🔹 Feature: Cross-Cluster Networking for Azure Kubernetes Fleet Manager
🔹 What It Does: Managed multi-cluster pod network for AKS — direct pod-to-pod communication across clusters
What Is It Giving You:
✅ Direct Pod-to-Pod Across Clusters: A flat virtual network across all member clusters. Traffic routes at native performance — no east-west gateways, no proxies, no encapsulation tax.
✅ Powered by Cilium Cluster Mesh: Requires Azure CNI powered by Cilium as the data plane, plus Advanced Container Networking Services (ACNS). Fleet manages the mesh lifecycle via a ClusterMesh profile — no manual cilium clustermesh connect operations.
✅ Global Services — Transparent Service Discovery + Load Balancing: Annotate a standard Kubernetes service with io.cilium/global-service: “true” and Cilium automatically discovers endpoints for that service across every cluster and load-balances traffic to ClusterIP across all of them. No app-level changes, no external service registry.
✅ High Availability & Automatic Failover: Run replicas of the same service across regions or availability zones. When endpoints in one cluster become unhealthy, traffic automatically routes to healthy endpoints in another. Cross-region failover with zero application changes.
✅ Shared-Services Pattern: Run tenant clusters isolated from each other but connected to a shared “services” cluster for secrets management, logging, monitoring, or DNS. Centralize what should be centralized without breaking tenant isolation.
✅ Stateful / Stateless Separation: Keep stateful workloads (databases, storage) in dedicated clusters and let stateless application clusters stay agile. Cluster lifecycle, security, and scaling all become simpler.
✅ Unified L3–L7 Network Policy Enforcement: Cilium network policies applied on one cluster are honored across every other cluster in the mesh. One identity-aware security posture, no manual policy replication.
✅ Multi-Cluster Observability: End-to-end visibility into east-west traffic between services regardless of which cluster the source or destination lives in. Container Network Logs aggregate flow data across the entire mesh — one view for troubleshooting, capacity planning, and traffic analysis.
✅ Scale: Up to 255 member clusters in a single cross-cluster network. One cluster per network at a time.
Relation to Azure Kubernetes Application Network:
🔸 Application Network = Ambient service mesh (L4/L7 mTLS) via east-west gateways — best when you need identity-based mTLS across clusters and network reachability isn’t flat.
🔸 Fleet Cross-Cluster Networking = Flat pod network via Cilium Cluster Mesh — best when you have peered VNets, want native performance, and need direct pod-to-pod without gateways.
🔸 Different tools for different jobs: Pick based on network topology and whether identity-based mTLS or raw performance is the priority.
🌐 https://learn.microsoft.com/en-us/azure/aks/cross-cluster-networking-fleet-use-cases
