Configuration and naming

Default runtime

You can use @Metrics, measure, and instrumentFunction without calling configureMetrics. The default runtime stores the latest 1,000 metric records and 1,000 log records in separate bounded buffers.

import { metrics } from '@haskou/metrics';

const snapshot = metrics.snapshot();
console.table(snapshot.metrics);
console.table(snapshot.logs);

metrics.clear();

Each buffer replaces its oldest record when full and increments its own discarded count. metrics.snapshot() only inspects these default buffers. If you configure an external adapter, inspect that backend instead.

Process-wide configuration

configureMetrics replaces the process-wide instrumenter and returns an idempotent restore function:

const restore = configureMetrics({ adapter, logger });

// Instrumented work uses adapter and logger.

restore();

Call it from the composition root before instrumented work starts. For isolated components or concurrent configurations, construct MetricsInstrumenter directly and inject it instead of changing process-wide state.

MetricsConfiguration

Property	Type	Default
adapter	MetricsPort	Required
attributes	MetricAttributes	{}
clock	ClockPort	Monotonic system clock
logger	LoggerPort	No-op after explicit configure
nameFormatter	(operation, kind) => string	operation.kind
onInstrumentationError	(error: unknown) => void	No callback
prefix	string	No prefix
resourceUsage	ResourceUsagePort	Node.js process resource usage

Shared attributes merge with method-level attributes. Method-level values win when both objects contain the same key.

Instrumentation catches errors from ports, clocks, resource sampling, and onInstrumentationError. Telemetry failures do not change the return value or error behavior of the instrumented operation.

Instrumentation options

Metrics, measure, and instrumentFunction accept InstrumentationOptions:

Option	Default	Effect
attributes	{}	Adds attributes to this operation
recordCalls	true	Increments the call counter
recordDuration	true	Observes elapsed milliseconds
recordFailures	true	Increments the failure counter
logCalls	true	Writes a structured call log
logFailures	true	Writes a structured failure log
captureStackTrace	true	Adds a stack trace to failure logs
recordCpu	false	Observes process CPU deltas
recordMemory	false	Observes process RSS, heap, and their deltas

Log switches still write nothing when the configured logger is a no-op.

Decorator names

@Metrics() infers ClassName.methodName:

class UserCreator {
  @Metrics()
  public create(): void {}
}

The operation name is UserCreator.create. Pass undefined to combine inference with options:

@Metrics(undefined, { recordCpu: true })

Use an explicit name for dashboards, alerts, and recording rules that must survive refactors or minification:

@Metrics('users.create')

Standard and legacy behavior

• Standard decorators resolve the class from the method receiver when the method runs. Calling an inherited method on a subclass uses the subclass name.
• Legacy decorators resolve the declaring class when TypeScript applies the decorator. An inherited method keeps the declaring class name.
• Static methods resolve the constructor name in both modes.
• A standard-decorated method called without a receiver falls back to Anonymous.methodName.
• Bundlers, minifiers, and class renames can change inferred names.

Explicit operation names avoid these differences.

Function instrumentation

Functions have no reliable class name, so measure and instrumentFunction require an explicit operation name:

const result = await measure('users.create', () => creator.create(command));
const measuredCreate = instrumentFunction('users.create', createUser);