K8s Healthcheck

Diagnosis rules

• Pending + no events: Likely scheduling issue — check node resources, taints, affinity
• Pending + FailedScheduling: Report which constraint failed (CPU, memory, node selector)
• ImagePullBackOff: Check image name, tag, and registry auth
• CrashLoopBackOff: Read logs from previous container, check OOMKilled
• Running but not Ready: Check readiness probe config and endpoint
• Evicted: Check node disk/memory pressure

Report format

Produce a structured report:

RESOURCE: <name> in <namespace>
STATUS: <Healthy|Degraded|Unhealthy>
PODS: <ready>/<total> ready

ISSUES:
- [SEVERITY] <issue description>
  EVIDENCE: <what you found>
  FIX: <recommended action>

Validation Criteria

• All relevant pods are inspected
• Root cause is identified (not just symptoms)
• Remediation steps are specific and actionable
• No false positives (don't flag healthy resources)

Example

Input: "Check the health of the nginx deployment in namespace demo"

Expected tool sequence:

1. k8s_get_resources (get deployment)
2. k8s_get_resources (list pods)
3. k8s_describe_resource (describe unhealthy pods)
4. k8s_get_pod_logs (get container logs)
5. k8s_get_events (check for error events)

Expected output: "RESOURCE: nginx in demo | STATUS: Healthy | PODS: 3/3 ready | No issues detected."