Question

A developer reports that a Pod named 'frontend' on node worker-1 cannot reach a Pod named 'backend' on node worker-2 by its Pod IP, though both Pods reach other Pods on their own node. There are no NetworkPolicies in the cluster, kube-proxy is running on all nodes, and both Pods are Running. You confirm the Pod IPs are in the expected Pod CIDR range. According to the Kubernetes network model, which explanation most likely accounts for this cross-node failure?

Accepted Answer

The CNI plugin's overlay/routing is broken between nodes, so Pod traffic cannot traverse the node-to-node network as the flat model requires.

Answer

A ClusterIP Service must be created for backend, because Pods cannot communicate using raw Pod IPs across nodes.

Answer

kube-proxy must be restarted, since it is responsible for establishing the routes that carry Pod IP traffic between nodes.

Answer

CoreDNS is misconfigured, preventing the frontend Pod from resolving the backend Pod's IP address across nodes.