Observability
Prometheus metrics
Section titled “Prometheus metrics”Enable the scrape endpoint with METRICS_ENABLED=true (runtime projects) or
directly through options:
app, err := runtime.New(runtime.Options{ Metrics: runtime.MetricsOptions{Enabled: true}, // serves GET /metrics})Every request is recorded as http_requests_total{method,route,status},
http_request_duration_seconds{method,route}, and http_requests_in_flight.
The route label uses the matched pattern (/tasks/:id, never /tasks/123),
so path parameters cannot explode label cardinality; unmatched requests are
grouped under unmatched. Register custom metrics on
app.Metrics().Registry().
Request-scoped logging
Section titled “Request-scoped logging”Handlers retrieve a logger already carrying the request ID, method, and path:
httpx.Logger(c).Info("task created", "task_id", task.ID)Inject your base logger through Options.Logger; access logs and the
request-scoped logger share it. Without the runtime middleware the accessor
falls back to slog.Default(), so handler code stays testable.
Profiling
Section titled “Profiling”PPROF_ENABLED=true (or runtime.PProfOptions{Enabled: true}) mounts the
standard library profiler under /debug/pprof.
Never expose /metrics or /debug/pprof publicly: keep them behind your
ingress rules or network policy. The pprof endpoints reveal process internals
and can degrade performance under profiling load.