Your OpenTelemetry fleet, from your Claude console

> Checkout spans have no service.version or http.route, so the latency dashboard lumps every checkout request together. Can you fix it at the collector?

Checkout spans pass through the prod-us-east node agents, which set the k8s attributes, so k8s-node-agent is the config to change.

$ telflo fleet deployments e9c47d12 --json: 4 deployments · k8s-node-agent 5d1b0c3a · config a41d7e2b · v10 · 42 nodes

$ telflo config pull a41d7e2b -o node-agent.yaml: wrote v10 to node-agent.yaml · pinned to otelcol-contrib 0.147.0

Read node-agent.yaml: traces pipeline is otlp → k8sattributes → batch → loadbalancing/gateway

kubectl get pods -n shop -l app=checkout --show-labels: 2 pods, both labelled app.kubernetes.io/version=2.14.3

k8sattributes extracts three k8s.* fields and no service.version, a one-line metadata field it derives from app.kubernetes.io/version.

Read 240 checkout server spans from the last 15 min via Grafana MCP: 0 of 240 carry http.route, names like POST /api/checkout/8841

The route is only in the span name. No transform in the traces pipeline copies it, so the dashboard groups on an empty http.route.

Two edits to node-agent.yaml: service.version under k8sattributes metadata, and http.route from the span name where it is missing.

node-agent.yaml · k8sattributes: processors: k8sattributes: extract: metadata: - k8s.namespace.name - k8s.pod.name - k8s.deployment.name +- service.version

node-agent.yaml · transform/route: +transform/route: +trace_statements: +- context: span +statements: +- set(attributes["http.route"], name) where attributes["http.route"] == nil and kind == SPAN_KIND_SERVER +- replace_pattern(attributes["http.route"], "^[A-Z]+ ", "") where kind == SPAN_KIND_SERVER and IsMatch(attributes["http.route"], "^[A-Z]+ ") +- replace_pattern(attributes["http.route"], "/\d+", "/{id}") where kind == SPAN_KIND_SERVER service: pipelines: traces: -processors: [k8sattributes, batch] +processors: [k8sattributes, transform/route, batch]

$ telflo ottl lint node-agent.yaml: invalid · processors.transform/route.trace_statements[0].statements[2] · 2 of 3 expressions parse · exit 4

Statement 3: OTTL rejects \d as an escape inside a string literal. Changed the pattern to /[0-9]+.

$ telflo ottl lint node-agent.yaml: 3 of 3 expressions parse · 420 ms

$ telflo validate node-agent.yaml --config a41d7e2b: valid · otelcol-contrib 0.147.0 · 2.1 s

Wrote route.test.yaml: 12 server spans, 8 checkout with no http.route, 4 cart that already carry /api/cart as the untouched control.

$ telflo test --spec route.test.yaml -f node-agent.yaml --config a41d7e2b: Success · run c6d2f8a1 · 4 of 4 rules passed · 12.6 s

pass: total count, 12 of 12 spans survived. pass: http.route on every span, all · attributes["http.route"] != nil · 12 of 12. pass: ids collapsed to {id}, 8 of 8 read /api/checkout/{id}. pass: untouched control, cart /api/cart, 4 of 4 unchanged

$ telflo config push a41d7e2b -f node-agent.yaml --publish --description "service.version from pod label, http.route from span name": appended v11 · published

$ telflo deploy --fleet e9c47d12 --config a41d7e2b: updated · deployment 5d1b0c3a now serves v11

prod-us-east · k8s-node-agent · v11: 42 of 42. 42 node agents pick up v11 · 38 s

Checkout spans from prod-us-east now carry service.version from the pod label and http.route with ids collapsed to {id}.

v10 is still published. telflo deploy --fleet e9c47d12 --config a41d7e2b --version 10 puts the fleet back on it.

> What do the attributes on a checkout span look like now?

Read 20 checkout server spans via Grafana MCP, 10 from 11:38 and 10 from 11:44. Showing one from 11:44.

Span attributes · checkout · POST /api/checkout/8841 · 212 ms

resource ├─ service.name checkout ├─ service.version new 2.14.3 ├─ k8s.namespace.name shop ├─ k8s.deployment.name checkout └─ k8s.pod.name checkout-6b9d8f7c4-k4xw2 span POST /api/checkout/8841 212 ms SERVER ├─ http.request.method POST ├─ http.route new /api/checkout/{id} ├─ http.response.status_code 200 └─ url.path /api/checkout/8841

10 of 10 spans from 11:44 carry service.version and http.route. 0 of 10 from 11:38, before the rollout, had either.

The latency dashboard can group checkout by http.route now, and p99 can split by service.version from the next release.