Rebuilding a Logistics Dispatch Engine for a YC-backed Delivery Network
Executive Summary
How we replaced a legacy database dispatch setup with a high-concurrency Redis queue, reducing transit matching latency by 85% during peak traffic spikes.
The Challenge
A rapid-delivery network was experiencing severe database write contention and transactional locks during peak transit hours. Their existing Node.js scheduler polled PostgreSQL coordinates every two seconds for thousands of active delivery drivers. As the fleet scaled past 2,000 drivers, transaction isolation conflicts caused dispatch match delays of up to 45 seconds, resulting in customer drop-offs and driver idle time.
The Approach
We decoupled the active coordination tracking from the transactional database. Driver coordinate updates were redirected to an in-memory Redis geospatial database, querying matches using radial lookups. We then built a dedicated queue dispatch system in Go that processes incoming delivery orders asynchronously. By moving booking state updates from immediate transactional writes to event-based queuing, database lock contention was completely eliminated.
The Impact
The architectural migration reduced coordination match times from a peak of 45 seconds down to a consistent sub-second response. API latency during scheduling spikes dropped by 85%, and fleet driver capacity scaled to over 10,000 active concurrent connections with zero transaction locks. This stability allowed the company to comfortably sustain a 3x increase in daily delivery volume.
What’s sitting on your engineering backlog?
Give us one task. We’ll take it from scope to production while your team keeps building the core product.