Standby in Combat: Scaling an App in the World's No.2 Classifieds
summary
The talk describes Avito’s multi-year experience using binary replication and standby replicas to scale a top-two global classified ads application. It explains a replication-based horizontal scaling approach, the challenges (like stale reads), and presents a pattern and Avito’s implementation for systems where stale reads are unacceptable, along with disaster-recovery plans for distributed systems.
AI-written summary · about
chapters (12)
- 0:00Introduction
- 3:18Replication history and evolution
- 7:14Async replication problems
- 10:01Lag issues workaround
- 15:44Aviто Smart Cache architecture
- 21:05Vacuum problems and fixes
- 24:42Archiving and compression issues
- 28:59Fault tolerance and recovery plans
- 31:51Logical replication for search
- 34:15Standby under load
- 36:07Zero-downtime DB updates
- 37:53Conclusion and replication future
related talks
28mPractical Approaches To Scaling Distributed Transactions
37mScaling a Distributed Stateful System: A LinkedIn Case Study
51mExperience Developing, Debugging, and Deploying a Hot-Swap Trading System
41mIntroduction to distributed systems
51mScaling yout Architecture with Services and Events
32mHow does your application behaves when everything goes wrong?
18mEngineering Resilience at Scale; Lessons from a Distributed Systems Transformation
48mReplication as a First-Class Citizen: The Art of Global Resilience