A subnetwork, or subnet, is an IP address range set aside for a networking purpose. An Anthos GKE cluster needs capacity for three kinds of addresses: nodes and API servers, Pods, and Services. Plan these ranges before deployment so the cluster can grow without exhausting its addresses. section 10.2.1 section 10.2.2
Size each range for what consumes it:
- Nodes and API servers each need an address on the LAN or VPC. For GKE on GCP, node addresses come from the primary subnet. Estimate the node pool from the number of clusters in a region and the maximum nodes per cluster; a single node subnet can serve multiple clusters. section 10.2.1 section 10.2.2
- Pods draw from a cluster Pod CIDR. The control plane divides that larger range into equal, node-specific ranges. Size it for both maximum nodes and Pods per node: a Pod moved to another node receives an address from its new node’s range. section 10.2.1 section 10.2.2
- Each ClusterIP Service needs an address from the service range. Its ClusterIP stays stable as the Pods behind it change and serves as an internal virtual IP; it is not reachable from outside the cluster. section 10.2.1 section 10.2.2
Dividing a Pod range
With a Pod CIDR of 10.0.0.0/16 and /24 allocated per node, the first node receives 10.0.0.0/24, the second 10.0.1.0/24, and subsequent nodes receive further slices. Reserve enough slices for the planned maximum node count. section 10.2.2
Deployment mode changes the overlap decision. In island mode, Pods communicate within a cluster but cannot be reached from outside it, allowing the same IP scheme across clusters. In VPC-native mode, cluster addresses are VPC addresses and must not overlap; plan distinct allocations, even where clusters share a node subnet. Pod reachability from outside a cluster also depends on its networking implementation. section 10.2.2 section 10.2.1
Use the Pods-per-node table to find the host bits needed when sizing the Pod CIDR alongside the maximum node count. section 10.2.2
