Optimizing Ruby's Load Path for Faster CI

Starting a new role at Intercom, I was tasked with enhancing the performance of the company's monolithic CI system. A crucial aspect of CI performance is the speed at which a Ruby process can be prepared to run tests. This becomes especially important when dealing with large test suites that need to be run in parallel. The setup phase, which includes fetching source code and preparing services, can significantly impact the overall time if not optimized.

To address this, I focused on reducing the setup time, particularly the application boot time. This led me to explore Bootsnap, a tool that optimizes Ruby's file loading process. Bootsnap's main feature is load path caching, which replaces Ruby's slow linear search with a faster hash lookup, significantly reducing the time spent on file loading.

However, adding a cache introduces the challenge of cache invalidation. Bootsnap handles this by recording the modification time (mtime) of directories, ensuring the cache is invalidated when files are added or removed. This approach, while effective, requires checking mtimes across directories, which can be costly.

To further optimize, I addressed the N+1 syscall issue in Bootsnap's load path scanner. By implementing a more efficient directory scanning method, I achieved a 2x speed improvement. Additionally, I proposed a new Ruby method, Dir.scan, to further enhance directory scanning performance.

Beyond Bootsnap, I examined Ruby's File.join method, which was significantly slower than a simple string interpolation due to unnecessary multi-byte encoding checks. By optimizing this and other path handling methods, I achieved a 7x speed improvement, making File.join faster than string interpolation in Ruby 4.1.

These optimizations not only improved the CI setup time but also demonstrated the potential for performance gains in other areas of Ruby's file handling methods.