The release of Ruby 3.2 brought many improvements to YJIT, both in terms of performance and memory usage. Importantly, the reduced memory footprint makes YJIT much more viable for real-world production deployments. Enough so that YJIT is no longer marked as experimental, and Shopify has chosen to deploy it on its worldwide StoreFront renderer infrastructure.

Making YJIT suitable for production is not just a matter of getting the highest performance numbers possible. In this talk, I want to go over the story and the process that has enabled us to bring YJIT to a state where we consider it to be production-ready, all the way from the original research ideas that lead to the creation of YJIT, to the choice of benchmarks and the curation of our benchmark suite, to the methodology used to benchmark YJIT, but also discussing specific key performance optimizations that YJIT performs, and how we were able to optimize YJIT for a real production workload using a data-driven approach.