oxygen914/kubeoncall/node-notready-triage

Node NotReady Triage

Diagnostic objective

Determine whether the node, kubelet, runtime, network path, or monitoring target is unavailable and identify derived workload impact.

Required live evidence

Confirm Ready condition and transition time, lease freshness, kubelet and runtime status, pressure conditions, node events, control-plane reachability, exporter health, and affected Pods. Compare with another node and monitoring path.

Decision branches

• Ready is false or unknown and lease is stale: prioritize node or kubelet reachability.
• Node is Ready but exporter is down: treat NodeDown as a monitoring-path failure until proven otherwise.
• Disk, memory, or PID pressure is present: hand off to the corresponding node resource Skill.
• Many Pod alerts share the node: treat them as derived symptoms unless independent Pod evidence exists.

Safe next step

Use the runbook to capture the node timeline, failure domain, redundancy, and workload impact before proposing a governed node action.

Prohibited actions

Do not cordon, drain, evict, restart kubelet, reboot, or modify node configuration from this Skill.

Recovery and escalation

Require stable Ready and lease signals, healthy runtime and monitoring, and recovered workloads across two windows. Escalate immediately for multiple nodes, control-plane reachability loss, or critical single-node workloads.