You do not need a full observability platform to sleep better. A focused Prometheus + Grafana setup answers three questions: is the service up, is it slow, and are we about to run out of disk?

1. Scrape what you will act on

  • Node exporter — CPU, memory, disk, network
  • App metrics — request rate, latency, error rate (RED)
  • Blackbox probes — HTTPS uptime from outside the host

Skip vanity metrics until the basics page on-call.

2. Four alerts that earn their keep

  • Instance down for > 2 minutes
  • Disk > 85% for 15 minutes
  • p95 latency above SLO for 10 minutes
  • 5xx error rate spike vs baseline

Every alert needs an owner and a runbook link. No runbook = noise.

3. Dashboards with a job

One “service health” dashboard beats ten decorative panels. Show golden signals, recent deploys, and saturation. Use Grafana folders per environment (staging / prod).

4. Retention and cost sanity

Keep local Prometheus retention modest (7–15 days) for small teams. Export long-term metrics only if you have a compliance or capacity-planning need. Back up Grafana dashboards as JSON in git.

Need monitoring that pages the right person?

We set up Prometheus, Grafana, and alert routing for cloud workloads.

Book Free Consultation