OpenTelemetry Tracing for a Critical Route

Add useful traces to one critical path. The goal is a trace a human can read during an incident, not a span on every function. APP Runtime: [NODE ON NEXT.JS | NODE ON A SERVER | PYTHON] Route or job: [NAME AND WHAT IT DOES] Downstream calls: [DB, HTTP APIS, QUEUE, CACHE] Collector: [HONEYCOMB | GRAFANA TEMPO | DATADOG | OTLP GENERIC] Symptoms we are trying to explain: [SLOW | ERROR SPIKES | OCCASIONAL TIMEOUTS] Cardinality traps in our domain: [USER IDS, EMAILS, RAW SQL] DELIVER 1) SDK setup: which packages, where to init so it runs before other imports, and the resource attributes (service.name, deployment.environment). 2) The span tree for this route only. Name each span as verb + object ("load subscription", "charge invoice"). Mark which are children vs links. 3) Attributes worth attaching, and a denylist (emails, tokens, raw SQL text, full request bodies). Use ids, status codes, cache hit, and row counts. 4) How errors land on the span (recordException, status code) without double-reporting to the error tracker. 5) Sampling: a rule that keeps errors and slow traces and samples the healthy majority, with numbers I can edit. 6) A worked example of one trace timeline in text (span, duration, attribute) for the slow case I described. 7) The one dashboard query: p95 of this route and error rate, grouped by a low-cardinality attribute. Do not wrap every internal helper in a span. If a span would always be under 1ms and has no attributes, leave it out and say so.