OpenTelemetry tracing for Node HTTP

Once services talk to other services, logs alone don’t tell you where time goes. I like OpenTelemetry because it’s vendor-neutral: export to Honeycomb, Datadog, Tempo—whatever your team uses. I keep the initial setup minimal: instrument http, express, and pg, set a service name, and start collecting traces. Then I add custom spans only where they add real value (like a slow third-party API call). The biggest practical lesson is context propagation: if trace context doesn’t flow through async boundaries and background jobs, you get broken traces that are hard to reason about. With good traces, latency investigations become much more deterministic.