Building Resilient Multi-Cluster Kubernetes (2026) | Karmada, Cluster API, Cilium Cluster Mesh, Submariner, Velero & More

Why Multi-Cluster, and What Resilience Requires

3 entries

Teams run more than one Kubernetes cluster for several reasons: to survive the loss of a zone, region or whole cloud, to keep data in a given jurisdiction, to isolate environments and tenants, to place workloads near users, and to limit the blast radius of upgrades and mistakes. Multi-cluster does not make an application resilient by itself, and it adds cost and complexity. Resilience comes from the whole design: stateless services that can run in more than one place, a plan for state, traffic management that can move users away from a failed cluster, repeatable cluster creation and GitOps so any cluster can be rebuilt, tested backups and restores, observability that spans clusters, consistent policy, and regular failure testing. A useful starting point is to define recovery time and recovery point objectives per workload, then choose the simplest pattern that meets them, such as active-passive before active-active. This page links to the main projects and documentation in each area.

01Production environmentKubernetes guidance on production-grade clusters, including control plane availability. 02Running in multiple zonesOfficial best practices for spreading a cluster across failure domains. 03Multi-cluster SIGKubernetes special interest group for multi-cluster APIs and concepts.

Fleet Management & Workload Placement

14 entries

Deploy and keep applications in sync across clusters.

04KarmadaCNCF project for running applications across multiple clusters with Kubernetes-native APIs. 05Karmada repositorySource code and releases. 06Open Cluster ManagementCommunity project for managing multiple clusters, which Red Hat's advanced cluster management is built on. 07OCM repositorySource code. 08Argo CD ApplicationSetsGenerate Argo CD applications across many clusters from templates. 09Argo CDDeclarative GitOps continuous delivery for Kubernetes. 10FluxGitOps toolkit that supports multi-cluster setups. 11Rancher FleetGitOps-style fleet management for large numbers of clusters. 12KubeStellarMulti-cluster configuration management. 13ClusternetProject for managing and deploying across clusters. 14LiqoDynamic multi-cluster peering that lets workloads offload to remote clusters. 15Cluster Inventory APISIG Multicluster work on a standard way to list and describe clusters. 16KubeFed (retired)The original Kubernetes federation project; archived in April 2023, so it is listed only as history. 17AdmiraltyMulti-cluster scheduler; the last repository commit was in November 2024, so check its activity before adopting it.

Cluster Lifecycle & Provisioning

6 entries

Create, upgrade and rebuild clusters repeatably.

18Cluster APIDeclarative cluster creation, upgrade and operation using Kubernetes-style APIs. 19Cluster API repositorySource code. 20CrossplaneControl plane for provisioning cloud infrastructure with Kubernetes APIs. 21KubermaticPlatform for managing clusters across clouds and on premises. 22vClusterVirtual clusters for isolation and cheaper test environments. 23kindRun local Kubernetes clusters in containers for testing multi-cluster setups.

Cross-Cluster Networking & Service Mesh

7 entries

Let services in different clusters find and call each other.

24Multi-Cluster Services APIStandard API for exporting and importing services across clusters. 25Cilium Cluster MeshConnect clusters with shared service discovery and network policy. 26Istio multiclusterInstall and operate Istio across several clusters. 27Linkerd multiclusterService mirroring across clusters. Check Linkerd's release information, because stable releases are published through Buoyant's distribution. 28SubmarinerDirect pod and service connectivity between clusters. 29SkupperApplication-layer connectivity across clusters and sites. 30Gateway APIStandard Kubernetes API for traffic routing.

Global Traffic & Failover

4 entries

Move users away from a failed cluster.

31k8gbCloud-native global server load balancing for Kubernetes. 32ExternalDNSPublishes Kubernetes services and ingresses to DNS providers. 33Topology spread constraintsSpread pods across zones and nodes inside a cluster. 34Pod disruption budgetsKeep a minimum number of replicas available during disruptions.

Backup, Disaster Recovery & State

4 entries

Protect and restore the stateful parts.

35VeleroBackup, restore and migration of cluster resources and volumes. 36Velero repositorySource code; the project moved from the vmware-tanzu organization. 37RookCloud-native storage orchestration, including Ceph. 38CockroachDB multi-regionMulti-region database design; check current licensing terms.

Observability, Policy & Chaos Testing

10 entries

See, govern and test the fleet.

39ThanosGlobal query view and long-term storage for Prometheus. 40Grafana MimirHorizontally scalable long-term metrics storage. 41CortexScalable, multi-tenant Prometheus storage. 42Prometheus federationBuilt-in way to pull selected metrics from other Prometheus servers. 43OpenTelemetryVendor-neutral telemetry standard. 44KyvernoKubernetes-native policy engine. 45OPA GatekeeperPolicy enforcement with Open Policy Agent. 46LitmusChaosChaos engineering for Kubernetes. 47Chaos MeshChaos engineering platform for Kubernetes. 48Google SRE bookReliability engineering practices behind recovery objectives and incident response.

Managed Fleet Services

6 entries

Cloud and vendor offerings.

49GKE fleet managementGoogle Cloud's documentation for managing groups of clusters. 50Azure Kubernetes Fleet ManagerMicrosoft's service for orchestrating updates and placement across AKS clusters. 51Azure Arc-enabled KubernetesManage Kubernetes clusters anywhere from Azure. 52Amazon EKS AnywhereAWS's option for running Kubernetes clusters on your own infrastructure. 53Red Hat Advanced Cluster ManagementMulti-cluster management for OpenShift and other Kubernetes. 54RancherMulti-cluster Kubernetes management platform.

12 entries

Neighbouring pages on this site.

55KubernetesKubernetes hub. 56Kubernetes Cluster Management ToolsTools for managing clusters. 57Multicloud Container PlatformsContainer platforms across clouds. 58Multicloud Management PlatformsManaging multiple clouds. 59Service MeshesService mesh implementations. 60Kubernetes Networking SolutionsNetworking for Kubernetes. 61Chaos Engineering in KubernetesChaos tools for Kubernetes. 62Kubernetes Monitoring ToolsMonitoring Kubernetes. 63Kubernetes Policy EnginesPolicy engines for Kubernetes. 64Cloud-native Data ProtectionBackup and recovery for Kubernetes. 65Edge Kubernetes PlatformsKubernetes at the edge. 66Site Reliability EngineeringSRE practices and tools.

Frequently Asked Questions

Why run more than one Kubernetes cluster?

For availability across zones, regions or clouds; data residency; isolation of environments or tenants; proximity to users; and smaller blast radius for upgrades and mistakes. It also adds cost and operational complexity, so use it when a requirement justifies it.

Is multi-cluster the same as multi-region?

No. A single cluster can span zones, and multiple clusters can sit in one region. Multi-cluster describes how many clusters you run, while multi-region describes where they are. Resilience to a regional failure needs capacity and data in more than one region.

What happened to Kubernetes Federation (KubeFed)?

KubeFed was archived in April 2023. Newer projects such as Karmada, Open Cluster Management and GitOps tools like Argo CD ApplicationSets cover similar needs.

Active-active or active-passive?

Active-passive is simpler and often enough when recovery time objectives allow a short failover. Active-active reduces failover time but needs applications and data stores that handle multiple writers or regions. Choose from your recovery objectives.

How do I know my setup is resilient?

Test it. Practice restoring from backup, rebuilding a cluster from code, failing traffic over, and injecting faults with chaos tools, and fix what breaks. Untested failover plans often fail when needed.

How current is this page?

Links and project activity were checked in October 2026. Multi-cluster Kubernetes tooling changes quickly, and several projects have moved, been retired or changed licences, so confirm status in each project's own documentation. Sites that block automated checks are kept and may need a manual click.

What does Karmada add for multi-cluster Kubernetes?

CNCF announced Karmada's graduation on 7 September 2026. It extends the standard Kubernetes API with centralized placement, propagation, failover and multi-cluster autoscaling.

What is the current Argo CD release?

On GitHub in October 2026 Argo CD v3.5.6 was marked latest (20 October), with v3.4.12 and v3.3.17 released the same week. Argo CD makes a minor release every three months and patches only the three most recent minor versions. The 3.4 to 3.5 upgrade notes say Helm was upgraded to 4.2.0, so Helm 4 now renders all charts and an app's helm.version v3 setting is ignored, plain-HTTP OCI registries need --insecure-oci-force-http, UI extensions must be rebuilt for React 19, and GnuPG signature verification was replaced by Source Integrity. Argo is a CNCF graduated project.