Why We Chose Rust over Go for Our New Caching Layer

A team that previously migrated from Node.js to Go found Go's garbage collector caused P99 latency spikes of 50–100ms in a high-throughput in-memory caching layer processing millions of requests per second. They rewrote the cache in Rust using DashMap, eliminating GC pauses entirely. Results: P99 latency dropped from 50ms to 1.5ms, memory usage fell 70%, and CPU usage halved. The conclusion is that Go remains the default for standard microservices, but Rust is the right tool for extreme, latency-sensitive workloads.