Down the Rabbit Hole: Uncovering the Reasons Behind a Mysterious Performance Regression

Kong Engineering adopts a rigorous approach to Gateway performance benchmarking by integrating a dedicated bare metal cluster into their CI/CD process, which facilitates extensive performance testing across various release stages to prevent and address performance regressions. A case study involving a recent regression of approximately 10% in Requests Per Second (RPS) highlighted the challenges of maintaining performance amidst fundamental changes to the core proxy path, particularly the cache key algorithm used by Kong's Rust-based router. Although the refactoring aimed to improve efficiency by generating cache keys using only configured conditions, it inadvertently led to performance issues due to LuaJIT's trace aborts during JIT compilation. This was resolved by replacing closures with local functions, resulting in a reduction of performance regression to within the margin of error. The team plans to enhance LuaJIT debuggability to better monitor and address such issues early in the development process, ultimately ensuring a robust and efficient platform for customers.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.