40 lines
2.8 KiB
YAML
40 lines
2.8 KiB
YAML
groups:
|
|
- name: EKS-dataengg-Pod-Alerts
|
|
rules:
|
|
- alert: Pod-CPU-Crossed-80%
|
|
expr: 100 * sum by (pod) (rate(container_cpu_usage_seconds_total{image!="",pod!=""}[1m])) / sum by (pod) (kube_pod_container_resource_limits{resource="cpu", job=~"kube-state-metrics-.*"}) * 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=~'dataengg'} > 80
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
annotations:
|
|
summary: High CPU Utilisation for EKS Pod - {{ $labels.pod }}
|
|
description: "Pod {{ $labels.namespace }}/{{ $labels.pod }} has been utilising CPU over {{ $value }}% in the last 1 min.\n VALUE = {{ $value }}\n LABELS = {{ $labels }}"
|
|
|
|
- alert: Pod-Memory-Crossed-80%
|
|
expr: 100 * sum by (pod) (container_memory_working_set_bytes{image!=""}) / sum by (pod) (kube_pod_container_resource_limits{resource="memory", job=~"kube-state-metrics-.*"} > 0) * 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=~'dataengg'} > 80
|
|
for: 1m
|
|
labels:
|
|
severity: critical
|
|
annotations:
|
|
summary: High Memory Utilisation for EKS Pod - {{ $labels.pod }}
|
|
description: "Pod {{ $labels.namespace }}/{{ $labels.pod }} has been utilising Memory over {{ $value }}% in the last 1 min.\n VALUE = {{ $value }}\n LABELS = {{ $labels }}"
|
|
|
|
- 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=~'dataengg'} == 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-.*"}[3m])) * 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=~'dataengg'} > 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 }}"
|
|
|