

[jacobsalway]

Could you go into a bit more detail on the specific compliance requirements? Is credential auto-mounting disabled in these clusters?

[Aransh]

@jacobsalway Sure Azure Microsoft defender reported a High Severity security finding as below

"Kubernetes clusters should disable automounting API credentials" - Disable automounting API credentials to prevent a potentially compromised Pod resource to run API commands against Kubernetes clusters. For more information, see https://aka.ms/kubepolicydoc.

So in order to mitigate the above finding need to disable automounting of service account token by setting automountServiceAccountToken = false on all pods. This is not possible with the chart currently, or with the operator in general (for generated spark app pods).

I will mention the idea of the policy is "while obviously some apps require serviceAccount tokens, it shouldn't be mounted by default to avoid misuse, and instead should only be manually (and explicitly) mounted as a volume when needed", example of manually mounting to a pod:

    volumes:
    - name: kube-api-access
      projected:
        defaultMode: 420
        sources:
        - serviceAccountToken:
            expirationSeconds: 3607
            path: token
        - configMap:
            items:
            - key: ca.crt
              path: ca.crt
            name: kube-root-ca.crt
        - downwardAPI:
            items:
            - fieldRef:
                apiVersion: v1
                fieldPath: metadata.namespace
              path: namespace
    volumeMounts:
    - mountPath: /var/run/secrets/kubernetes.io/serviceaccount
      name: kube-api-access
      readOnly: true

Anyway, this is a required policy for work with Azure Government cloud

[jacobsalway]

@Aransh Thanks appreciate the details and the links. It would be easy enough to add this as a configurable field to the controller and webhook deployment specs in the Helm chart, however for the actual Spark driver pod to have this field it would require a change to Spark core or to the webhook in the operator or in a pod template spec.

Are both required for compliance in this environment? I would imagine so given the driver also needs a service account in order to request and watch executor pods.

[Aransh]

@jacobsalway Yup, both are required for compliance