OpenTelemetry Semantic Conventions and How to Avoid Broken Observability
summary
Semantic changes to OpenTelemetry (e.g. deployment.environment → deployment.environment.name, HTTP.method → http.request.method) silently break pipelines and caused customers to lose red metrics; about 30% of attendees run the collector in production. The community builds a checker and generator called Weaver (resolve/check/generate/diff, OTLP listener for CI) and a runtime Schema Processor that rewrites incoming OTLP to a target schema (example: 1.27 → 1.26) so dashboards keep working. Limitations: schema URL is optional so some payloads are ignored, and merges/splits still require manual handling.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:21Broken observability in production
- 2:46OpenTelemetry overview
- 4:55Semantic conventions
- 6:22Semantic Convention Registry
- 8:31Observability issues in production
- 10:39Tooling for breaking changes
- 13:08Weaver commands
- 15:14Registry policies and extensibility
- 17:22Policy enforcement with Rego
- 18:45Generating docs and SDKs
- 19:25Registry diff for versions
- 20:51Collector pipelines
- 23:00Semantic changes breaking dashboards
- 24:45Schema processor
- 27:14Schema processor limits and future
- 28:37Next steps and community
- 31:31Conclusion and Q&A
related talks
38mThe Hater's Guide To OpenTelemetry
18mOpenTelemetry and Observability: What, Why, and Why Now?
36mObservability For You and Me with OpenTelemetry
04mOpen Telemetry Project Updates
20mObservability and Open Telemetry
50mOpen Telemetry: The Rise of Open Standards in Observability
53mOpenTelemetry Metrics: observability’s holy grail
28mOpenTelemetry developments in Prometheus over the last year