Buraq
High-performance distributed task queue architected over Redis Streams. Delivers 3.2ms median latency with idempotent delivery, backpressure handling, and automatic dead-letter re-queuing.
The Problem
Legacy RabbitMQ-based queues introduced single points of failure and unpredictable latency under 10K+ ops/second, causing message drops and financial inconsistencies for a $50M+ payment processor.
System Architecture
Rewritten in Go with Redis Streams as the durable event log, consumer groups for parallel fan-out, and gRPC for low-latency inter-service communication. A custom weighted scheduler prioritizes VIP tenant workloads without starving other consumers.
System architecture diagram — coming soon
Technical Challenges & Trade-offs
Guaranteeing exactly-once execution semantics and zero data loss during pod evictions. Built a Dead Letter Queue topology with automatic re-queuing and implemented strict graceful-shutdown hooks.
Business Impact & Metrics
Benchmarked at 12,000+ tasks/sec at 3.2ms median latency. Zero message loss over 8 months in production. Eliminated $180K/year in queue infrastructure costs while sustaining $50M+ annual transaction volume.