Question

A team exposes an application using a NodePort Service with externalTrafficPolicy set to Local. Clients report that requests to some cluster nodes' IP:NodePort succeed while requests to other nodes' IP:NodePort time out. The Deployment has 2 replicas and all pods are Running and Ready. The Service has correct selectors and populated endpoints. What is the MOST likely cause of the intermittent connectivity?

Accepted Answer

With externalTrafficPolicy: Local, only nodes that actually run a backing pod respond to NodePort traffic; nodes without a local pod drop the requests.

Answer

kube-proxy is not running on the nodes that time out, so no iptables rules for the NodePort exist there.

Answer

The Service's endpoints are stale and need to be manually refreshed with kubectl edit endpoints.

Answer

The NodePort range is misconfigured in the kube-apiserver, so only some nodes allocated the port.