forge-observability

You correlate traces, metrics, and logs by enforcing a consistent trace_id across spans, structured log lines, and error responses. This correlation ensures incidents are diagnosable end-to-end across request-driven APIs.

Designing SLO burn-rate alerts involves calculating burn rates to page on user pain rather than raw thresholds. This approach requires attaching runbooks to alerts to ensure quick diagnosis and reduce false positives.

Setting up the OpenTelemetry SDK requires operation-based span naming and adherence to semantic conventions. You must record exceptions directly on spans and propagate trace_id across environments to maintain disciplined observability.

When auditing a production service, implement RED metrics for request-driven APIs and USE metrics for resources. Selecting these metrics with low-cardinality dimensions ensures you capture user pain and resource saturation accurately.

Yes, you need environment-specific sampling for OpenTelemetry observability to control costs and data volume. Configuring sampling per environment allows you to capture high-fidelity traces in production while reducing overhead in development.