-
Notifications
You must be signed in to change notification settings - Fork 576
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Hard coding semconv.SchemaURL references, inconsistencies and brittleness #3657
Comments
This does not sound like changes we can accept. The schema compatibility of a resource is something defined by the specification. Not checking that compatibility would be invalid behavior. It sounds like schema translation is what is needed here. |
Thanks for the feedback @MrAlias. I suspected that to be the case (regarding non-acceptance). Equally so, should there not be any "opinion" on specific versions of the schema URL adopted by packages held within here – or at the very minimum, function arguments to select a responding resource with a specific schema URL? Examples I would highlight include:
Both these explicitly make use of the schema URL As noted elsewhere, an approach that hard codes schema URLs in place within packages inhibits safely upgrading applications that make use of those packages as all parts of an application will need to be amended in a single hit. Further, without having branches + tags with management for different schema URLs, security management of dependencies of this package then is tied to schema URL management. Thinking through this the last couple of weeks it screams at me that any function returning a resource should be required, by convention within the Go implementation of OTel, to provide the requested schema URL that is desired to be used in the returned resource. Is this something that the otel-go (contrib or core) management team been investigating? I note that even the core library explicitly sets the expected resource schema its own detector uses. https://github.com/open-telemetry/opentelemetry-go/blob/main/sdk/resource/container.go#L50 This would tend to indicate that any other schema is invalid according to the core lib for that particular release. It would also imply, wrt semantic versioning (at the least) that a shift to |
Here's a new repo I created to handle schema translations: https://github.com/MrAlias/otel-schema-utils Only thing supported currently is resource translations. |
Similar issue open-telemetry/opentelemetry-go#3769 |
## Which problem is this PR solving? When using detectors with mis-matched Schema URLs, resource.New will return an error. Previously, this was dropped and consequently it looked like configuration worked but many resource attributes would be missing in the resulting telemetry. With the recent addition of #48, this error is much easier to hit. For example, the detectors built into opentelemetry-go [import](https://github.com/open-telemetry/opentelemetry-go/blob/main/sdk/resource/builtin.go#L25) semconv `v1.17.0` but this project [pulls in](https://github.com/honeycombio/otel-config-go/blob/main/otelconfig/otelconfig.go#L21) `v1.18.0`. This means that adding something like `otelconfig.WithResourceOption(resource.WithHost())` results in a mis-configured OTel and no error to warn you what happened. This is all … very bad. This PR is really the bare minimum that needs to be done here. That is, it at least causes a runtime error so that you can tell what's going on — but it doesn't solve the problem that you fundamentally cannot use this library with _any_ other library that doesn't happen to use semconv `v1.18.0`. There are [a number](open-telemetry/opentelemetry-go#2341) of [open](open-telemetry/opentelemetry-go#3769) [issues](open-telemetry/opentelemetry-go-contrib#3657) in OTel and adjacent repositories regarding this problem, which probably warrant further investigation into a long-term sustainable solution. ## Short description of the changes Return the `err` from `resource.New` up through the stack. Update tests to expect this.⚠️ I guess technically this could be considered a backwards incompatible change, in that folks may have configurations that are "working" today but will throw an error after this change. I put that in scare quotes though because it's not _really_ working — it's just not throwing an error. I feel like actually returning an appropriate error here and letting folks know their configuration is likely not working as expected is the right choice. ## How to verify that this has the expected result There's a new test that specifically uses a detector with a different schema URL. Co-authored-by: Vera Reynolds <[email protected]>
And another round of #footgun emerges :( |
Here is the full rundown of the differing semconvs hard coded in the current
|
Hi y'all, hopefully this isn't the wrong place to raise this discussion.
I'm making use of the AWS detectors and it's the root cause of issues I'm facing, in part because the SchemaURL of the detector is hardcoded, and the OTEL specification explicitly negates cross-schema connection of resources.
Adopting a different semconv schema url in my application's instrumentation (in this instance
v1.18.0
) causes breaking to occur due to:Code details
Now, while I can appreciate there is a level of alignment required to schema changes, there should be the opportunity to either "migrate" a resource to a different schema, or at the very least, the documentation around available detectors (or anything else with explicit
semconv.SchemaURL
references) should be documented.open-telemetry/opentelemetry-go#2341
As it stands, the current most up to date release of the Schema URL is
v1.18.0
, and there's no possible way to ask a detector (as it stands) to return a resource with a defined schema (for example, as an attribute). Not only that, the schemas used in this repo also vary (it's not allv1.17.0
), and dependency updates don't seem to really support patched releases (for example, security related) for older semconv schema urls.I have a draft PR that I'll link shortly that removes the explicit
semconv.SchemaURL
references, however I'm wondering if that's considered "bad form" from an Open Telemetry for Golang "best practice".Seeking guidance and advice (other than "use v1.17.0" – until something else breaks because the brittleness of the implementation of OTEL prevents mixed schema operation.
The text was updated successfully, but these errors were encountered: