Bad API Observability
summary
Announces Ty adding open Telemetry distributed tracing to its API Gateway after 10–15 customer sessions and shows concrete tooling: OpenTelemetry collector, span metric connector, Jaeger and Prometheus. Calls out seven anti-patterns like relying on external API monitors alone and using access logs instead of distributed traces; demonstrates surprises such as an API monitor reporting green while traces show 74 requests 41 errors. Recommends starting traces at the gateway, capturing client-side errors (4xx and GraphQL payload errors), and routing observability via pipelines per team.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:05Why API observability matters
- 2:53Reasons APIs need observability
- 5:21OpenTelemetry observability stack
- 7:11Anti-pattern: API monitoring only
- 9:20Anti-pattern: Forgetting API users
- 12:09Anti-pattern: Client-side errors ignored
- 12:31Anti-pattern: Relying on API logs
- 14:19Distributed tracing early
- 15:02Anti-pattern: Disconnected teams
- 17:33OpenTelemetry pipelines
- 18:38Anti-pattern: One-size monitoring
- 19:42Observability for GraphQL
- 22:34Anti-pattern: Production-only observability
- 23:57Traces for testing and debugging
- 25:00Key takeaways
- 25:23Q&A
related talks
43mPractical Introduction to OpenTelemetry Tracing
08mGraphQL Observability
52mGraphQL Observability with Elastic and OpenTelemetry
32mAntipatterns in Observability: Lessons Learned and How OpenTelemetry Solves Them
33mThe Observant Developer - OpenTelemetry from code to prod
38mPractical introduction to OpenTelemetry tracing
22mOpen Telemetry for API Monitoring
33mWhat Developers Should Know About Observability