Worked on the milvus-io/milvus repository, focusing on backend stability and reliability over a two-month period. Addressed a thread-safety issue in the ResourceManager recovery process by implementing robust locking mechanisms in Go, preventing infinite loops and reducing downtime risk. Expanded test coverage to ensure correct behavior under various concurrent scenarios, strengthening production resilience. Additionally, improved the access logging subsystem by introducing defensive handling for missing trace IDs, eliminating panics and enhancing observability reliability. The work emphasized concurrent programming and thorough testing, resulting in a more stable backend system that better supports production workloads and aligns with ongoing reliability initiatives.