Monitoring Kubernetes: A Practical Guide for Small Teams

Kubernetes monitoring guides often assume you have a dedicated SRE team and an enterprise observability budget. Most teams don't. Here's the lean approach.

The 80/20 of Kubernetes Monitoring

What Matters Most

1. Are user-facing services accessible? — External HTTP monitoring
2. Are pods healthy? — Restart counts and readiness
3. Are resources sufficient? — CPU and memory pressure
4. Are deployments succeeding? — Rollout status

What Can Wait

• Per-pod CPU/memory graphs
• Network policy monitoring
• Detailed etcd metrics
• Custom resource monitoring

The Lean Monitoring Stack

External Monitoring (Start Here)

Monitor your Ingress endpoints from outside the cluster. This is your single most valuable check — if users can reach your service and it responds correctly, the cluster is working.

• HTTP monitors on every exposed service
• Keyword checks to verify correct content
• Response time thresholds
• Multi-region checks

Pod Health

Monitor for common Kubernetes failure modes:

• CrashLoopBackOff — Pod keeps crashing and restarting
• OOMKilled — Pod exceeded memory limits
• Pending pods — Can't be scheduled (resource constraints)
• High restart counts — Pods restarting frequently

Resource Pressure

• Node CPU utilization > 80%
• Node memory utilization > 85%
• Persistent volume usage > 80%
• Pod resource requests vs limits

Deployment Health

• Heartbeat after successful deployments
• Rollout status monitoring
• Error rate comparison pre/post deployment

The 30-Minute Setup

1. Add HTTP monitors for all Ingress endpoints (10 min)
2. Set up a basic resource alert on nodes (5 min)
3. Add heartbeat to your deployment pipeline (5 min)
4. Configure Slack alerts for all monitoring (5 min)
5. Create a status page for external users (5 min)

You can add complexity later. Start with what catches 80% of problems in 30 minutes.