Buraq
High-Performance Distributed Task Queue processing 12k+ tasks/sec at <5ms P99 latency. Engineered DLQs and graceful shutdowns for $50M+ in annual transactions.
The Problem
The legacy RabbitMQ task queue became a bottleneck, dropping messages under heavy load and causing financial inconsistencies for a high-volume payment processor.
System Architecture
Completely rewritten in Go using Redis Streams for the event log and gRPC for blazing-fast microservice communication. Features a custom scheduler algorithm to prioritize VIP tenant workloads.
System architecture diagram — coming soon
Technical Challenges & Trade-offs
Ensuring exact-once execution and zero data loss during pod evictions. Engineered an advanced Dead Letter Queue (DLQ) topology and implemented strict graceful shutdown hooks to let inflight tasks complete before termination.
Business Impact & Metrics
Benchmarked at 12,000+ tasks/sec at sub-5ms P99 latency. Now reliably powers $50M+ in annual transaction processing with zero dropped payloads.