Observability for Bun and Elysia Applications

Add logs, metrics, request IDs, and tracing boundaries to Elpod applications; wire vendor exporters through application providers.

Elpod provides request identity and observability boundaries for native Elysia routes on Bun. Use them to connect application logs, metrics, and traces to your chosen telemetry SDK and exporter.

The idea

Production is a dark forest when all you have is “it failed”. Logs tell the story, metrics show the shape, and traces connect the scenes. Good observability is a trail of safe breadcrumbs—not a copy of every request and secret.

How Elpod provides it

Elpod’s observability API is vendor-neutral. Pass a Logger, Tracer, or Metrics adapter to bootstrap()/start() and keep your vendor SDK in application infrastructure.

import { bootstrap, consoleLogger } from "@elpod/core";

const server = await bootstrap(app, {
  logger: consoleLogger(),
  tracer: tracerAdapter,
  metrics: metricsAdapter,
});

The built-in logger emits safe structured request events with method, path, status, duration, request ID, correlation ID, and error code. It does not log bodies, credentials, headers, or query values. Request spans use http.server; metrics include http.server.requests and http.server.duration_ms.

Valid W3C traceparent/tracestate headers become parent context. HttpClient can inject outbound context. Database, cache, event, job, security, and scheduler boundaries accept optional operation telemetry where implemented.

OpenTelemetry bridge

openTelemetryTracer(api) and openTelemetryMetrics(provider) adapt compatible OpenTelemetry API objects without making the framework depend on an SDK.

import { openTelemetryMetrics, openTelemetryTracer } from "@elpod/core";

const server = await bootstrap(app, {
  tracer: openTelemetryTracer(otelApi, { name: "orders" }),
  metrics: openTelemetryMetrics(otelMeterProvider, { name: "orders" }),
});

You still install/configure the SDK, exporters, resource attributes, sampling, and collector. Elpod is not an OpenTelemetry distribution.

Common mistakes

Putting full URLs, query values, tokens, email addresses, or payloads in span attributes.
Assuming telemetry adapters are always available or never fail.
Creating high-cardinality metric labels from user IDs.
Using an in-memory inspector endpoint in production without access control.

Production notes

Use sampling and retention appropriate to your data policy, redact at the adapter boundary, and make telemetry failures non-blocking. Correlate logs, traces, and metrics with request/correlation IDs, but do not make those IDs authorization credentials.

Errors · Health and shutdown · Deployment.