Spinnaker Running a Script with Run Job (Manifest)

External

• https://kb.armory.io/s/article/Run-a-Generic-Shell-Script-with-Spinnaker
• https://spinnaker.io/docs/guides/user/kubernetes-v2/run-job-manifest/
• https://blog.spinnaker.io/extending-spinnaker-with-kubernetes-and-containers-5d16ec810d81

Internal

• Spinnaker Concepts
• Run Job (Manifest)

Overview

The execution phase is implemented as a Run Job (Manifest) stage.

⚠️ Make sure you use Run Job (Manifest) and NOT Deploy (Manifest)!

Run Job (Manifest) only accepts Job manifests, so a ConfigMap to specify the script cannot be used.

The stage contains the definition of a Kubernetes ConfigMap, which includes the script as data, and a Kubernetes Job, which executes the script within a container whose image is specified in the job definition. A simpler approach would be to only specify the container command, without a script in the ConfigMap and without the ConfigMap. This works if we control the image and we can release new versions of the image to include the functionality we need. The stage will deploy the job manifest and wait until it completes, thus gating the pipeline’s continuation on the job’s success or failure. The stage may optionally collect the Job output and inject it back into the pipeline. If the pipeline is intended to be run repeatedly, which is the case for most of the pipelines, it must also contain a Delete (Manifest) stage to delete the Job and associated resources after execution, otherwise an attempt to re-run it fails on account of not being able to modify the Job. An alternative is to specify a time to live for the Job.

Solution

Run Job (Manifest) Definition

Add a "Run Job (Manifest)" stage and name it "Script Job" or similar.

The Account is the target Kubernetes cluster.

In Manifest Configuration. set Manifest Source: Text, and use this:

apiVersion: v1
data:
  script.sh: |-
    echo "Hello world!"
kind: ConfigMap
metadata:
  labels:
    app: script-job
  name: script-job-cm
  namespace: REPLACE-WITH-TARGET-NAMESPACE-NAME
---
apiVersion: batch/v1
kind: Job
metadata:
  labels:
    app: script-job
  name: script-job
  namespace: REPLACE-WITH-TARGET-NAMESPACE-NAME
spec:
  backoffLimit: 2
  template:
    spec:
      containers:
        - command:
            - sh
            - /opt/script/script.sh
          image: 'bitnami/kubectl:1.12'
          name: script
          volumeMounts:
            - mountPath: /opt/script
              name: volume01
              readOnly: true
      restartPolicy: Never
      volumes:
        - configMap:
            name: script-job-cm
          name: volume01

Notes:

• Replace 'REPLACE-WITH-TARGET-NAMESPACE-NAME' with the target namespace for both the ConfigMap and Job.
• Both manifests should use an app:script-job label, so they can be deleted at the same time with one selector.

Delete (Manifest) Definition

Add a Delete (Manifest) stage, name it "Script Job Clean Up".

Use the same account and namespace.

Selector: Match target(s) by label. In the Kinds control, select both job and configMap. For Labels, use key: "app", value "script-job"

This is the equivalent of manually executing:

kubectl -n of-test delete job,cm  --selector app=script-job

Execution Considerations

This "Script Job" stage deploys a Kubernetes Job, which gets executed immediately. The logic to be executed is provided as the content of a ConfigMap, declared in the same deployment artifact manifest.

If the pipeline fails in the Deploy (Manifest) stage with an error message that includes "field is immutable", it is because an older Job has been left around.

A failed job execution leaves Jobs around.

Upon deployment, the Job will run the script.

Viewing Execution Logs

https://spinnaker.io/docs/guides/user/kubernetes-v2/run-job-manifest/#viewing-execution-logs

The job execution log is captured and can be accessed in the pipeline execution details → Deploy (Manifest) stage → Deploy Status → Console Output (raw).

If the Spinnaker instance is configured as such, logs can be forwarded to dedicated infrastructure.

Also see:

Operations | Log Management

Generating and Using Execution Results

I haven't found a way to inject the stout of the script back into the pipeline, as an artifact. The only way the output of the script can be useful is to write it in S3 or GitHub and read it in a later stage as an artifact.