Astra Dynamic Chunks: How We Saved by Redesigning a Key Part of Astra

This post details the redesign of Astra's chunk management system to implement dynamic chunk sizing. The changes involved modifying the Cluster Manager and Cache nodes to utilize bin packing algorithms for more efficient allocation of cache node capacity. Specifically, it describes the redesign of cache nodes to advertise capacity instead of slots, and the manager's use of first-fit bin packing to assign chunks to cache nodes based on metadata and assignments persisted in Zookeeper. This resulted in a reduction of cache nodes by up to 50% and a 20% reduction in cache node costs.