ArgoCD
Topics in this category
App of Apps Applications GitOps Principles Multi-cluster RBAC Sync & Health Sync Waves & Hooks
ArgoCD

Multi-cluster

29 question(s)

Do workloads keep running if the ArgoCD hub goes down?

Beginner
Yes. ArgoCD manages desired state, but running workloads on spoke clusters continue operating independently of the hub. A hub outage stops new syncs and drift correction, not the already-deployed applications, so it's a control-plane, not a data-plane, dependency.
Real-world example During hub maintenance, production apps keep serving traffic; only new deployments pause until the hub returns.

Common follow-ups: What stops during a hub outage? | Why is this a control-plane dependency?

Sync & Health Applications Multi-cluster

How do you onboard a new cluster into an existing GitOps setup?

Intermediate
Register the cluster (Secret with proper labels), and if using ApplicationSet cluster generators, matching apps deploy automatically. With App of Apps, add a root for the cluster. Labeling correctly ensures the right app set lands without hand-writing each Application.
argocd cluster add new-ctx
# label it so generators pick it up
Real-world example Adding and labeling a new cluster causes the standard platform apps to deploy to it with no extra manifests.

Common follow-ups: What makes onboarding automatic? | Why label at registration?

Applications App of Apps Multi-cluster

How does controller sharding distribute cluster load?

Advanced
The application controller can run multiple replicas, each assigned a shard of clusters (by a sharding algorithm). This spreads reconciliation work so no single controller is overloaded managing hundreds of clusters, improving throughput and resilience at scale.
env:
  - name: ARGOCD_CONTROLLER_REPLICAS
    value: "3"
Real-world example Three controller shards each handle a third of the 300-cluster fleet, keeping reconciliation timely.

Common follow-ups: How are clusters assigned to shards? | Why shard at large scale?

Applications Sync & Health Multi-cluster

How do you deploy different app versions to different clusters (e.g., staged rollout)?

Intermediate
Pin each cluster's app to a specific revision via per-cluster config (values/overlays selected by cluster generator), so you can advance one cluster to a new version while others stay put. Promotion then moves the version cluster-by-cluster via PRs.
Real-world example A canary cluster runs v2 while the rest stay on v1, controlled by per-cluster tag values.

Common follow-ups: How do you pin per-cluster versions? | How is a staged rollout advanced?

Applications App of Apps Multi-cluster

What are security considerations for a hub managing many clusters?

Advanced
The hub holds credentials to every spoke, making it a high-value target. Mitigate with least-privilege service accounts per spoke, network segmentation, short-lived credentials, strict RBAC and SSO on ArgoCD, audit logging, and possibly splitting sensitive clusters to a separate instance to reduce blast radius.
Real-world example The hub uses minimally-scoped, rotating credentials per spoke and strict SSO/RBAC to limit the impact of a compromise.

Common follow-ups: Why is the hub high-value? | How do you limit blast radius?

RBAC App of Apps Multi-cluster

How do you remove a cluster from ArgoCD?

Beginner
Use `argocd cluster rm <server-or-name>` or delete its cluster Secret. First ensure no Applications target it (or they'll go Unknown/fail). Removing the Secret revokes ArgoCD's access and drops it from the destinations list.
argocd cluster rm prod-eu
Real-world example A decommissioned cluster is removed after migrating its apps, cleaning up its credentials.

Common follow-ups: What should you do before removing a cluster? | What does removal revoke?

Applications RBAC Multi-cluster

How do you view aggregated application status across clusters?

Intermediate
The ArgoCD UI and `argocd app list` show all Applications with their target cluster, sync, and health, giving a fleet-wide view. Filtering by cluster/project/label helps spot which cluster has failing apps across the whole estate from one console.
argocd app list -o wide
Real-world example A single dashboard reveals that only the ap-southeast cluster has OutOfSync apps after a rollout.

Common follow-ups: How do you filter by cluster? | Where's the fleet-wide view?

Sync & Health Applications Multi-cluster

How does ApplicationSet auto-manage apps as clusters come and go?

Advanced
The cluster generator watches registered clusters; when one is added (matching the selector) it generates that cluster's Applications, and when one is removed it prunes them. This keeps the deployed app set aligned with the live cluster inventory automatically.
generators:
  - clusters:
      selector: { matchLabels: { env: prod } }
Real-world example Registering a new prod cluster automatically creates its apps; deregistering it removes them.

Common follow-ups: What triggers app creation/removal? | How does the selector scope this?

Applications App of Apps Multi-cluster

Can you run ArgoCD per cluster instead of a central hub, and why?

Intermediate
Yes—an ArgoCD per cluster gives isolation (a failure or compromise affects only that cluster), avoids cross-cluster network/credential needs, and localizes blast radius, at the cost of operating many instances. It's chosen when isolation and resilience outweigh central convenience.
Real-world example A security-sensitive org runs ArgoCD in each cluster so no single hub holds all clusters' credentials.

Common follow-ups: What's the trade-off vs hub-spoke? | When is per-cluster preferred?

App of Apps RBAC Multi-cluster