Clarify how the default value of OTEL_EXPORTER_JAEGER_ENDPOINT should work

[abe545]

What are you trying to achieve?
My organization is trying to switch from opentracing to opentelemetry, and currently the .net exporter will not work with our infrastructure, because our jaeger collector listens on a different path than /api/traces. With the opentracing jaeger client, we were able to specify the endpoint and have everything wired up correctly. With the current spec, the default value of the OTEL_EXPORTER_JAEGER_ENDPOINT env var makes it seem like libraries need to append /api/traces themselves to correctly send spans over http.

What did you expect to see?
I'd like it to be clear in the spec how this should behave - should libraries append /api/traces - or should the default value include the path in the uri, and expect developers to configure the full url to export to. If libraries should append the path, can we add another configuration option to override it (this is what my organization needs to adopt otel in .net).

Additional context.
The .net sdk is currently implemented to append /api/traces to the endpoint, but the golang, javascript, and ruby sdks all default the value of the endpoint to include the path, and require users to specify the path in the value.
I opened a PR to try to make the .net sdk configurable: open-telemetry/opentelemetry-dotnet#2847
This led to a discussion where one of the maintainers felt that the golang implementation is wrong: open-telemetry/opentelemetry-go#2599

[pellared]

Related issue in OTel Go SDK open-telemetry/opentelemetry-go#2599

[carlosalberto]

Re-assigning this to @yurishkuro who may know better.

[yurishkuro]

The environment variable should be assumed to contain the actual URL, including the path. The SDK SHOULD NOT append anything to the URL, except, in come cases, the parameters in the query string, such as ?service=*** for remote sampling URL.

If this is not clear in the spec, we should make it clear, and I think the .NET implementation is incorrect.

[abe545]

The environment variable should be assumed to contain the actual URL, including the path. The SDK SHOULD NOT append anything to the URL, except, in come cases, the parameters in the query string, such as ?service=*** for remote sampling URL.

If this is not clear in the spec, we should make it clear, and I think the .NET implementation is incorrect.

Thanks for the clarification @yurishkuro. Would you like me to open a PR making it clear in the documentation?

[yurishkuro]

I am creating one now, will tag you for review.

[yurishkuro]

Actually, something is odd here. The default value in the spec is http://localhost:14250, which is a gRPC port number in Jaeger and thus does not need the path. The default Thrift over HTTP endpoint is http://jaeger:14268/api/traces, which does need the path. Which one are you (@abe545) and the spec are referring to?

[abe545]

@yurishkuro - the docs only mention one: https://github.com/open-telemetry/opentelemetry-specification/blob/9555f9594c7ffe5dc333b53da5e0f880026cead1/specification/sdk-environment-variables.md#jaeger-exporter
And it specifically says it is for HTTP.

[yurishkuro]

I know, what the spec says currently seems inconsistent, it says HTTP but lists the gRPC port number.

[abe545]

Yeah - I'm trying to configure it to for HTTP.

[pellared]

Can we also document the OTEL_EXPORTER_JAEGER_PROTOCOL env var?

I think it is related.

Origin open-telemetry/opentelemetry-dotnet-instrumentation#355 (comment).

How about making the spec similar to https://github.com/open-telemetry/opentelemetry-specification/blob/main/specification/protocol/exporter.md ?

[yurishkuro]

I think the open question to me is what data format other SDKs are using currently. I checked the Go SDL, it uses Thrift over HTTP, and they even made a change to use a different default URL in open-telemetry/opentelemetry-go#1898, with the correct port.

Here's the proposed fix #2333