Missing trace_id inside http_server/client_*

[JecminekTomas]

Hello, I'd like to pair Prometheus metrics with Tempo traces, but there is no trace_id, what so ever.

I've saw Prometheus connection is being setup to show an exemplar.

But, inside http_server_/http_client_duration_miliseconds_*, there is no trace_id. So there is no way for me to join those records.

There are some examples:

• http_server_duration_milliseconds_bucket{job="some-backend",le="0",method="GET",path="/lists/countries",service_name="some-backend",status="304"}
• http_client_duration_milliseconds_sum{http_flavor="1.1",http_method="GET",http_status_code="404",job="some-backend",net_peer_name="api.stripe.com",net_peer_port="443",service_name="some-backend"}
• http_server_request_count_requests_total{job="soe-backend",method="GET",path="/lists/currencies",service_name="some-backend"}

Am I doing something wrong? In advance, thank you for your help.

[zeitlinger]

You have enable exemplars - they are the dots on the bottom:

This is a view of using the example app

[JecminekTomas]

@maryliag, could you please clarify, it is quite important for me.

[zeitlinger]

You can also reach out to the community: https://opentelemetry.io/community/

[JecminekTomas]

@zeitlinger, thank you. I will.

[maryliag]

@JecminekTomas I apologize, for some reason I didn't get a notification for this. About the JS SDK, I can see that you're using auto-instrumentation, and that initialized the instrumentation-http that generates http.server.request.duration.
That metric was generated previously with milliseconds, then we added the stable metric that sends in seconds, but currently you need to turn that on.
When we make a change in metrics like this, we need to keep the default for several months, then we can turn that old behaviour off and keep just the latest on by default (The plan for this one is to change the default behaviour in June 2026 so only stable is sent). Since you said you're receiving the metric, just with the wrong unit, I suspect you just need to turn that option on.

To do this, use the OTEL_SEMCONV_STABILITY_OPT_IN environment variable.

• http: emit the new (stable) v1.23.0 semantics
• http/dup: emit both the old v1.7.0 and the new (stable) v1.23.0 semantics
• By default, if OTEL_SEMCONV_STABILITY_OPT_IN includes neither of the above tokens, the old v1.7.0 semconv is used.

You can find this info on the README of the instrumentation-http package here

[JecminekTomas]

@maryliag, no worries. I'm very thankful, you've responded! :) Thank you, so much for your help. Now, both seconds and milliseconds are coming from instrumentation-http, just to make super my current setup is not being broke.

Hower, @zeitlinger, I was now able to replicate you're setup, but unfortunately, still no traces are being connected with metrics. Even when I did turned out examplars.