Cloud & Compute Infrastructure

Google Kubernetes Engine introduced preview in-place migration to Dataplane V2, allowing eligible clusters to transition without parallel cluster setups.

· ComputeLabs Research · from the November 19, 2026 edition

Google introduced preview migration from Google Kubernetes Engine (GKE) Dataplane V1 to Dataplane V2 for eligible virtual private cloud (VPC)-native clusters. The stated purpose is to avoid the parallel-cluster setup previously associated with migrating production workloads that have established network-address allocations and stateful applications.

Dataplane V2 uses extended Berkeley Packet Filter (eBPF) and open-source Cilium, replacing traditional iptables routing and kube-proxy. Google reports support for up to 15,000 nodes with network-policy enforcement, and up to 65,000 nodes without enforcement, explicitly noting that the latter configuration requires recreating clusters.

The dataplane supports GPU and Cloud Tensor Processing Unit (TPU) clusters, as well as GKE managed DRANET for Dynamic Resource Allocation (DRA) of network devices. Security and observability features include Kubernetes NetworkPolicy, hierarchical ClusterNetworkPolicy, support for sidecarless ambient networking, network-policy logging, live traffic topology, and Hubble flow metrics.

Migration preflight checks include verifying sufficient Pod address capacity to recreate nodes and using Active Assist to identify incompatible configurations, such as custom eBPF programs on managed interfaces. The excerpt describes initiating migration through the Google Cloud command-line interface, but does not supply the full workflow or establish that migration is disruption-free.

  • Google Kubernetes Engine
  • Dataplane V2

All 19 stories from November 19, 2026