Monoceros: Faster and Predictable Services through In-pod Load-balancing

Explore the optimization of multi-process Python applications in this 15-minute conference talk from SREcon23 Europe/Middle East/Africa. Discover how Reddit engineers achieved a 20% latency improvement, decreased CPU variance, and enhanced CPU utilization. Follow their journey from a Ruby-based system to the development of Monoceros, a solution designed to address Python's limitations in large-scale Kubernetes deployments. Gain insights into the initial challenges that led to improved load-balancing and observability techniques. Compare SO_REUSEPORT load balancing with full load balancers, and learn valuable lessons for efficient system performance at scale.