When using auto-instrumentations-node, some values of OTEL_NODE_RESOURCE_DETECTORS are treated as invalid

[qbedard]

What version of OpenTelemetry are you using?

@opentelemetry/[email protected]
@opentelemetry/[email protected]
@opentelemetry/[email protected]
@opentelemetry/[email protected]
@opentelemetry/[email protected]
@opentelemetry/[email protected]
@opentelemetry/[email protected]
@opentelemetry/[email protected]
@opentelemetry/[email protected]

What version of Node are you using?

18.12.1

What did you do?

Use auto-instrumentations-node with OTEL_NODE_RESOURCE_DETECTORS set to aws,env,host,os.

What did you expect to see?

AWS resource detector is enabled.

What did you see instead?

Invalid resource detector "aws" specified in the environment variable OTEL_NODE_RESOURCE_DETECTORS

Additional context

This conditional is evaluating to false-y and appears to do so for all resource detectors that are of type Detector or Detector[]. Those that have values of DetectorSync or DetectorSync[] do work, so the issue is likely that a Detector instance is evaluating to false-y and the linked conditional or the detector map needs to be updated to evaluate those values properly.

[trentm]

@qbedard Thanks for the issue, and the PR.

I believe what is happening here is that those Invalid resource detector "aws" specified ... diag.error messages are actually coming from the underlying @opentelemetry/sdk-node package that @opentelemetry/auto-instrumentations-node is using.

For example:

% cat foo.js
console.log('hi')
[16:59:00 trentm@peach:~/tmp/asdf]
% OTEL_NODE_RESOURCE_DETECTORS=foo,gcp,alibaba,aws,env,host,os node -r @opentelemetry/auto-instrumentations-node/register foo.js
Invalid resource detector "foo" specified in the environment variable OTEL_NODE_RESOURCE_DETECTORS
Invalid resource detector "foo" specified in the environment variable OTEL_NODE_RESOURCE_DETECTORS
Invalid resource detector "gcp" specified in the environment variable OTEL_NODE_RESOURCE_DETECTORS
Invalid resource detector "alibaba" specified in the environment variable OTEL_NODE_RESOURCE_DETECTORS
Invalid resource detector "aws" specified in the environment variable OTEL_NODE_RESOURCE_DETECTORS
OTEL_TRACES_EXPORTER is empty. Using default otlp exporter.
OpenTelemetry automatic instrumentation started successfully
hi
Process is not running on K8S [Error: ENOENT: no such file or directory, access '/var/run/secrets/kubernetes.io/serviceaccount/token'] {
  errno: -2,
  code: 'ENOENT',
  syscall: 'access',
  path: '/var/run/secrets/kubernetes.io/serviceaccount/token'
}

Observe that there are two diag error messages about the invalid "foo" resource detector name.
What is happening in:

1. auto-instrumentations-node runs its getResourceDetectorsFromEnv (

   opentelemetry-js-contrib/metapackages/auto-instrumentations-node/src/register.ts

   Line 30 in 9dc55da

   resourceDetectors: getResourceDetectorsFromEnv(),

   ), which successfully loads the "gcp", "alibaba", and "aws" detectors.
2. Then it calls new NodeSDK (from @opentelemetry/sdk-node). NodeSDK's constructor calls its getResourceDetectorsFromEnv, even if configuration.resourceDetectors is passed in: https://github.com/open-telemetry/opentelemetry-js/blob/a037f84b679386daefd726133be6265c3c43de27/experimental/packages/opentelemetry-sdk-node/src/sdk.ts#L123-L132

I believe an appropriate fix for this would be to improve the NodeSDK constructor to not do the defaultDetectors processing if configuration.resourceDetectors is passed in. That would get rid of some unnecessary processing and would get rid of the diag.error messages.

---

Would you be interested in attempting a PR that does that?