groups:
- name: EKS-infra-Pod-Alerts
  rules:
  - alert: Pod-Not-in-Ready-State
    expr: kube_pod_status_phase{phase!='Running', phase!='Succeeded', phase!='Pending', job=~'kube-state-metrics-p-.*'}  * on(pod) group_left(label_bu, label_team, label_service, label_env, label_priority, namespace, cluster) kube_pod_labels{job=~"kube-state-metrics-.*", label_env=~'prod|prd',label_bu=~'infra'} == 1
    for: 3m
    labels:
      severity: critical
    annotations:
      summary: EKS Pod - {{ $labels.pod }} not in Ready State
      description: "Please connect with DevOps-POC of this service for this Issue. Pod {{ $labels.namespace }}/{{ $labels.pod }} is in {{ $labels.phase }} state for last 3 mins.\n  STATE = NotReady\n  LABELS = {{ $labels }}"

  - alert: Pod-Restarted-Multiple-Times
    expr: sum by (pod,namespace) (increase(kube_pod_container_status_restarts_total{job=~"kube-state-metrics-.*"}[2m])) * on(pod) group_left(label_bu, label_team, label_service, label_env, label_priority, namespace, cluster) kube_pod_labels{job=~"kube-state-metrics-.*", label_env=~'prod|prd',label_bu=~'infra'} > 1
    for: 3m
    labels:
      severity: critical
    annotations:
      summary: EKS Pod - {{ $labels.pod }} Restarted Multiple Times
      description: "Pod {{ $labels.namespace }}/{{ $labels.pod }} has Restarted for more than {{ $value }} times in last 3 mins.\n  VALUE = {{ $value }}\n  LABELS = {{ $labels }}"