It was working well before adding the following change in the spark driver pod yaml specification to make driver pod use a PVC backed by externally provisioned storage for spilling files to a custom path:
spec:
containers:
env:
- name: SPARK_LOCAL_DIRS
value: /ephemeral
volumeMounts:
- name: spark-local-dir-1
mountPath: /ephemeral
volumes:
- name: spark-local-dir-1
persistentVolumeClaim:
claimName: shared-persistence
Then the pod gets stuck at ContainerCreating state and there aren't any events or errors in kubectl describe pod
. PVC shared-persistence
is bound.
One potential reason I can think of is that we also have ephemeral-storage definitions in the resources section:
containers:
- resources:
limits:
cpu: '1'
ephemeral-storage: 4Gi
memory: 2457Mi
requests:
cpu: '1'
ephemeral-storage: 500Mi
memory: 2457Mi
Are they conflicting? It is difficult to remove the ephemeral-storage definitions due to some issues in our codebase so I would like to confirm first. Thanks.