Add Service and Environment dimensions to EMF metrics

[syed-ahsan-ishtiaque]

Summary

When both OTEL_AWS_APPLICATION_SIGNALS_ENABLED and OTEL_AWS_APPLICATION_SIGNALS_EMF_EXPORT_ENABLED are set to true, EMF metrics will now include Service and Environment as dimensions:

• Service: Extracted from service.name resource attribute, falls back to "UnknownService"
• Environment: Extracted from deployment.environment resource attribute, falls back to "lambda:default"
• Dimensions are not added if user already set them (case-insensitive check)

For Lambda, both env vars default to true via the otel-instrument wrapper script.

Changes

1. src/exporter/aws/metrics/emf-exporter-base.ts

   • Added isApplicationSignalsEmfExportEnabled() to check both env vars
   • Added hasDimensionCaseInsensitive() for case-insensitive dimension checking
   • Added addApplicationSignalsDimensions() to inject Service/Environment dimensions
   • Modified createEmfLog() to call the new method

2. lambda-layer/packages/layer/scripts/otel-instrument

   • Added default OTEL_AWS_APPLICATION_SIGNALS_EMF_EXPORT_ENABLED="true" for Lambda

3. test/exporter/aws/metrics/aws-cloudwatch-emf-exporter.test.ts

   • Added 12 new test cases covering all scenarios

Test plan

• npm run lint passes (0 errors)
• npm run compile builds successfully
• All 544 unit tests pass
• Integration test on Lambda verified Service/Environment dimensions in CloudWatch metrics

Manual Testing

Lambda Code

import { metrics, trace } from '@opentelemetry/api';
import { S3Client, ListBucketsCommand } from '@aws-sdk/client-s3';

const tracer = trace.getTracer('test-tracer');
const meter = metrics.getMeter('test-meter');
const requestCounter = meter.createCounter('test.requests', {
  unit: '1',
  description: 'Test request counter'
});

const s3Client = new S3Client({});

export const handler = async (event, context) => {
  // Custom metric emission
  requestCounter.add(1, { operation: 'test' });

  // AWS SDK call
  try {
    await s3Client.send(new ListBucketsCommand({}));
  } catch (e) {
    console.log('S3 call error (expected if no permissions):', e.message);
  }

  // Custom span with HTTP call
  const span = tracer.startSpan('my-custom-span');
  span.setAttribute('key1', 'value1');
  try {
    const response = await fetch('https://aws.amazon.com');
    span.setAttribute('http.status_code', response.status);
    console.log('HTTP call to aws.amazon.com status:', response.status);
  } catch (e) {
    console.log('HTTP call error:', e.message);
    span.setAttribute('error', true);
  }
  span.end();

  return {
    statusCode: 200,
    body: JSON.stringify({
      message: 'Metrics and traces emitted',
      functionName: process.env.AWS_LAMBDA_FUNCTION_NAME
    })
  };
};

CW Metrics

Application Signals Related Metrics

🤖 Generated with Claude Code

[jj22ee]

Is this because we don't want to set serviceName as unknown_service:${process.argv0}, so we fallback to UnknownService instead?

[syed-ahsan-ishtiaque]

This is what we're doing in python as well.
we want to fallback to UnknownService instead of the OTel default unknown_service:node. Do you think this is correct?

[jj22ee]

If the purpose of OTEL_AWS_APPLICATION_SIGNALS_EMF_EXPORT_ENABLED is to only addApplicationSignalsDimensions to EMF, can we use a different Env Var for this?