Jaeger Deep Dive
Talkat KubeCon + CloudNativeCon North America 2020 Virtual · on Thursday, 2020-11-19· at 14:55· Intrado Virtual Event Platform,·
summary
Walkthrough of Jaeger architecture, sampling and OpenTelemetry integration with concrete choices: span size ~2 KB and 10k QPS across 100 instances gives 170 petabytes per day, motivating sampling; head‑based (probabilistic, rate‑limit) vs tail‑based sampling tradeoffs and observed instrumentation overheads (~5% on high‑QPS services); Jaeger defaults to a remote sampler pulled from collectors; Kafka decouples ingestion and ingester; backend rebased on OpenTelemetry Collector but keeps storage plugins (Cassandra, Elasticsearch, Badger); Kubernetes deploys via Helm, YAML or operator.
AI-written summary · about
chapters (22)
- 0:00Intro
- 2:30Metrics - Cardinality
- 3:27Context Propagation & Distributed Tracing
- 4:08Service dependencies diagram
- 5:02Transitive Service Graphs
- 5:48Trace timeline - A blocking operation
- 6:11Trace timeline - Sequential operations
- 7:18Comparing trace structures - Unified diff
- 9:36Jaeger, a Distributed Tracing Platform
- 10:21Instrumentation not included
- 11:00Jaeger and your system
- 11:52Architecture 2017: Push
- 14:29Technology Stack
- 15:52Why do we sample
- 18:33Sampling techniques
- 21:34Head-based sampling in Jaeger
- 23:26Jaeger sampling configuration
- 24:18Tail-based (post-trace) sampling
- 27:15Tail-based sampling in Jaeger
- 28:39Jaeger with OpenTracing
- 30:27Jaeger with OpenTelemetry
- 34:42Deploying Jaeger on Kubernetes
