Controller-Native Trace Propagation for Delivery Analytics
summary
Implements controller-native trace propagation on a Tekton/Conflux delivery platform: controllers inject W3C trace context into Kubernetes resource annotations (using the Tekton Snapshot anchor for cross-stage lifetime), downstream reconcilers extract it to create child spans, and OpenTelemetry renders one trace per delivery. Rules enforce creator-writes reconciler-reads, a carrier-format contract, separate dedupe keys, and end-to-end tests. Result: build/test/release appear in one trace with wait vs execution spans, enabling accurate mean-time-to-build without a custom correlation store.
AI-written summary · about
chapters (19)
- 0:00Controller-native tracing intro
- 1:04Request propagation failures
- 3:33Resource-carried trace context
- 5:39Tekton and Conflux case studies
- 8:48W3C and OpenTelemetry standards
- 11:17Tekton’s tracing support
- 13:05Choosing trace context resource
- 14:54Trace context rules
- 18:00Stable trace coordination
- 21:53Writer ownership prevents corruption
- 22:57Trace continuity and lifetime
- 24:25Deduplication vs propagation
- 25:52Manual retries handling
- 27:16Reliable tracing operations
- 29:02Testing trace correctness
- 29:43Mean time to build
- 33:20Wait and execution spans
- 35:05Delivery latency analysis value
- 36:28Key takeaways
related talks
22mBuilding a Kubernetes Operator for Automated Distributed Tracing
55mCloud-Native Observability with OpenTelemetry
34mReduce Complexity with Distributing Tracing
37mThrough the looking glass: Effective observability for cloud native applications
54mA TRACE TO REMEMBER: COMPOSITIONAL TRACING
32mDistributed Tracing in Action: Our Journey with OpenTelemetry
16mProgressive Delivery with Observability
33mOpenTracing and Containers: Depth, Breadth, and the Future of Tracing