Skip to content

Diagnostics

How to see, attribute, and respond to what's happening inside a running program — language-agnostic principles, with concrete examples in languages/.

Sections

The original three

  • Debugging — interactive debuggers, post-mortem analysis, core dumps, time-travel debugging, when prints beat breakpoints.
  • Logging — levels, structured logs, correlation IDs, log volume vs signal, sampling, retention.
  • Error Handling — exceptions vs result types, sentinel errors, error wrapping, stack traces, recovery vs propagation.

Observability pillars (the rest of the triangle)

  • Metrics — counters, gauges, histograms; cardinality; Four Golden Signals / RED / USE; OpenTelemetry Metrics SDK.
  • Tracing — spans, context propagation, OpenTelemetry SDK, sampling; tying traces to logs and metrics.

Production diagnostics

  • Crash Reporting — Sentry / Crashlytics / Bugsnag flows; symbolication, deduplication, release tagging.
  • Diagnostic Endpoints — /debug/pprof, JMX, JFR, health / readiness probes, in-process REPLs, runtime config toggles.
  • Panic & Recovery — invariant violations, unwinding, signals, "let it crash," recover-at-boundary patterns.
  • Post-Mortem Analysis — core dumps, heap dumps, thread / goroutine dumps, JFR recordings, offline reproduction.
  • Audit Logging — security- and compliance-grade logs; tamper-evidence, retention, sampling-forbidden discipline; separation from operational logs.