Building a Scalable, Resilient System for Real-Time Fraud Detection
summary
RS Software replaces a brittle DB stack with YugabyteDB plus a Kafka consumer, deploying two active-active clusters (30 data nodes × 64 vCPU, 3 masters each) to hit 40–80k records/sec per site and sustain up to 1 billion rows/day with read latencies under 30 ms. Daily range partitioning (600–1,000M rows/partition) and delaying secondary index builds until after day-one preserves ingestion; automatic tablet splitting and YSQL compatibility enable analytics while reducing ops overhead.
AI-written summary · about
chapters (17)
- 0:00Fraud detection system intro
- 1:25RS Software’s payments role
- 2:28Database infrastructure challenges
- 4:13New database requirements
- 5:38Benchmarking YugabyteDB vs CockroachDB
- 6:46Choosing YugabyteDB
- 7:07New system architecture
- 9:39Real-time data with Kafka
- 10:23Payment channel handling
- 11:07Data partitioning strategy
- 12:31PostgreSQL compatibility & performance
- 13:13Cluster scalability benefits
- 14:19Performance metrics & results
- 15:49Secondary indexes impact
- 16:55Index usage optimization
- 17:40Auto-splitting operational overhead
- 18:02Conclusion & future outlook
related talks
29mEnterprise Scale: How a Distributed Architecture Can Deliver Scale and Resilience
25mHow Sharekhan Scaled Trading Platform with Yugabyte's Distributed SQL
18mScaling Without Limits: KFIN's Journey to Distributed SQL
11mChallenges Deploying Modern Real-Time Payment Systems on Traditional RDBMS
23mUsing a Modern Database with ML to Prevent Financial Fraud
23mScalable Architecture for Processing UPI Real-Time Transactions Using Distributed SQL
28mDisrupting the Status Quo with Distributed SQL to Deliver Data-Driven Solutions
17mBuilding a Scalable and Performant SaaS Platform with YugabyteDB