Skip to main content

Sync Policies

Manual vs automated​

By default, ArgoCD detects drift but does not apply changes automatically. Enable automated sync only where you trust the Git state completely.

syncPolicy:
  automated:
    prune: true       # delete resources removed from Git
    selfHeal: true    # re-apply if cluster state drifts
prune: true

With pruning enabled, removing a file from Git will delete the corresponding Kubernetes resource. Double-check before enabling in production.

Sync options​

syncPolicy:
  syncOptions:
    - CreateNamespace=true          # create destination namespace if missing
    - PrunePropagationPolicy=foreground  # wait for cascading delete
    - PruneLast=true                # delete after creating new resources
    - Replace=true                  # use kubectl replace instead of apply
    - ServerSideApply=true          # use SSA (handles large CRDs)
    - RespectIgnoreDifferences=true # apply ignoreFields during sync

Ignore differences​

Ignore fields managed by external controllers to prevent constant drift detection:

spec:
  ignoreDifferences:
    - group: apps
      kind: Deployment
      jsonPointers:
        - /spec/replicas          # managed by HPA
    - group: ""
      kind: Service
      jsonPointers:
        - /spec/clusterIP         # assigned by Kubernetes
    - group: autoscaling
      kind: HorizontalPodAutoscaler
      jqPathExpressions:
        - .spec.metrics[] | select(.type == "ContainerResource")

Manual sync triggers​

# Sync with pruning
argocd app sync my-app --prune

# Sync a specific resource only
argocd app sync my-app --resource apps:Deployment:api-server

# Dry run
argocd app sync my-app --dry-run

# Force replace (use with caution)
argocd app sync my-app --replace

Refresh​

Refresh fetches the latest Git state without applying:

argocd app get my-app --refresh

Force hard refresh (busts the manifest cache):

argocd app get my-app --hard-refresh