Quickstart
Instrument your entire Kubernetes cluster in 5 minutes with zero code changes.
Quickstart
Go from zero to full cluster observability in 5 minutes.
Prerequisites
- A running Kubernetes cluster (any provider)
kubectlconfigured and pointing to the cluster- An Obtrace API key (
obt_live_...)
Step 1: Install the CLI
Verify:
Step 2: Discover what would be instrumented
Before installing anything, see what the operator would do:
This is a dry run — nothing is installed yet.
Step 3: Install the operator
The CLI will:
- Check cluster connectivity
- Install the operator via Helm (falls back to
kubectl applyif Helm is not available) - Create the
obtrace-systemnamespace - Deploy the webhook, RBAC, and default ObtraceInstrumentation
Output:
Step 4: Trigger a rollout
Existing Pods are not modified — instrumentation is applied on Pod creation. Trigger a rollout restart to instrument existing workloads:
New Pods will be created with instrumentation automatically injected.
Step 5: Verify
You can also check individual Pods:
Or inspect annotations on a specific Pod:
Step 6: See telemetry in the Obtrace dashboard
Open the Obtrace frontend. Within seconds of the rollout completing, you will see:
- Traces — HTTP server spans for every inbound request, with method, route, status code, and duration
- Logs — Errors, warnings, and uncaught exceptions with stack traces, correlated to the trace that caused them
- Metrics — Memory usage, CPU time, GC pressure, and HTTP latency histograms per service
All of this without touching application code.
Scoping to specific namespaces
If you prefer to instrument only specific namespaces:
Or instrument one namespace at a time:
Choosing a strategy
By default, Obtrace Zero uses auto — SDK for interpreted languages, eBPF for compiled. You can force a specific strategy:
See Strategies for details on when to use each.
Next steps
- How it works — understand the mutation and injection flow
- Language agents — what each agent captures
- CRD reference — advanced configuration via GitOps
- Troubleshooting — common issues and fixes