root_cause_category: infrastructure
required_keywords:
  - node
  - Unknown
  - kubelet
forbidden_categories:
  - healthy
  - unknown
  - resource_exhaustion
  - code_defect
  - data_quality
optimal_trajectory:
  - get_eks_node_health
  - list_eks_pods
  - get_eks_events
max_investigation_loops: 3
ruling_out_keywords:
  - healthy node
  - not at fault
required_queries:
  - get_eks_node_health
  - list_eks_pods
model_response: |
  ROOT_CAUSE: Node ip-10-0-1-73.us-east-1.compute.internal has stopped reporting kubelet heartbeat to the control plane; the node condition has flipped to Ready=Unknown with MemoryPressure=True, and pods scheduled to that node show phase=Unknown with condition Ready=Unknown reason=NodeLost. Sibling node ip-10-0-1-42 is Ready and its pod is Running, isolating the failure to the node rather than the workload.
  ROOT_CAUSE_CATEGORY: infrastructure

  VALIDATED_CLAIMS:
  - Node ip-10-0-1-73.us-east-1.compute.internal condition Ready=Unknown with MemoryPressure=True. [evidence: eks_node_health]
  - Node ip-10-0-1-42.us-east-1.compute.internal is Ready=True with no pressure conditions. [evidence: eks_node_health]
  - Pod on the failing node shows phase=Unknown reason=NodeLost; the pod on the healthy node is Running and Ready. [evidence: eks_pods]
  - Warning event NodeNotReady 'Node ip-10-0-1-73 status is now: NodeNotReady' corroborates the node-level failure. [evidence: eks_events]

  NON_VALIDATED_CLAIMS:
  - Underlying cause is infrastructure-side (kubelet crash, control-plane connectivity loss, or EC2 instance impairment) — workload owners cannot resolve this from inside the pod.
  - Cordoning and draining the unhealthy node, then letting the deployment reschedule, would restore capacity.

  CAUSAL_CHAIN:
  - Kubelet on ip-10-0-1-73 stops reporting to the control plane, the node controller flips Ready to Unknown after the grace period, the scheduler marks pods on that node phase=Unknown, a NodeNotReady event is emitted, pod-eviction-timeout has not yet elapsed, and capacity stays degraded while the healthy node serves all traffic.