Project
Kubernetes observability
Prometheus and Grafana monitoring tuned so alerts are actionable — including distinguishing node-level from container-level failures.
What it is
Monitoring and alerting for Kubernetes infrastructure: the rules, dashboards and signal choices that make the difference between an alert that helps during an incident and one that just makes noise.
Why it exists
Default metrics are plentiful and mostly useless on their own. An alert needs to tell the person on call what broke, where, and whether it’s still happening.
How it works
- Prometheus rules encode the signals worth reacting to.
- OOM and memory-pressure alerts distinguish node-level pressure from container-limit violations.
- Alert freshness is considered so already-resolved problems don’t page anyone.
- Alerts carry context — node, namespace, workload — so diagnosis starts from the alert itself.
- Grafana dashboards cover the infrastructure and the workloads running on it.
Notes
- A repeated lesson: the value is in what you don’t alert on.
- See the draft write-up.
Status
Active / ongoing.