OpenTelemetry instrumentation for serverpod

License OpenTelemetry Specification

otel_serverpod is an OpenTelemetry instrumentation for serverpod that easily adds standard observability to applications that use serverpod. This package depends on dartastic_opentelemetry_api.

This package can be used with any OTel backend because it's compliant with the OpenTelemetry Specification.

This package supports Linux, macOS and Windows, for versions of serverpod that support those platforms.

This Apache-2.0 package is professionally maintained and supported with 💙 by Michael Bushe, Dart and Flutter Maintainer at OpenTelemetry and the creator of dartastic_opentelemetry and dartastic_opentelemetry_api. This package is professionally supported by Dartastic.io.


Dartastic.io: AI Code Healing and Real User Monitoring for Flutter

See what your users experience. Find problems with Flutter apps in production. Fix them with AI.

Dartastic.io brings AI code healing, real user monitoring, and OpenTelemetry observability together in one platform built for Dart and Flutter. (Patents pending)

Dartastic AI answering "Why did crashes jump after the last release?" over a Flutter App Health dashboard

Dartastic AI — From Production Problems to Code Fixes

Dartastic AI analyzes production telemetry alongside your application source code to diagnose problems and propose fixes.

  • Investigate production issues using crashes, errors, traces, metrics, and real user telemetry.
  • Find the offending code by connecting observed problems with your application's source in Dartastic Pub or your source code repos.
  • Generate code patches that address the underlying problems.
  • Create actionable issues in your issue tracker, complete with findings and the proposed code patch.

Dartastic RUM — See What Your Users See

Understand how people actually experience your Flutter application.

  • Replay user sessions in the Dartastic Observatory session player.
  • Choose your replay fidelity, from privacy-preserving widget outlines to pixel-level reproductions of the user experience.
  • Investigate real user problems using session replay and production observability data.

Dartastic RUM Session Replay

Dartastic Observatory — Observability Built for Flutter

A complete observability backend where Flutter and Dart are first-class citizens.

  • Flutter-specific dashboards for application performance, health, and reliability.
  • Alerts and notifications through email, Slack, and PagerDuty.
  • Dartastic Cloud Observatory — Start with a 30-day free trial on isolated, shared infrastructure, as low as $39/mo.
  • Dartastic Hosted Observatory — Run the observability stack in a private managed environment, sized from Mini to Enterprise XL.
  • Dartastic Self-Hosted Observatory — The whole platform inside your own network. Your keys, your cluster, with nothing leaving your perimeter.

Dartastic Native OTel Runtime — Native OpenTelemetry Performance

Collect production telemetry without burdening Flutter's main isolate.

  • Native OpenTelemetry runtime that moves telemetry processing off the main isolate.
  • Native crash reporting for problems that Dart can't catch.
  • Widget performance diagnostics to identify janky widgets.
  • On-device PII filtering before sensitive data leaves your application.
  • Source-level error diagnostics powered by the Dartastic Symbolizer API.
  • Platform-native metrics for iOS, Android, and Linux, including iOS MetricKit and Android Vitals.
  • OpenTelemetry compatibility — Send telemetry to Dartastic Observatory or any OpenTelemetry platform.

Dartastic Labs — OpenTelemetry Integrations

Instrument more of your application without writing everything yourself.

  • 650+ open-source integrations for Dart and Flutter libraries and services, including Dio, Shelf, Logger, Anthropic, AWS, Azure, and Stripe. Look for otel_* on pub.dev.

More Dartastic Tools

Dartastic Pub

A private Dart package registry for sharing packages and plugins with your team, partners, and customers.

Dartastic Symbolizer

Turn production errors into lines of source code through a web API, while keeping your source code artifacts private.

Professional Support and Training

Dartastic.io offers commercial support for the open-source dartastic_opentelemetry and dartastic_opentelemetry_api packages and the complete Dartastic product suite.

  • Professional support with available around-the-clock coverage and response times under four hours on qualifying plans.
  • OpenTelemetry training for teams instrumenting Dart, Flutter, mobile, and web applications.

Open Standards. No Mandatory Backend.

The Dartastic OpenTelemetry API and SDK are open source and standards-based. You can use them independently of Dartastic's commercial services and send telemetry to any compatible OpenTelemetry backend.

Dartastic's commercial products add native performance, advanced diagnostics, session replay, AI-assisted remediation, and managed observability services.

Explore Dartastic.io


OpenTelemetry instrumentation for Serverpod — server-side complement to the various client-side wrappers (http, dio, grpc, chopper). One http.server span per endpoint method call, with W3C traceparent extracted from inbound headers so distributed traces stitch end-to-end, plus the standard HTTP server metrics. Everything follows the OpenTelemetry semantic conventions and is standard OTLP.

import 'package:dartastic_opentelemetry/dartastic_opentelemetry.dart'
    show OTel;
import 'package:otel_serverpod/otel_serverpod.dart';
import 'package:serverpod/serverpod.dart';

Future<void> run(List<String> args) async {
  // Endpoint + auth come from the standard OTEL_EXPORTER_OTLP_*
  // environment variables.
  await OTel.initialize(serviceName: 'mypod');

  final pod = Serverpod(
    args,
    Protocol(),
    Endpoints(),
    // Database spans (`find person`, `insert order`, ...) per session.
    databaseInterceptor: OtelServerpod.databaseInterceptor,
  );

  // Endpoint spans and request metrics.
  OtelServerpod.install(pod);

  await pod.start();
}

This package depends on the OpenTelemetry API only; your app chooses and initializes the SDK (above, dartastic_opentelemetry). Initialize it before OtelServerpod.install(pod): without an SDK the API hands out no-op tracers and meters, so nothing is recorded.

What one call gives you, zero config

  • One SERVER span per endpoint method call, named POST endpoint.method (semconv {method} {route} form, e.g. POST greeting.hello) with http.route = endpoint.method — Serverpod routes POST /<endpoint> with the method in the JSON body, so the span is renamed once dispatch resolves the route.
  • W3C context extraction (traceparent/tracestate/baggage) from inbound headers. A client using otel_http, otel_dio, or otel_chopper injects those headers automatically, and the endpoint span stitches into the same trace.
  • http.server.request.duration histogram (registry bucket advice, labeled by method, route, and status code) and the http.server.active_requests up-down counter.
  • Exceptions are recorded and rethrown, never swallowed. 5xx flips span status to Error; 4xx does not (per OTel HTTP semconv).

Database spans, one more line

OtelServerpod.databaseInterceptor is a Serverpod constructor parameter, so install cannot retrofit it:

Serverpod(args, Protocol(), Endpoints(),
    databaseInterceptor: OtelServerpod.databaseInterceptor);

Every typed database operation becomes a CLIENT span named {operation} {table} (find person, insert order) parented to whatever span is active — the endpoint span for session operations inside an endpoint call — plus a db.client.operation.duration measurement. transaction wraps its callback so inner operations become children of the transaction span automatically.

db.query.text privacy default

db.query.text is stamped ONLY on the unsafe* raw-SQL operations and ONLY when you opt in with OtelServerpod.install(pod, captureDbStatements: true) — raw SQL can carry sensitive literals, so the default is OFF. The typed CRUD operations never carry query text.

Future calls

Each future-call run becomes an INTERNAL span named after the registered future call, with code.function.name. Its database operations are child spans. Nothing changes in your FutureCall classes; the spans come from the same databaseInterceptor line, plus one diagnostic handler so a call that throws is recorded:

final pod = Serverpod(
  args,
  Protocol(),
  Endpoints(),
  databaseInterceptor: OtelServerpod.databaseInterceptor,
  experimentalFeatures: ExperimentalFeatures(
    diagnosticEventHandlers: [OtelServerpod.diagnosticEventHandler],
  ),
);

Serverpod 4.0 runs future calls through generated wrappers and has no invoke hook. Each run gets a fresh FutureCallSession, though, and the span follows that session: it opens when Serverpod builds the session's database, right before the call body, and ends when Serverpod closes the session. A call that throws is reported by Serverpod as a diagnostic event, which the handler records on the span (exception event, error.type, Error status).

The call body does not run inside the span's zone, so other client spans it starts (an HTTP call, say) are not children of it. Database operations are, because they go through the session.

What is instrumented, exactly

Signal Name Notes
Span POST {endpoint}.{method} kind server, http.request.method, http.route, url.path, url.scheme, network.protocol.version, http.response.status_code, rpc.system.name = serverpod, rpc.method, error.type on failure
Span {operation} {table} (e.g. find person) kind client, db.system.name, db.operation.name, db.collection.name, server.address, server.port; db.query.text only on unsafe* ops and only with captureDbStatements: true (default off)
Span future-call name kind internal, code.function.name, via databaseInterceptor + diagnosticEventHandler
Metric http.server.request.duration histogram, seconds, registry bucket advice
Metric http.server.active_requests up-down counter
Metric db.client.operation.duration histogram, seconds

Names not covered by the semconv registry enums are exported as consts on OtelServerpodSemantics.

Caveats

  • Serverpod version: requires the 4.0 line (>=4.0.0-beta.1 <4.1.0). Serverpod 3.4.x is not supported — the databaseInterceptor constructor parameter and the stable Database surface this package wraps exist only on 4.0.
  • Hot reload (dev mode) re-initializes Serverpod's endpoint connectors, which drops the span naming (POST greeting.hello reverts to POST) until you restart the server. Middleware and metrics keep flowing.
  • Websocket method streams (MethodStreamConnector) pass through untouched in this release — streaming keeps working, stream spans are a tracked follow-up.

License

Apache 2.0

Libraries

otel_serverpod
OpenTelemetry instrumentation for Serverpod.