Linkerd Jaeger Tracing w/ Open Telemtry Collector

Linkerd General Discussion
1

We are using the grafana-agent to handle incoming traces, potentially do things to them, and then forward them on to Grafana Tempo. The Grafana agent is basically running the Open Telemtry Collector under the hood.

This setup is working for all of our applications, but linkerd is not sending traces and I assume its some misconfiguration on my end.

There is a service, listening for OTLP HTTP and GRPC traffic

kubectl get svc -n grafana-agent
NAME            TYPE        CLUSTER-IP     EXTERNAL-IP   PORT(S)                    AGE
grafana-agent   ClusterIP   172.20.18.93   <none>        80/TCP,4317/TCP,4318/TCP   27h

The configuration of this OTLP collector, ends in sending the traces to Tempo via GRPC

    otelcol.exporter.otlp "to_tempo" {
      client {
        endpoint = "tempo-distributed-distributor.tempo.svc.cluster.local:4317"
        tls {
            insecure             = true
            insecure_skip_verify = true
        }
      }
    }

This is working for our applications, I can view traces in Grafana etc.

nginx-ingress for example, it defaults otlp-collector-port to 4317

    enable-opentelemetry: "true"
    otlp-collector-host: "grafana-agent.grafana-agent.svc.cluster.local"

I have linkerd-jaeger deployed via the helm chart, like this, ive tried both 4318 and 4317, neither appear to work.

collector:
  enabled: false
jaeger:
  enabled: false

webhook:
  externalSecret: true
  injectCaFrom: linkerd-jaeger/jaeger-injector
  collectorSvcAddr: grafana-agent.grafana-agent:4317
  collectorSvcAccount: grafana-agent

I can see this is configured on my linkerd-proxy

      LINKERD2_PROXY_TRACE_COLLECTOR_SVC_ADDR:                  grafana-agent.grafana-agent:4317
      LINKERD2_PROXY_TRACE_COLLECTOR_SVC_NAME:                  grafana-agent.grafana-agent.serviceaccount.identity.linkerd.cluster.local

I see errors on the linkerd-proxy that would be sending a trace

{"timestamp":"[  1351.438936s]","level":"WARN","fields":{"message":"Service failed","error":"endpoint 172.20.18.93:4317: channel closed"},"target":"linkerd_reconnect","spans":[{"name":"opencensus"},{"addr":"grafana-agent.grafana-agent.svc.cluster.local:4317","name":"controller"},{"addr":"172.20.18.93:4317","name":"endpoint"}],"threadId":"ThreadId(1)"}
{"timestamp":"[  1394.200939s]","level":"WARN","fields":{"message":"Service failed","error":"endpoint 172.20.18.93:4317: channel closed"},"target":"linkerd_reconnect","spans":[{"name":"opencensus"},{"addr":"grafana-agent.grafana-agent.svc.cluster.local:4317","name":"controller"},{"addr":"172.20.18.93:4317","name":"endpoint"}],"threadId":"ThreadId(1)"}
{"timestamp":"[  1416.017672s]","level":"WARN","fields":{"message":"Service failed","error":"endpoint 172.20.18.93:4317: channel closed"},"target":"linkerd_reconnect","spans":[{"name":"opencensus"},{"addr":"grafana-agent.grafana-agent.svc.cluster.local:4317","name":"controller"},{"addr":"172.20.18.93:4317","name":"endpoint"}],"threadId":"ThreadId(1)"}
{"timestamp":"[  1502.128800s]","level":"WARN","fields":{"message":"Service failed","error":"endpoint 172.20.18.93:4317: channel closed"},"target":"linkerd_reconnect","spans":[{"name":"opencensus"},{"addr":"grafana-agent.grafana-agent.svc.cluster.local:4317","name":"controller"},{"addr":"172.20.18.93:4317","name":"endpoint"}],"threadId":"ThreadId(1)"}
2

AFAIK Linkerd supports only the Opencensus protocol for span collection.
I don’t know how it looks on the Grafana Agent, but I use OpenTelemetry Collector and I push the Linkerd spans to port 55678 on the collector.

The receiver config for this is:

receivers:
    opencensus: {}
3

Linkerd now natively supports the OpenTelemetry protocol! This was added in Linkerd edge-24.10.1, so it’s also in Linkerd 2.17 – check out Distributed tracing with Linkerd | Linkerd for details, but that should help you here, I’d think.