Question

A retail company runs a customer-facing web application on a VPC-native GKE cluster exposed through an external HTTP(S) load balancer. During load testing, the network team notices that traffic first reaches a node's kube-proxy and is then redirected to a Pod on a different node, adding an extra hop and causing uneven backend utilization and inaccurate health checks. The team wants the load balancer to distribute traffic directly to the Pods and to see accurate per-Pod health. What should they implement?

Accepted Answer

Enable container-native load balancing by using standalone or Ingress-managed Network Endpoint Groups (NEGs) so the load balancer targets Pod IPs directly.

Answer

Change the Kubernetes Service type to NodePort and add more nodes to the cluster so kube-proxy can spread the load more evenly.

Answer

Set externalTrafficPolicy to Local on a LoadBalancer Service so traffic is only sent to nodes that host a Pod.

Answer

Convert the cluster to a non-VPC-native (routes-based) cluster so Pod IPs are advertised through the VPC route table to the load balancer.