Event-based system: Uklon practice
summary
Uklon processes about 0.5 billion events per day (~1.5 TB) from mobile apps and up to 20k drivers online, and implemented an in-house streaming traffic-jam system instead of using Google. The system is built on Kafka, .NET and Python on Kubernetes, using actor models and stateful stream aggregation for matching and ETA. The talk argues that choreography, Event Sourcing and treating Kafka as the only backbone create hidden cyclic dependencies and accidental complexity, and recommends explicit orchestration with a single source of truth instead of fragile event coupling.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 1:29Uklon architecture
- 8:20Event myths and problems
- 13:50Bidirectional to unidirectional
- 18:12Streaming route calculation
- 23:56Order and tracking interaction
- 31:44Event sourcing errors
- 36:33Event architecture conclusions
- 44:58Q&A
related talks
25mMixing data streams at scale
54mSo You Want to Build An Event Driven System?
01h 04mState or Events? Which Shall I Keep?
29mStreaming & Eventing for Top Management
43mEvent driven development
36mWhen ordering matters
38mEvery Event, Everywhere, All at Once
51mBe proactive about state in a reactive system