An open-data platform schedules Pods successfully on every node, bu…

Question

An open-data platform schedules Pods successfully on every node, but Pods on different nodes cannot ping each other. Control-plane and kubelet appear healthy. What should the administrator investigate first?

Accepted Answer

The CNI plugin and its datapath configuration, because cross-node Pod connectivity depends on a working cluster network implementation. — If Pods land on nodes but can't talk across the building, the hallway wiring — the CNI — is the first place to look, not HPA or TLS Ingress.

Answer

Whether every Pod has been assigned a ClusterIP Service even when no Service was requested.

Answer

Whether HorizontalPodAutoscaler is disabled, because HPA is required for cross-node ICMP.

Answer

Whether Ingress TLS secrets exist in kube-system, because TLS terminates all Pod-to-Pod traffic.

An open-data platform schedules Pods successfully on every node, but Pods on different nodes cannot ping each other. Control-plane and kubelet appear healthy. What should the administrator investigate first?

Short Explanation

Full Explanation