What you don’t measure, you can’t manage. Here is a complete guide to monitoring.

Three Pillars of Observability¶

• Metrics — numerical data (CPU, latency, error rate)
• Logs — text records of events
• Traces — the path of a request through the system

Metrics — Prometheus¶

The Complete Guide to Monitoring¶

Counter — monotonically increasing (requests_total) Gauge — current value (temperature) Histogram — distribution (request_duration_seconds) Summary — percentiles

Logs — Loki¶

Structured JSON logs → central storage → query and alerting.

Traces — Jaeger/Tempo¶

Distributed tracing tracks a request across all microservices. Essential for debugging distributed systems.

SLI/SLO/SLA¶

• SLI (Indicator) — what you measure (P99 latency, availability)
• SLO (Objective) — target (99.9% availability)
• SLA (Agreement) — contract with the client (99.9% + penalties)

Error Budgets¶

SLO 99.9% = 43 minutes downtime/month = error budget. If you exhaust it, stop new features and fix reliability.

Recommended Stack¶

• Metrics: Prometheus + Grafana
• Logs: Loki + Promtail + Grafana
• Traces: Tempo or Jaeger
• Alerting: Alertmanager + PagerDuty/OpsGenie
• All-in-one: Grafana Cloud (free tier)

Principle¶

Monitor symptoms (error rate, latency), not causes (CPU). Alert on what affects users.