unknown

Uber evolved its database overload protection from simple quota-based rate limiting to an intelligent load management system. The journey involved replacing static CoDel queues with Cinnamon, a priority-aware load shedder that uses PID-based control to dynamically adjust queue timeouts and inflight limits. The final architecture unifies multiple overload signals (concurrency, memory, write bytes, follower lag) into a single pluggable decision-making engine that sheds traffic by priority while maintaining fairness across tenants. This resulted in 80% higher throughput under overload, ~70% reduction in P99 latency, and ~93% fewer goroutines during stress.

How Uber Conquered Database Overload: The Journey from Static Rate-Limiting to Intelligent Load Management

SeyoungCho

<p>Cool. But still remember that your app isn’t uber so just use something simpler first before over engineering into 999 microservices and load balancers</p>
