#Export OpenTelemetry traces somewhere else

1 messages · Page 1 of 1 (latest)

vague cedar
#

Is it possible to export OpenTelemetry traces elsewhere, in addition to Dagger Traces?

lament elk
vague cedar
#

But will it still send to Dagger Tracing too?

lament elk
#

if you add an exporter and you're logged it to Dagger Cloud or using a DAGGER_CLOUD_TOKEN, we'll send the telemetry both to our and your system

vague cedar
lament elk
#

@vague cedar I wasn't able to repro this. The AI-assisted issue above is also incorrect given that the telemetry gets sent from the client and not from the engine. This is the code that takes care of propagating the env variables https://github.com/dagger/dagger/blob/7c83409eee972e0e5eacb71427f3553158d33dfa/cmd/dagger/run.go?plain=1#L136.

The way that I verified this works is by doing the following:

  1. Start a random dev http server:

python -m http.server 8080

  1. Run dagger and point the metrics endpoint to that server

OTEL_EXPORTER_OTLP_METRICS_ENDPOINT=http://localhost:8080/v1/metrics dagger -c 'container | from alpine | with-exec "echo, hello"

You'll see in the server access logs that Dagger effectively tries to post the metrics there:

127.0.0.1 - - [18/Feb/2026 10:05:47] code 404, message File not found
127.0.0.1 - - [18/Feb/2026 10:05:47] "GET /v1/metrics HTTP/1.1" 404 -
127.0.0.1 - - [18/Feb/2026 10:05:54] code 501, message Unsupported method ('POST')
127.0.0.1 - - [18/Feb/2026 10:05:54] "POST /v1/metrics HTTP/1.1" 501 -
127.0.0.1 - - [18/Feb/2026 10:05:55] code 501, message Unsupported method ('POST')
127.0.0.1 - - [18/Feb/2026 10:05:55] "POST /v1/metrics HTTP/1.1" 501 -
vague cedar
#

Ah I was trying to send Otel from the server. 😅

#

The docker container does log many spans I think so it’s aware of various metrics would be good to be able to get them. For now I’m parsing logs which works but ugly of course.

vague cedar
#

Is there anything different about running from a GitHub action? I just can't figure it out... It works from local, it works from manually sending traces via curl from CI. But when I run Dagger in CI (via GitHub action), no traces show up.

lament elk
vague cedar
# lament elk Nope. There shouldn't be any difference? Mind sharing a snippet about how you're... 
      - name: Run Dagger CI
        env:
          OUR_APP_GITHUB_TOKEN: ${{ steps.app-token.outputs.token }}
        uses: dagger/dagger-for-github@662d9b66af7c40b8745fb3131785a53129193ffe
        with:
          call: ci-all ... many args ...
          dagger-flags: --progress=dots
          cloud-token: ${{ secrets.DAGGER_CLOUD_TOKEN }}
          version: ${{ env.DAGGER_VERSION }}
          summary-path: "/tmp/dagger-summary.txt"
          enable-github-summary: "true"
          workdir: /codebuild/output/src/foo
lament elk
#

let me do a quick test

vague cedar
#

I am setting the OTEL vars on the runner env, which gets passed down to the action, I am pretty sure, as other values did.

#

In your example, in the dagger.yaml file you are not even setting the traces env, how does it work?

#

Or was it just an example and you already took it out? I see on the run it's set:

screenshot-20260219T213300-Evs5Tlqw2x.png
#

Also, in your examples you've set metrics endpoint. I actually got metrics to work. I am looking for traces tho.

#

Metrics are too basic and don't give anything beyond I can already pretty much collect from the runner host.

#

I am looking for spans to be able to see bottlenecks in the pipeline. I want to see which tasks and why take too long and then optimize them.

lament elk
lament elk
vague cedar
#

I'm setting:

OTEL_EXPORTER_OTLP_TRACES_ENDPOINT=http://.../v1/traces
OTEL_EXPORTER_OTLP_TRACES_PROTOCOL=http/protobuf

Seems to be correct?