@viktorbarzin/metrics

Prometheus metrics for an app: one prom-client registry per process with the default process metrics, the HTTP request metrics every app exports, and /metrics on a port of its own that the ingress never routes (point the pod's prometheus.io/port annotation at it).

import { getLogger } from "@viktorbarzin/log";
import { createHandlerLabeler, getRegistry, observeRequest, parseMetricsPort, startMetricsServer } from "@viktorbarzin/metrics";
import { Counter } from "prom-client";

// at boot (Bun)
const metrics = startMetricsServer({ port: parseMetricsPort(process.env.METRICS_PORT) });

// request middleware
const label = createHandlerLabeler({ known: ["/api/items", "/healthz"] });
const done = observeRequest(request, { label, accessLog: getLogger("myapp.http") });
// ... handle ...
done(response.status);

// app metrics
new Counter({ name: "myapp_things_total", help: "...", registers: [getRegistry()] });

| Export | | |---|---| | getRegistry() | the process registry, with collectDefaultMetrics | | httpMetrics() | http_requests_total{method,handler,status} (status as 2xx), http_request_duration_seconds{method,handler} | | createHandlerLabeler({ known, buckets? }) | known paths by name; default buckets /api/auth/ → /api/auth, /api/ → /api/other, /_serverFn/ → /_serverFn, else page | | observeRequest(request, { label, accessLog?, accessLogPrefix? }) | raises the idle timeout, tracks the request, returns done(status) that records the metrics and one access-log line {method, path, status, duration_ms} under the prefix (default /api/) | | metricsResponse(registry?) | the text format | | parseMetricsPort(raw) | a port or null (ERROR logged for junk) | | startMetricsServer({ port, hostname?, registry? }) | Bun.serve answering /metrics only; null without a port |

Peers: prom-client ^15.1, @viktorbarzin/log, @viktorbarzin/runtime.