Grafana

Deploys a complete trace sink into a local Kind cluster: an OpenTelemetry Collector that accepts OTLP on host ports 30317 (gRPC) and 30318 (HTTP), Grafana Tempo storing the traces it receives, and Grafana on host port 30300 with the Tempo datasource already provisioned.

Point any OpenTelemetry-instrumented application at the collector and the traces become searchable in Grafana within seconds – no datasource setup, no login.

To send traces from a Gravitee stack running in the same cluster, compose grafana/base onto an APIM or AM context instead – see Composing with a Gravitee stack below.

Install gck#

go install github.com/gravitee-io-labs/gck@latest

For other installation methods, see Installation .

Grafana is reachable on http://localhost:30300 with no extra setup. If you would rather serve it on a hostname, pass --enable-route (see Serving Grafana on a hostname ); that turns on local DNS, which needs a one-time OS setup (may require sudo):

gck setup dns

See the Networking guide for details.

Usage#

Create#

gck create --from grafana/standalone

Cleanup#

gck delete

Quick Start#

Open Grafana and go to Explore – the Tempo datasource is selected by default, so Search lists every trace received so far:

open http://localhost:30300

Anonymous access is enabled with the Admin role, so no login is required. The built-in admin / admin account still works if you want a named user.

Send a trace to confirm the pipeline end to end:

curl -X POST http://localhost:30318/v1/traces \
  -H 'Content-Type: application/json' \
  -d '{"resourceSpans":[{"resource":{"attributes":[{"key":"service.name","value":{"stringValue":"demo"}}]},"scopeSpans":[{"spans":[{"traceId":"5b8efff798038103d269b633813fc60c","spanId":"eee19b7ec3c1b174","name":"GET /demo","kind":2,"startTimeUnixNano":"1700000000000000000","endTimeUnixNano":"1700000000050000000"}]}]}]}'

Then search for the demo service in Explore. Traces are also printed to the collector’s own logs, which is the quickest way to tell whether something is arriving at all:

kubectl logs -f deploy/otel-collector

Traces are held in an emptyDir, so they are lost when the pod restarts and compacted blocks are dropped after an hour. That is deliberate for local development – raise blockRetention if you need a longer window.

Serving Grafana on a hostname#

http://localhost:30300 works everywhere and needs nothing set up, which makes it the right default. For a demo you may prefer a real hostname. Pass --enable-route to serve Grafana through the Kubernetes Gateway API with a local DNS record instead:

gck create --from grafana/standalone --enable-route

Grafana is then available at http://grafana.gck.local, alongside the NodePort. Override the hostname with --set hostname=grafana.demo.gck.local.

The flag turns on gck’s gateway and dns features, which brings up a cloud-provider-kind load balancer and the local DNS server. On macOS the load balancer needs a packet tunnel, so gck create prompts for your password, and gck setup dns must have been run once beforehand.

Composing with a Gravitee stack#

grafana/base is the same layer without a cluster of its own: the collector, Tempo and Grafana are deployed into an observability namespace, leaving the product namespace to the product. Compose it onto any APIM or AM context and pass --enable-otel-collector to point the gateway’s exporter at the collector:

gck create \
  --from gravitee-io/oss/apim/jdbc/postgres \
  --from grafana/base \
  --enable-otel-collector

Grafana comes up on http://localhost:30300 in that cluster too – the layer brings its own host port mapping, so it works alongside the APIM ports without any port-forward. Add --enable-route for http://grafana.gck.local.

Call an API through the gateway, then search Tempo for the gio-apim-gateway service in Explore.

The gateway only emits spans for APIs that have tracing switched on. Enabling it on the gateway is not enough: for each v4 API, open Reporter Settings in the console and turn on tracing (analytics.tracing.enabled), then redeploy the API. Until you do, the pipeline is healthy but carries nothing. Requests that match no API — a bare curl against the gateway returning 404 — never produce a span either.

Compose grafana/base, not grafana/standalone. The standalone variant declares its own Kind cluster and would rename the cluster the Gravitee context created. The abstract base declares no cluster of its own, so it layers cleanly on top.