Worked on optimizing HDFS-backed index writing in the apache/celeborn repository, focusing on improving reliability and reducing unnecessary HDFS access. Developed a buffering mechanism for index writes to minimize repeated reads and writes, and implemented robust error handling to ensure proper resource cleanup. Introduced a retry mechanism with configurable backoff for creating auxiliary index and success files, addressing transient HDFS namenode failures and enhancing overall indexing stability. Utilized Java and Scala to deliver these improvements, applying skills in distributed systems and configuration management. The work targeted increased resilience and efficiency in distributed file operations without introducing new bugs during the period.