k8s-resource-log-selector.yaml

# This example demonstrates how to include and obtain logs from pods created by a
# custom resource submitted via resource template. Note that this feature is only
# available in v3.3 and above.
#
# This is particularly useful since Argo Workflows does not know how
# other CRDs (Kubeflow training CRDs, Spark application CRD, etc.) work
# and thus could not pull the logs from the pods created by those CRDs.
apiVersion: argoproj.io/v1alpha1
kind: Workflow
metadata:
  generateName: k8s-jobs-log-selector-
spec:
  entrypoint: tf-jobtmpl
  templates:
  - name: tf-jobtmpl
    resource:
      action: create
      successCondition: status.succeeded = 2
      failureCondition: status.failed > 0
      # You can also create any other custom K8s resource here. This is an example
      # of using Kubeflow TFJob as an illustration.
      manifest: |
        apiVersion: kubeflow.org/v1
        kind: TFJob
        metadata:
          name: tfjob-examples
        spec:
          tfReplicaSpecs:
             Worker:
               replicas: 2
               restartPolicy: Never
               template:
                 metadata:
                   # We add this label to the pods created by TFJob custom resource to inform Argo Workflows
                   # that we want to include the logs from the created pods. Once the pods are created with this
                   # label, you can then use `argo logs -c tensorflow` to the logs from this particular container.
                   # Note that `workflow.name` is a supported global variable provided by Argo Workflows.
                   #
                   # The Kubeflow training controller will take this CRD and automatically created worker pods with
                   # labels, such as `job-role` and `replica-index`. If you'd like to query logs for pods with
                   # specific labels, you can specify the label selector explicitly via `argo logs -l <logs-label-selector>`.
                   # For example, you can use `argo logs -c tensorflow -l replica-index=0` to see the first worker pod's logs.
                   labels:
                     workflows.argoproj.io/workflow: {{workflow.name}}
                 spec:
                   containers:
                     - name: tensorflow
                       image: "Placeholder for TensorFlow distributed training image"