Opentelemetry is not collecting http_server_request_duration_bucket metrics for Nextjs endpoints

[estigma88]

Link to the code that reproduces this issue

https://github.com/estigma88/opentelemetry-nextjs-issue

To Reproduce

1. Clone the following project https://github.com/vercel/opentelemetry-collector-dev-setup/tree/main
2. Run the docker-compose up to start up Prometheus and the otel-collector
3. Clone https://github.com/estigma88/opentelemetry-nextjs-issue
4. Run the nextjs app with npm run dev
5. Go to http://localhost:3000/ several times
6. Wait for a while for the metrics to be collected
7. Go to Prometheus in http://0.0.0.0:9090/query
8. Explore the metrics, you will see http_client_request_duration_bucket but not http_server_request_duration_bucket

Current vs. Expected behavior

Http client metrics are collected but not Http server metrics.

The following is the Opentelemetry configuration:

const sdk = new opentelemetry.NodeSDK({
  resource: resourceFromAttributes({
    [ATTR_SERVICE_NAME]: 'next-app',
  }),
  traceExporter: new OTLPTraceExporter(),
  metricReader: new PeriodicExportingMetricReader({
    exporter: new OTLPMetricExporter(),
  }),
  instrumentations: [getNodeAutoInstrumentations()],
});

sdk.start();

Same configuration using Express works fine.

Provide environment information

Operating System:
  Platform: linux
  Arch: x64
  Version: #63-Ubuntu SMP PREEMPT_DYNAMIC Tue Apr 15 19:04:15 UTC 2025
  Available memory (MB): 31731
  Available CPU cores: 12
Binaries:
  Node: 20.16.0
  npm: 10.8.1
  Yarn: N/A
  pnpm: N/A
Relevant Packages:
  next: 15.4.0-canary.29 // There is a newer canary version (15.4.0-canary.69) available, please upgrade! 
  eslint-config-next: N/A
  react: 18.3.1
  react-dom: 18.3.1
  typescript: 4.7.4
Next.js Config:
  output: N/A
 ⚠ There is a newer canary version (15.4.0-canary.69) available, please upgrade! 
   Please try the latest canary version (`npm install next@canary`) to confirm the issue still exists before creating a new issue.
   Read more - https://nextjs.org/docs/messages/opening-an-issue

Which area(s) are affected? (Select all that apply)

Instrumentation

Which stage(s) are affected? (Select all that apply)

next dev (local)

Additional context

This is how metrics look in Prometheus:

Interestingly enough, traces are collected fine for endpoints:

[stephenliang]

I've found that the OpenTelemetry nodejs auto instrumentations (HTTP being the relevant for this issue) must be run before Next.js itself starts up. This is because the server will start before the http module is monkey patched. Which means you'd have to use the node --register method (but this causes other issues, such as .env not being loaded because that happens as part of the server start)

Alternatively, if you do a top-level registration in instrumentation.ts outside of the register function, it will cause the instrumentations to run before the server executes. When this happens, the server is monkey patched first and then Next.js will call to create a server which is now instrumented. With that, the http.server.request.duration metric will be emitted by the HTTP instrumentation.

I'm using something like this in instrumentation.ts and it appears to be working:

if (process.env.NEXT_RUNTIME === 'nodejs') {
  await import('./instrumentation.node');
  await import('@opentelemetry/auto-instrumentations-node/register');
}

export async function register() {
  // Stubbed to enable opentelemetry instrumentation module BEFORE next.js itself loads
}

[estigma88]

Thanks for spending time debugging the issue.

I tried what you suggest in the instrumentation.ts, but it doesn't work, I don't see any http.server.request.duration, could you share your code? I might be missing something. Btw, I am using npm run dev to test it locally.

Regarding the node --register option, how can I do that on NextJS? is there a way to add it to the package.json scripts?

Interestenly enough, seems like traces work fine, so, I assume instrumentation is working at some degree?