Stop Performance Regressions Before Release: Valkey’s CI Benchmarking

This talk gives the open source community a reproducible, automated pattern for continuous performance validation applicable to any performance-critical system. By sharing the valkey-perf-benchmark tooling, workflows, and lessons learned, we help maintainers catch regressions before releases, improve release confidence, and reduce “performance surprise” bugs reported by users. The framework embeds performance testing into the development lifecycle using GitHub Workflows on self-hosted runners, runs benchmarks after merges and on-demand for PRs, backfills historical commits to build baselines, and implements a scalable observability stack with PostgreSQL, Grafana, and Kubernetes. It supports both local and CI environments and demonstrates how system-level tuning (e.g., IRQ affinity and CPU/process pinning) can yield more deterministic, higher performance results. By leveraging widely adopted open source components, this approach is portable across projects of any size, and attendees will be able to adopt or adapt these practices immediately, fostering higher quality databases, messaging systems, and other performance-critical services built on open source.