📷 "599 - Network connect timeout error" by GirlieMac is licensed under CC BY 2.0. To view a copy of this license, visit https://creativecommons.org/licenses/by/2.0/.
Multi-Cluster GitOps with ArgoCD 2026: ApplicationSets, Architecture, and Best Practices
Managing multiple Kubernetes clusters has long been a reality for most companies: according to recent surveys, enterprise organizations will run between five and fifty Kubernetes clusters in parallel by 2026 – for different environments, regions, or compliance requirements. This multiplication of clusters causes deployment process complexity to grow exponentially. This is exactly where GitOps with ArgoCD comes in: what began as a tool for individual clusters has become the central platform for multi-cluster deployments.
Architecture Paradigms: Hub-and-Spoke vs. Federated
In the multi-cluster context, two fundamental architecture patterns are available:
Hub-and-Spoke (centralized): A single ArgoCD instance in a management cluster manages multiple target clusters. This model provides a central view of all environments, simplifies RBAC, and allows global policy control from one dashboard. The downside: the management cluster holds highly privileged credentials for all managed clusters – a security risk that must be hardened through VPC peering, private endpoints, and external secret managers.
Federated (decentralized): Each cluster runs its own ArgoCD instance, governed by a meta-ArgoCD layer. This model suits organizations with more than 20 clusters, strict network isolation, or teams needing full autonomy. The credential blast radius remains limited to the individual cluster.
The choice between the two depends largely on the number of clusters, security requirements, and team structure.
ApplicationSets: The Key to Multi-Cluster Automation
The heart of ArgoCD’s multi-cluster capabilities is ApplicationSets. They dynamically generate multiple Application resources from a template and a generator – one per cluster, one per Git directory, or one per team.
The most important generator types at a glance:
List Generator: Explicit enumeration of target environments. Ideal for small, stable sets of clusters.
generators:
- list:
elements:
- cluster: staging
url: https://k8s-staging.example.com
values.replicas: "1"
- cluster: production
url: https://k8s-prod.example.com
values.replicas: "3"
Cluster Generator: Automatically generates one Application per registered cluster. New clusters are selected via labels like environment=production. As soon as a cluster is registered and labeled in ArgoCD, the ApplicationSet deploys the defined application automatically – without manual intervention.
Git Generator (Directory Mode): Generates one Application per subdirectory in the Git repository. This enables self-service onboarding: a team adds a new directory apps/neuer-service/, and ArgoCD automatically creates and synchronizes the corresponding Application.
Matrix Generator: Forms the Cartesian product of two generators. This allows deploying all applications on all clusters without having to define each combination individually. Example: 5 clusters × 8 applications = 40 Applications from a single ApplicationSet.
Merge Generator: Overrides base values for specific clusters. For instance, high-traffic clusters receive more replicas or larger resource quotas, while the base configuration remains the same for all other clusters.
Security in Multi-Cluster Operations
Managing multiple clusters significantly increases the attack surface. Best practices include:
- Dedicated service accounts with minimal permissions per cluster
- Private API endpoints – never expose Kubernetes APIs publicly
- Regular credential rotation for all cluster accesses
- Namespace isolation through ArgoCD projects that restrict target clusters and namespaces
ArgoCD 3.3 also brought important security improvements in 2026, including PreDelete hooks for controlled Application deletions and an improved OIDC token refresh that prevents session timeouts during long monitoring sessions.
GitOps in Production Use in 2026
GitOps has become the standard for Kubernetes deployments – over 64% of companies use it as their primary delivery method, according to surveys. ArgoCD leads the market with its visual application topology, ApplicationSets, and broad ecosystem (Argo Rollouts, Workflows, Events, Image Updater). Competition from FluxCD is strong in decentralized and edge environments, but for centrally managed multi-cluster environments, ArgoCD remains the top choice.
In 2026, the challenge is shifting from “ArgoCD or Flux?” to “How do we run GitOps at fleet scale?” – with a focus on OCI/Helm maturity, progressive delivery, and multi-cluster consistency.
Conclusion
Multi-cluster GitOps with ArgoCD is no longer a nice-to-have, but an operational necessity for companies running Kubernetes at scale. ApplicationSets are the key tool for managing this complexity – they make deployment processes scalable, traceable, and secure. The choice of the right architecture (hub-and-spoke vs. federated) should always be made based on specific security requirements, cluster count, and team structure.
Sources
- ArgoCD Best Practices for Multi-Cluster Management — OneUptime, Februar 2026
- Argo CD ApplicationSet: Multi-Cluster Deployment Patterns (2026) — Coding Protocols, Mai 2026
- ArgoCD vs FluxCD: Which GitOps Tool Should You Use in 2026? — DEV Community, 2026
- GitOps Tools Comparison 2026: ArgoCD vs Flux vs Jenkins X vs Codefresh — NomadX, April 2026
- Multi-Cluster Mgmt Strategies K8s — Kubernetes Recipes, April 2026
🌐 Machine-translated from the German original, editorially reviewed. 🤖 Written with AI assistance.