Chasing the Rainbow: Towards Unified Service Metrics
summary
Zendesk moves from a monolith to ~1–2k codebases and hundreds of services and adopts Kubernetes plus a service mesh to unify observability. Centralizing on Datadog APM plus Istio + Envoy sidecars delivers consistent instrumentation across languages, emits source-side vs destination-side metrics (double-reported latencies and errors so upstream “no upstream instances” and proxy-enforced timeouts surface), and reveals retries, circuit-breaker and AZ-specific failures. The claim: mesh-driven, centrally configured metrics replace ad-hoc libs and dashboards and make cross-service failures measurable. Datadog APM semantic metrics
AI-written summary · about
chapters (9)
- 0:00Introduction and evolution
- 2:20From monolith to services
- 7:16Early observability journey
- 14:25Adopting APM
- 17:17Challenges in the cloud
- 20:54Implementing service mesh
- 26:55Metrics and dashboards
- 30:04Future of observability
- 34:17Q&A
related talks
33mEffective observability in microservice architectures
14mMonitoring Network Traffic in Kubernetes
46mObservability in the real world: Improve systems w/iterative approach
24mObservability for Istio Service Mesh
50mIntuit Journey To Unified Observability at Scale: Challenges, Be...
31mHow the Metrics Backend Works at Datadog
30mReliable Observability at Scale: Error Budgets for 1,000+
16mMonitoring Microservices The Right Way