Araina contributed a targeted backend enhancement to the apache/kudu repository, focusing on memory-aware rowset compaction to address out-of-memory risks during background operations. Using C++ and backend development expertise, Araina modified the compaction process to factor in undo delta sizes when selecting rowsets, effectively integrating these metrics into the budgeting logic. This approach enabled the system to maintain resident memory around 1GB, a substantial reduction from previous usage, and ensured stable compaction on memory-constrained nodes. The solution was validated across multiple memory scenarios, demonstrating careful engineering and a deep understanding of resource management in large-scale distributed storage systems.