Observability
Observability is the degree to which a system's internal state can be understood from its external outputs — typically metrics, logs, and traces — allowing engineers to diagnose novel, unanticipated problems without having to ship new code…
Definition
Observability is the degree to which a system's internal state can be understood from its external outputs — typically metrics, logs, and traces — allowing engineers to diagnose novel, unanticipated problems without having to ship new code to add visibility.
Overview
Traditional monitoring answers questions you already thought to ask in advance — is CPU above 90%, is the service returning errors. Observability aims at a broader goal: giving engineers enough rich, correlated data about a system that they can answer questions they didn't anticipate when they built it, which matters enormously in complex microservices systems where failure modes are often novel and specific to a particular combination of conditions. The discipline is commonly described around three foundational pillars: metrics (numeric time-series data, often visualized in Grafana dashboards fed by systems like Prometheus), log aggregation (centralized, searchable event records), and distributed tracing (the path a single request takes across services). OpenTelemetry has emerged as the standard way to instrument applications to produce all three in a vendor-neutral format, which can then flow into platforms like Datadog, New Relic, or Dynatrace. Observability underpins nearly every other practice in the site reliability engineering (SRE) toolkit: you can't set a meaningful Service Level Objective (SLO) without reliable metrics, you can't run chaos engineering experiments without a way to observe what actually happened, and effective incident management depends entirely on being able to quickly localize a novel failure. Courses like CI/CD with GitHub Actions touch on the delivery side of this pipeline, but observability itself is what makes the running system legible once code is deployed.
Key Concepts
- Built on three foundational pillars: metrics, logs, and distributed traces
- Designed to answer novel, unanticipated questions, not just pre-defined dashboards
- Standardized instrumentation increasingly via OpenTelemetry across languages
- Correlates data across pillars to speed up root cause analysis
- Underpins SLO tracking, error budgets, and effective incident response
- Distinguishes 'known unknowns' monitoring from 'unknown unknowns' investigation