← All snippets
prometheusmonitoringkubernetespromql

Prometheus: Copy-Paste PromQL Queries for Kubernetes

Ready-to-use PromQL for pod restarts, ArgoCD health, node memory, and CPU load alerts.

Pods restarting or not running

rate(kube_pod_container_status_restarts_total[5m]) > 0
or
kube_pod_status_phase{phase!="Running", phase!="Succeeded"}

ArgoCD unhealthy apps

argocd_app_info{health_status!="Healthy"} > 0

ArgoCD out-of-sync apps

argocd_app_info{sync_status="OutOfSync"} > 0

Node CPU load (15m avg, normalized)

avg(node_load15) / count(node_cpu_seconds_total{mode="idle"}) > 0.8

Node memory > 85%

avg(
  (node_memory_MemTotal_bytes - node_memory_MemFree_bytes
   - node_memory_Buffers_bytes - node_memory_Cached_bytes)
  / node_memory_MemTotal_bytes
) > 0.85

Grafana alert rule (Terraform)

pods_crashing = {
  name            = "Pod Restarting Alert"
  no_data_state   = "OK"
  condition       = "rate(kube_pod_container_status_restarts_total[5m]) > 0"
  triggered_after = "5m"
  annotations = {
    severity = "critical"
    summary  = "Pod keeps restarting for 5+ minutes"
    todo     = "kubectl logs <pod> -n <ns> --previous"
  }
  labels = {
    team    = "devops"
    send_to = "teams_channel_devops"
  }
}

Gotcha

  • Always use a for duration (e.g. 5m) in alert rules — otherwise rolling deployments trigger false alerts
  • node_load15 > 1 is meaningless on multi-core nodes — normalize by CPU count
  • The ArgoCD queries require the ArgoCD metrics exporter to be enabled (server.metrics.enabled: true)