Worked on the pingcap/tidb-engine-ext repository to deliver a targeted optimization for Raftstore’s compaction process, specifically adjusting the compaction order to process the write column family before the default column family. This approach reduced the workload of deletions during compaction when a compaction-filter was active on the write column family, resulting in lower I/O and improved write-path efficiency for deletion-heavy workloads. The work demonstrated a strong understanding of database internals, distributed systems, and performance optimization, utilizing Rust and integrating with RocksDB. The changes enhanced the reliability and maintainability of the compaction pipeline within the Raftstore component.