opentelemetry

SKILL.md

OpenTelemetry with Grafana

Docs: https://grafana.com/docs/opentelemetry/

Vendor-neutral instrumentation pipeline. Apps speak OTLP → Alloy (or direct) → Grafana Cloud (Mimir / Loki / Tempo / Pyroscope).

Backends

Signal
Backend

Metrics
Grafana Mimir

Logs
Grafana Loki

Traces
Grafana Tempo

Profiles
Grafana Pyroscope

Prerequisites

Grafana Cloud stack OR self-hosted Mimir / Loki / Tempo

Cloud OTLP endpoint: https://otlp-gateway-<region>.grafana.net/otlp

Basic-auth credentials: numeric instance ID + API token with MetricsPublisher + LogsPublisher + TracesPublisher

An app to instrument

Common Workflows

1. Authenticate to the Grafana Cloud OTLP endpoint

# 1. Build the auth header
INSTANCE_ID=123456
API_KEY="glc_eyJ..."
export OTEL_EXPORTER_OTLP_ENDPOINT=https://otlp-gateway-prod-us-east-0.grafana.net/otlp
export OTEL_EXPORTER_OTLP_PROTOCOL=http/protobuf
export OTEL_EXPORTER_OTLP_HEADERS="Authorization=Basic $(echo -n "${INSTANCE_ID}:${API_KEY}" | base64)"
export OTEL_RESOURCE_ATTRIBUTES="service.name=myapp,service.namespace=myteam,deployment.environment=prod"

# 2. Smoke-test creds with a curl POST against the OTLP traces endpoint (empty body)
curl -s -o /dev/null -w "%{http_code}\n" \
  -X POST -H "Content-Type: application/x-protobuf" \
  -H "Authorization: Basic $(echo -n "${INSTANCE_ID}:${API_KEY}" | base64)" \
  "$OTEL_EXPORTER_OTLP_ENDPOINT/v1/traces" --data-binary '\n'
# Expect 400 (malformed payload) — NOT 401 (auth) or 404 (wrong endpoint).

2. Auto-instrument a Java app + verify

# 1. Download the Grafana JVM agent (single jar)
curl -sLO https://github.com/grafana/grafana-opentelemetry-java/releases/latest/download/grafana-opentelemetry-java.jar

# 2. Run with the agent + env from step 1
java -javaagent:./grafana-opentelemetry-java.jar -jar myapp.jar

# 3. Generate traffic, then verify in Grafana → Explore → Tempo:
#    TraceQL: { resource.service.name = "myapp" }
#    Expect spans within ~30s. Also verify metrics:
#    PromQL: count by (service_name)({service_name="myapp"})

3. Auto-instrument a Python app

pip install "opentelemetry-distro[otlp]"
opentelemetry-bootstrap -a install

# Same env vars as step 1, then:
opentelemetry-instrument python app.py

# Verify the same way — Explore → Traces filter service.name=myapp.

4. Add Alloy as a buffering / sampling collector

# Application points at local Alloy (gRPC fastest)
export OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4317
export OTEL_EXPORTER_OTLP_PROTOCOL=grpc

# Alloy environment for forwarding to Cloud
export GRAFANA_CLOUD_OTLP_ENDPOINT=https://otlp-gateway-prod-us-east-0.grafana.net/otlp
export GRAFANA_CLOUD_INSTANCE_ID=$INSTANCE_ID
export GRAFANA_CLOUD_API_KEY=$API_KEY
alloy run /etc/alloy/config.alloy

# Verify Alloy received and forwarded
curl -s http://localhost:12345/metrics | grep otelcol_exporter_sent_spans

Full Alloy config + tail-sampling block + OTel Collector YAML + K8s Operator install: references/collector-config.md.

SDK-by-language details (Go full code, Node manual setup, .NET ASP.NET Core, all the env-var quirks): references/instrumentation.md.

5. Kubernetes — auto-inject via the Operator

apiVersion: opentelemetry.io/v1alpha1
kind: Instrumentation
metadata: { name: my-instrumentation }
spec:
  exporter: { endpoint: http://otelcol:4317 }
  propagators: [tracecontext, baggage]
  java:
    image: us-docker.pkg.dev/grafanalabs-global/docker-grafana-opentelemetry-java-prod/grafana-opentelemetry-java:2.3.0-beta.1
  nodejs: {}
  python: {}

Then annotate pods:

metadata:
  annotations:
    instrumentation.opentelemetry.io/inject-java: "true"
    # or: inject-nodejs, inject-python, inject-dotnet

# Verify the operator injected the agent
kubectl describe pod <pod> | grep -A2 'opentelemetry-auto-instrumentation'
# Then run the same Grafana Explore checks.

Sampling — when to pick which

# Head sampling (cheap, decided at start; may lose rare errors)
export OTEL_TRACES_SAMPLER=parentbased_traceidratio
export OTEL_TRACES_SAMPLER_ARG=0.1   # 10%

Tail sampling (decides after seeing the whole trace — keep errors + sample the rest) requires an Alloy / OTel-Collector tail_sampling processor; full block in references/collector-config.md.

Key environment variables

Variable
Example

OTEL_EXPORTER_OTLP_ENDPOINT
https://otlp-gateway-prod-us-east-0.grafana.net/otlp

OTEL_EXPORTER_OTLP_PROTOCOL
grpc or http/protobuf

OTEL_EXPORTER_OTLP_HEADERS
Authorization=Basic <base64>

OTEL_RESOURCE_ATTRIBUTES
service.name=app,service.namespace=team,deployment.environment=prod

OTEL_SERVICE_NAME
shorthand for service.name

OTEL_TRACES_SAMPLER / _ARG
parentbased_traceidratio / 0.1

Troubleshooting

401 from OTLP gateway → instance ID is not numeric, or API key missing publisher roles

404 → endpoint URL wrong (must end with /otlp)

Spans missing → check OTEL_EXPORTER_OTLP_PROTOCOL matches transport (Cloud OTLP gateway = http/protobuf, Alloy local = grpc)

Node.js auto-instrumentation broken after bundling → bundlers like @vercel/ncc defeat the require hooks

Python under Gunicorn / uWSGI shows no spans → reinit OTel providers in a post-fork hook

Resources

Grafana OTel docs

Grafana Cloud OTLP

Grafana JVM agent

Grafana .NET SDK

OTel Operator