Post Install Actions

Table of contents

  1. Overview
  2. Supported Actions
    1. restart
    2. delete
  3. Configuration
  4. Configuration Options
    1. action (Required)
    2. kind (Required)
    3. namespace
    4. name
    5. labelSelector
    6. group
    7. version
  5. Examples
    1. Restart by Name
    2. Restart by Label Selector
    3. Restart Multiple Resources
    4. Restart Different Resource Types
    5. Delete by Name
    6. Delete by Label Selector
    7. Combined Workflow
  6. How It Works
    1. Restart Action
    2. Delete Action
  7. Validation
    1. Required Fields
    2. Action-Kind Compatibility
  8. Execution Order
    1. Full Execution Flow
  9. Important Notes
    1. Label Selector Precedence
    2. Core API Group
    3. No Rollback
  10. Troubleshooting
    1. Resource Not Found
    2. Label Selector Not Matching
    3. Action Not Supported
    4. Permission Errors
  11. Best Practices
    1. Use with Helm Charts
    2. Clean Up After Manifests
    3. Use Label Selectors for Flexibility
    4. Validate Labels First

Overview

Post-install actions allow you to perform automated operations on Kubernetes resources after the cluster setup is complete. This is useful for:

  • Restarting deployments after configuration changes
  • Cleaning up temporary pods
  • Forcing rollouts to pick up new configurations
  • Removing debug resources

Actions run after Helm charts are installed, manifests are applied, and patches are applied.

Supported Actions

restart

Performs a rollout restart on the specified resource (equivalent to kubectl rollout restart).

Supported resource kinds:

  • Deployment
  • StatefulSet
  • DaemonSet

delete

Deletes the specified resource (equivalent to kubectl delete).

Supported resource kinds:

  • Pod

Configuration

Add actions under the postInstallActions key:

postInstallActions:
  - action: restart
    kind: Deployment
    name: argocd-server
    namespace: argocd
    
  - action: delete
    kind: Pod
    name: old-pod
    namespace: default

Configuration Options

action (Required)

Type: string
Values: restart | delete
Description: The action to perform on the resource.

action: restart

kind (Required)

Type: string
Description: The kind of Kubernetes resource.

  • For restart action: Deployment, StatefulSet, or DaemonSet
  • For delete action: Pod
kind: Deployment

namespace

Type: string
Optional: Yes
Default: default
Description: The namespace where the resource(s) exist.

namespace: argocd

name

Type: string
Optional: Either name or labelSelector must be provided
Description: The name of a specific resource to target.

name: argocd-server

If both name and labelSelector are provided, labelSelector takes precedence and name is ignored.

labelSelector

Type: object (map of label key-value pairs)
Optional: Either name or labelSelector must be provided
Description: Labels to select multiple resources. All resources matching the labels will be affected.

labelSelector:
  app.kubernetes.io/name: argocd-server
  app.kubernetes.io/component: server

group

Type: string
Optional: Yes
Description: The API group of the resource.

Defaults:

  • Pod: `` (empty string / core API)
  • Deployment, StatefulSet, DaemonSet: apps

You typically don’t need to specify this unless working with custom resources.

group: apps

version

Type: string
Optional: Yes
Default: v1
Description: The API version of the resource.

version: v1

Examples

Restart by Name

Restart a specific deployment by name:

postInstallActions:
  - action: restart
    kind: Deployment
    name: argocd-server
    namespace: argocd

This is equivalent to:

kubectl rollout restart deployment/argocd-server -n argocd

Restart by Label Selector

Restart all deployments matching specific labels:

postInstallActions:
  - action: restart
    kind: Deployment
    namespace: argocd
    labelSelector:
      app.kubernetes.io/name: argocd-applicationset-controller

This is equivalent to:

kubectl rollout restart deployment -n argocd -l app.kubernetes.io/name=argocd-applicationset-controller

Restart Multiple Resources

Use multiple labels to target specific resources:

postInstallActions:
  - action: restart
    kind: Deployment
    namespace: production
    labelSelector:
      app: myapp
      environment: production
      version: v2

Restart Different Resource Types 

postInstallActions:
  - action: restart
    kind: Deployment
    name: web-app
    namespace: default
    
  - action: restart
    kind: StatefulSet
    name: database
    namespace: default
    
  - action: restart
    kind: DaemonSet
    name: monitoring-agent
    namespace: monitoring

Delete by Name

Delete a specific pod:

postInstallActions:
  - action: delete
    kind: Pod
    name: temporary-job
    namespace: default

This is equivalent to:

kubectl delete pod/temporary-job -n default

Delete by Label Selector

Delete all pods matching specific labels:

postInstallActions:
  - action: delete
    kind: Pod
    namespace: kube-system
    labelSelector:
      app: cleanup
      temporary: "true"

This is equivalent to:

kubectl delete pod -n kube-system -l app=cleanup,temporary=true

Combined Workflow

Restart deployments and clean up temporary pods:

postInstallActions:
  # Restart Argo CD components to pick up new config
  - action: restart
    kind: Deployment
    namespace: argocd
    labelSelector:
      app.kubernetes.io/part-of: argocd
  
  # Remove debug pods
  - action: delete
    kind: Pod
    namespace: default
    labelSelector:
      debug: "true"
  
  # Remove specific temporary pod
  - action: delete
    kind: Pod
    name: init-job-xyz
    namespace: default

How It Works

Restart Action

When you restart a resource, BeKind:

  1. Gets the resource using the Kubernetes dynamic client
  2. Updates the pod template annotation: 
    kubectl.kubernetes.io/restartedAt: "2024-11-16T10:30:00Z"
  3. This triggers Kubernetes to perform a rolling restart

The restart is graceful and follows the resource’s update strategy.

Delete Action

When you delete a resource, BeKind:

  1. Finds the resource(s) matching the criteria
  2. Deletes each resource using the Kubernetes API
  3. The resource is removed immediately (subject to grace periods)

Validation

BeKind validates your configuration before executing:

Required Fields

  • action must be specified
  • kind must be specified
  • Either name or labelSelector must be provided

Action-Kind Compatibility

  • restart action only works with: Deployment, StatefulSet, DaemonSet
  • delete action only works with: Pod

Using invalid combinations will result in an error.

Execution Order

Actions are executed in the order they appear in your configuration:

postInstallActions:
  - action: restart
    kind: Deployment
    name: first         # Executed first
    
  - action: restart
    kind: Deployment
    name: second        # Executed second
    
  - action: delete
    kind: Pod
    name: cleanup       # Executed third

Full Execution Flow

  1. KIND cluster is created
  2. Docker images are loaded (if configured)
  3. Helm charts are installed (if configured)
  4. Post-install manifests are applied (if configured)
  5. Post-install patches are applied (if configured)
  6. Post-install actions are performed ← You are here

This order allows you to patch resources first, then trigger actions like restarts to pick up the patched configurations.

Important Notes

Label Selector Precedence

If both name and labelSelector are provided, labelSelector takes precedence and name is ignored.

# This will use labelSelector and ignore name
postInstallActions:
  - action: restart
    kind: Deployment
    name: ignored-name                    # This is ignored
    labelSelector:
      app: myapp                          # This is used

Core API Group

For Pods (core API resources), use an empty string for the group or omit it entirely:

# Both are correct
postInstallActions:
  - action: delete
    kind: Pod
    group: ""           # Explicit empty string
    
  - action: delete
    kind: Pod           # Group omitted (uses default)

No Rollback

Actions are executed once and not tracked. There is no automatic rollback if something fails.

Troubleshooting

Resource Not Found

If you get “resource not found” errors:

  1. Check resource exists: 
    kubectl get deployment argocd-server -n argocd
  2. Verify namespace: 
    kubectl get deployment --all-namespaces | grep argocd-server
  3. Check spelling: Resource names and namespaces are case-sensitive

Label Selector Not Matching

If label selectors don’t find resources:

  1. Check labels on resources: 
    kubectl get deployment -n argocd --show-labels
  2. Verify label syntax: 
    # Correct
labelSelector:
  app.kubernetes.io/name: argocd
   
# Wrong - values must be strings
labelSelector:
  replicas: 1  # This won't work
  3. Test selector manually: 
    kubectl get deployment -n argocd -l app.kubernetes.io/name=argocd

Action Not Supported

If you get “action not supported” errors:

  1. Check action name: Must be exactly restart or delete

  2. Verify kind is supported:

    • restart: Only Deployment, StatefulSet, DaemonSet
    • delete: Only Pod

Permission Errors

If you get permission errors:

  1. Check cluster access: 
    kubectl auth can-i update deployments -n argocd
kubectl auth can-i delete pods -n default
  2. Verify RBAC: Ensure your kubeconfig has appropriate permissions

Best Practices

Use with Helm Charts

Restart deployments after Helm installs to pick up new configurations:

helmCharts:
  - url: "https://argoproj.github.io/argo-helm"
    repo: "argo"
    chart: "argo-cd"
    release: "argocd"
    namespace: "argocd"
    wait: true

postInstallActions:
  - action: restart
    kind: Deployment
    namespace: argocd
    labelSelector:
      app.kubernetes.io/part-of: argocd

Clean Up After Manifests

Remove temporary resources created by manifests:

postInstallManifests:
  - "file:///path/to/init-job.yaml"

postInstallActions:
  - action: delete
    kind: Pod
    namespace: default
    labelSelector:
      job: init

Use Label Selectors for Flexibility

Label selectors are more flexible than names:

# Good - works even if deployment name changes
postInstallActions:
  - action: restart
    kind: Deployment
    labelSelector:
      app: myapp

# Less flexible - breaks if name changes
postInstallActions:
  - action: restart
    kind: Deployment
    name: myapp-deployment-v1

Validate Labels First

Before using label selectors in BeKind, test them:

# Check what will be affected
kubectl get deployment -n argocd -l app.kubernetes.io/part-of=argocd

# Verify it's what you expect
kubectl get deployment -n argocd -l app.kubernetes.io/part-of=argocd -o name