Instrumentation

Definition

Instrumentation is the code, runtime configuration, or platform setup that makes software emit telemetry. It is where observability actually begins: everything downstream, from dashboards to alerts to AI assisted debugging, can only work with what instrumentation captured in the first place.

What it means in observability

The quality ceiling of an observability practice is set at instrumentation time. No query can recover an attribute that was never recorded, and no backend can reconstruct a span that was never created. The good news is that the effort curve is friendly: auto instrumentation now covers the standard surface area of most applications, HTTP servers and clients, databases, queues, and frameworks, so the marginal effort goes into the part only you can write, the business level signals that make telemetry speak your domain language.

How it works in practice

Instrumentation arrives in three layers. Auto instrumentation attaches at the runtime or framework level and produces spans, metrics, and logs with no code changes. Instrumentation libraries ship inside or alongside popular packages, covering specific clients and frameworks. Manual instrumentation uses the OpenTelemetry API directly to create custom spans, record business metrics, and attach the attributes that matter to your domain. Beneath all three, resource attributes identify the emitting service, version, and environment so every signal lands with context.

Where it gets hard

Coverage gaps are the silent failure mode: the un-instrumented service in the middle of a call chain, the background job nobody traced, the language whose support lags the rest of the estate. Over-instrumentation is the opposite trap, generating volume that costs money without answering questions. And custom attributes need governance, because without naming discipline the estate drifts into a private dialect that only its authors can query.

Where Tsuga fits

Tsuga runs on stock OpenTelemetry instrumentation with nothing proprietary to add. Because ingestion is OTLP native, instrumentation investment stays portable, and Tsuga's quality reports help spot the services and signals where coverage falls short.