Deep Dive: Jaeger

on YouTube ↗28m

" summary

Jaeger uses a host agent + collector model, then moves to an asynchronous ingestion via Kafka pipeline with a Flink job that builds dependency graphs and protects Cassandra from spikes. Sampling includes an adaptive sampling PID controller that recomputes probabilities ~every minute to hit a target QPS. New features: HTTP client submission, trace comparison UI showing deltas (one node +815 ms), an operator for Kubernetes, and internal scale runs across 3,000 microservices. Support for Cassandra, Elasticsearch, Go/Java clients, and React UI exists.

AI-written summary · about

" chapters (16)

1. 0:00Start
2. 0:26Agenda
3. 1:00Distributed Tracing Systems
4. 2:53Technology Stack
5. 3:24Architecture 2017
6. 6:38Architecture Today
7. 8:00Data model
8. 9:17Configuration
9. 10:35Performance Settings
10. 11:29Monitoring
11. 13:28Adaptive Sampling
12. 14:29Asynchronous Ingestion
13. 15:44Trace Comparisons
14. 17:24Post Trace Sampling
15. 18:01Data Pipeline
16. 18:39Incomplete Span Support

" related talks

• 40mJaeger Deep DivePavol Loffay, Yuri Shkuro · KubeCon + CloudNativeCon North America 2019
• 41mJaeger: Intro and Deep DiveJonah Kowall · KubeCon + CloudNativeCon North America 2021
• 37mJaeger Project Deep DiveJuraci Paixão Kröhling · KubeCon + CloudNativeCon Europe 2018
• 37mDeep Dive: JaegerSteve Flanders · KubeCon + CloudNativeCon + Open Source Summit China 2019
• 37mDeep Dive: Jaeger – Steve FlandersSteve Flanders · KubeCon + CloudNativeCon China 2019
• 36mJaeger Deep DivePavol Loffay, Yuri Shkuro, Annanay Agarwal · KubeCon + CloudNativeCon North America 2020 Virtual
• 16mGetting Started with JaegerJoe Elliott · KubeCon + CloudNativeCon North America 2020 Virtual
• 28mJaeger Intro and Deep DivePrithvi Raj Venkat Raj, Joe Elliott · KubeCon + CloudNativeCon Europe 2021