observability

The ability to understand the internal state of a system by examining its outputs — logs, metrics, and traces (the 'three pillars' of observability). Observability goes beyond monitoring (alerting on known failure modes) to enable debugging unknown problems: if you can ask an arbitrary question about system behavior and answer it from instrumentation, the system is observable. Modern observability tools include Datadog, Honeycomb, Grafana, and OpenTelemetry.

When you'd see it: Platform engineering discussions, incident post-mortems, SRE team scope, vendor procurement.

Why it matters: Unobservable systems make incidents longer and more expensive to resolve. Observability is the difference between debugging in the dark and debugging with a full picture of what the system was doing.

Common mistakes: Conflating observability with monitoring. Monitoring checks known conditions (is the server up? is latency below threshold?). Observability enables answering novel questions about unknown problems.

Related logging, metrics, distributed tracing, SLO