Writing

Thoughts & notes

On backend systems, databases, and running software in production — the notes I wish I'd found at 2 a.m.

2026
Designing a Public Feed for 10 Million Users: Security, Caching, and Launch-Day Scale on GCPA social platform's public feed has to survive a launch-day traffic surge while never being the thing that leaks private data. Here's the two-project GCP architecture we designed—separate VPCs, an event-driven read model, unsigned media URLs, and a three-layer cache funnel that turns ten million requests into a few thousand database queries.22 min
2025
CliffsTwo notes written eight years apart, on the same metaphor — a jump between cliffs. What changed in between wasn't the cliff. It was whether you can afford to keep hold of what you're carrying while you're still in the air.2 min
2024
Tripping on Purpose: A Circuit Breaker for Go ServicesWhen a dependency is failing, the worst thing your service can do is wait for it. How cascading failures spread through a fleet of polite services, and the small open-source Go package we built to stop them: three states, six knobs, one Redis key shared across replicas, a fallback, and retries you can shape.7 minBuilding a Fault-Tolerant Communication Service for 150K+ Monthly EventsHow we architected a centralized notification system that evolved from a simple queue-based design to a robust, priority-driven platform with guaranteed delivery—a technical journey through failure modes, scheduler patterns, and production resilience.11 min
2023
Case Study: How We Cut MySQL CPU Usage from 87% to 40% with ProxySQL and Read ReplicasA production MySQL 5.7 instance serving 800+ GB and 100+ requests per second kept hitting 85%+ CPU every few days. Here's the full case study of how we evaluated InnoDB Cluster, Percona XtraDB, and Cloud SQL—and why a master-replica architecture with ProxySQL won.8 min