feat(resources): implements service.instance.id

[maryliag]

Which problem is this PR solving?

Implements service.instance.id

Short description of the changes

From the Semantic Convention on Service, we have the values of service.name as required, as service.instance.id, service.version and service.namespace as recommended.
We are not able to set the value of version and namespace by ourselves, but following the semantic convention for service.instance.id, the value can be generated using a randomUUID (we have a similar implementation of this in other SDKs, such as the Java one).
This commit adds a value for service.instance.id by default, which can still be overwritten by the environment variable or cloud provider.

This commit also updates some deprecated strings and add missing sdk.shutdown to all sdk.start on the sdk test file.

How Has This Been Tested?

• Added test on sdk.test
• Tested manually

Checklist:

• Followed the style guidelines of this project
• Unit tests have been added

[codecov]

Codecov Report

Merging #4608 (8628f8a) into main (c046867) will decrease coverage by 2.06%.
Report is 6 commits behind head on main.
The diff coverage is 80.00%.

❗ Current head 8628f8a differs from pull request most recent head 54baf54. Consider uploading reports for the commit 54baf54 to get more accurate results

Additional details and impacted files

@@            Coverage Diff             @@
##             main    #4608      +/-   ##
==========================================
- Coverage   92.83%   90.77%   -2.06%     
==========================================
  Files         329       90     -239     
  Lines        9528     1930    -7598     
  Branches     2053      417    -1636     
==========================================
- Hits         8845     1752    -7093     
+ Misses        683      178     -505     

Files	Coverage Δ
packages/opentelemetry-resources/src/Resource.ts	100.00% <100.00%> (ø)
.../platform/browser/ServiceInstanceIDDetectorSync.ts	100.00% <100.00%> (ø)
...lemetry-resources/src/detectors/EnvDetectorSync.ts	94.11% <50.00%> (ø)

... and 242 files with indirect coverage changes

[trentm]

Thanks for this!

I also consider adding this to the default SDK value, but that would unnecessarily force a call for a randomUUID creation on cases that are not required, so I opted for its creation only when not value is set during the environment detector checks.

I kinda think it should be on the default Resource (

opentelemetry-js/packages/opentelemetry-resources/src/Resource.ts

Lines 52 to 62 in c046867

	static default(): IResource {
	return new Resource({
	[SemanticResourceAttributes.SERVICE_NAME]: defaultServiceName(),
	[SemanticResourceAttributes.TELEMETRY_SDK_LANGUAGE]:
	SDK_INFO[SemanticResourceAttributes.TELEMETRY_SDK_LANGUAGE],
	[SemanticResourceAttributes.TELEMETRY_SDK_NAME]:
	SDK_INFO[SemanticResourceAttributes.TELEMETRY_SDK_NAME],
	[SemanticResourceAttributes.TELEMETRY_SDK_VERSION]:
	SDK_INFO[SemanticResourceAttributes.TELEMETRY_SDK_VERSION],
	});
	}

), but I'm curious for other's opinions.

I think it is a bit odd that one would only get service.instance.id if using the EnvDetector because this default service.instance.id doesn't have anything to do with process.env. values.

I don't think the extra crypto.randomUUID() call is a real perf concern. This suggests it takes on the order of 0.1 microseconds on my machine:

% cat timeuuid.js
const { randomUUID } = require('crypto');
const ben = require('ben');
const ms = ben(1e6, function () {
    randomUUID();
});
console.log('crypto.randomUUID(): %dms per iteration', ms);

% node timeuuid.js
crypto.randomUUID(): 0.000105ms per iteration

% node timeuuid.js
crypto.randomUUID(): 0.000095ms per iteration

% node timeuuid.js
crypto.randomUUID(): 0.00009ms per iteration

It would only be called once per Resource.default() call, which is only once in a basic usage of NodeSDK.

[maryliag]

ah good to hear it doesn't cause such a perf impact!
The place you suggested is exactly the place I tried first. I can add there, just a heads up that it breaks a lot of tests, so the PR will be quite big to fix all places, but hopefully it's easy to review 😄

[trentm]

just a heads up that it breaks a lot of tests, so the PR will be quite big

Thanks for slogging through those.

[maryliag]

Implementation and PR description updates. I will focus on the tests now 😄

[pichlermarc]

Short summary of what we discussed in the SIG Meeting today 🙂

• we'll move the detection of this attribute to it's own resource detector in the @opentelemetry/resources package.
• people that currently set up their SDKs manually can use that new resource detector in the setup code (this is their opt-in)
• we add the detector in NodeSDK if an env var is set AND no list of resource detectors is passed via the resourceDetectors constructor option
  • we can use OTEL_NODE_RESOURCE_DETECTORS for this (which contains a comma seperated list of resource detectors to turn on), the values for existing resource detectors should match this list, we'll discuss the value for the new detector on the PR - instance or serviceinstance comes to mind.

Implementing the resource detector and adding the env var behavior in NodeSDK can be done in seperate PRs to speed up the review process. 🚀

Separately from that:

• I'll implement a way for users to opt-out of auto-merging with the default resource in MeterProvider and TracerProvider akin to [prototype] feat(sdk-trace): add option to opt-out from merging the resource with Resource.default() #4617
• I'll also implement the change in behavior on the next branch (SDK 2.0) which will then be fully spec compliant
  • (I actually committed the metrics part of the crime against the spec in fix(sdk-metrics): use default Resource to comply with semconv #3411 😢 Sorry about that 🙁 )

[maryliag]

Latest changes (and some questions):

1. Created serviceInstanceIDDetector
2. The Browser version does nothing and the Node version sets the value with randomUUID
3. The default resource is working the same way. I removed the changes I had made there (just keep the updates on deprecated values)
4. I added the new detector on the default list of detectors only if the env var OTEL_NODE_EXPERIMENTAL_DEFAULT_SERVICE_INSTANCE_ID is set. The other condition will come on a later PR. Or should I completely remove this part for this PR?

Looking for some help: I added a few tests for a few cases with value coming fro env var VS should get set based on the experimental flag and one of the tests is failing but I don't quite get why.
The test failing is where I don't pass a service.instance.id on env var, but I do set to true the OTEL_NODE_EXPERIMENTAL_DEFAULT_SERVICE_INSTANCE_ID, so I should see a random UUID for its value, but on the test itself, it's returning undefined. When I print here the value of resources, I don't see anything getting set for service.instance.id. This problem is not particular to the implementation of using the detectors, because I was having the same error when setting the value on the defaultResources

[maryliag]

while working on the following PR I realize it makes more sense to have all the logic of the env var on the other one, so I remove those from here. This PR focus on the creation of the service instance ID detector. It currently is not called anywhere, but we have a test showcasing that it works as expected if the user initializes their NodeSDK with it.

The following PR (#4626) will add the OTEL_NODE_EXPERIMENTAL_DEFAULT_SERVICE_INSTANCE_ID and OTEL_NODE_RESOURCE_DETECTORS with their respective logics

[pichlermarc]

Generally looks good 🙂

We can drop the Detector and use DetectorSync only. Other than that, only comments about marking things as @experimental.

Thanks for sticking with us through the review-process 🙂

[maryliag]

@pichlermarc let me know when you want to merge and I can rebase, otherwise as soon as I rebase it's already out of date again 😅

[pichlermarc]

I'll wait a bit with merging to give @trentm another chance to take a look as he initially reviewed the PR. 🙂

(Approvers: feel free to rebase/merge if you come across this and I have not done it yet 🙂 - I'm OOO until Monday)

[trentm]

FYI: A currently open otel spec issue (by some OTel Java implementors) asking about whether the SDK should handle service.instance.id and veering into whether service.instance.id should vary between process restarts: open-telemetry/opentelemetry-specification#3136

[trentm]

LGTM. Dealer's choice in my OCD naming nit.

[trentm]

OCD nit: In other class names "id" has been capitalized as "Id" rather than "ID":

% rg -I 'export class' | rg ID

% rg -I 'export class' | rg Id
export class TraceIdRatioBasedSampler implements Sampler {
export class RandomIdGenerator implements IdGenerator {
export class RandomIdGenerator implements IdGenerator {
export class TraceIdRatioBasedSampler implements Sampler {
export class RandomIdGenerator implements IdGenerator {
export class RandomIdGenerator implements IdGenerator {

Though, given OSDetectorSync consistency is difficult.

[maryliag]

OCD crisis adverted! 😄 Changed to Id