How to restrict runner autoscale based of runner labels?

[rahul9170]

Hello,

Currently the runner autoscales(pull based) based on metric type: TotalNumberOfQueuedAndInProgressWorkflowRuns which scales up the runner as soon as it sees a job queued. Can we restrict this so it autoscales based on the runner label mentioned in the job to prevent unnecessary autoscaling?

Thanks,
Rahul Sharma

[toast-gear]

Have you tried v0.23.0?

[toast-gear]

You shouldn't need to do anything, label support was added in v0.23.0 and having just re-tested it it's working fine in my environment. What does all of your yaml look like? Use code snippets:

[rahul-FLS]

Here are the yamls

apiVersion: v1
items:

• apiVersion: actions.summerwind.dev/v1alpha1
  kind: RunnerDeployment
  metadata:
  generation: 3
  labels:
  app: release-27a27040-595d-3961-9348-3e105b625be2-runner
  chart: actions-runner-0.1.0-0.4
  release: release-27a27040-595d-3961-9348-3e105b625be2
  name: actions-runner
  namespace: actions-runner-system
  resourceVersion: "504580"
  uid: 12ae65f3-59d5-490e-a5be-f952e2ab42a5
  spec:
  effectiveTime: null
  replicas: 0
  selector: null
  template:
  metadata: {}
  spec:
  dockerEnabled: false
  dockerdContainerResources: {}
  env:
  - name: RUNNER_FEATURE_FLAG_EPHEMERAL
  value: "true"
  - name: DISABLE_RUNNER_UPDATE
  value: "true"
  ephemeral: true
  group: poc
  image: actionsRunner
  imagePullSecrets:
  - name: secret
  labels:
  - container-runner
  organization: orgName
  resources:
  limits:
  cpu: "1"
  memory: 2Gi
  requests:
  cpu: 500m
  memory: 500Mi
  securityContext:
  runAsUser: 1000
  serviceAccountName: actions-runner-sa
  volumeMounts:
  - mountPath: /tmp
  name: tmp
  volumes:
  - emptyDir: {}
  name: tmp

apiVersion: v1
items:

• apiVersion: actions.summerwind.dev/v1alpha1
  kind: HorizontalRunnerAutoscaler
  metadata:
  creationTimestamp: "2022-05-03T11:29:58Z"
  generation: 1
  name: actions-runner-autoscaler
  namespace: actions-runner-system
  resourceVersion: "504533"
  uid: d1c9d7f6-9be2-44c2-83c2-fa2a491f55e3
  spec:
  maxReplicas: 5
  metrics:
  • repositoryNames:
    • repoName
      type: TotalNumberOfQueuedAndInProgressWorkflowRuns
      minReplicas: 0
      scaleTargetRef:
      name: actions-runner

[toast-gear]

Format this so it's in a code block, this is incredibly hard to read. What behaviour are you actually seeing? I expected 2 runner*.yaml files as I assume you aren't getting only the matching labeled runner scaling?

[rahul9170]

Sorry about that. Here are the formatted ones.

HorizaontalRunnerAutoscaler.yaml

RunnerDeployment.yaml

I could see runners are getting scaled up if the controller sees any job(irrespective to the job label) queued up in GitHub.

[toast-gear]

mmmmm, container-runner is unique? Can provide what an offending workflow looks like? What do the controller logs say, I'm espcially interested in the Suggested desired replicas of $CALCULATED_REPLICA_COUNT by TotalNumberOfQueuedAndInProgressWorkflowRuns lines and their surrounding logs. If that label is unique to this bank of runners and you are referencing it in a workflow then I've tested this scenario this morning and it worked OK in my environment so it's slightly perplexing! 🤔

[mumoshu]

Could you also share the kubectl get deployment -o yaml output for ARC's deployment?

[rahul9170]

Hey this seems to be working now not sure what changed. I'll test it more and shout in case of any issue. Thanks again :)