recipes_collection/cluster/k8s.yaml

# Original Copyright (c), NVIDIA CORPORATION. Modifications © Amazon.com

pullPolicy: Always # policy to pull container, can be Always, IfNotPresent and Never
restartPolicy: Never # restart policy
namespace: default # the namespace to submit job
# create customized labels for the PytorchJob and Pods deployed jobs.
# Example:
#   custom_labels:
#     label-key-1: label-value-1
#     label-key-2: label-value-2
custom_labels: null
# create customized annotations for the jobs.
# Example:
#   annotations:
#     annotation-key-1: annotation-value-1
#     annotation-key-2: annotation-value-2
annotations: null
# add service account to job pods
# Example:
#  serviceAccountName: service_account
service_account_name: null
# priorityClassName for Kueue scheduler to decide jobs priority
priority_class_name: null

# temp volume, usually used to mount temp directory
# Example:
#  volumes:
#    - volumeName: data1
#      hostPath: "/data"
#      mountPath: "/data"

volumes: null

# persistent volume, usually used to mount FSx
# Example:
# persistent_volume_claims:
#       - claimName: null
#         mountPath: null
#       - claimName: null
#         mountPath: null

# persistent volumes, usually used to mount FSx
persistent_volume_claims:
  - null
  # This claim should be created before running. Example:
  # - claimName: fsx-claim
  #   mountPath: data

# Create k8s NodeAffinity to select nodes to deploy jobs which matches required and preferred labels
# Structure:
#   label_selector:
#     required: <required label key-values pair>
#     preferred: <preferred label key-values pair>
#     weights: <weights list used by preferred labels to get nodes priority>
# Example:
#   label_selector:
#     required:
#       example-label-key:
#         - expected-label-value-1
#         - expected-label-value-2
#     preferred:
#       preferred-label-key:
#         - preferred-label-value-1
#         - preferred-label-value-2
#     weights:
#       - 100
label_selector: null

# The clean up policy after the job completes or fails.
cleanPodPolicy: null