Worked on the pinterest/starrocks repository to address a critical bug affecting materialized view refresh operations on Paimon base tables. Focused on optimizing query performance and maintaining data integrity, the developer implemented a fix that ensures partition predicates are correctly pushed down during refresh cycles. This adjustment improves the efficiency of materialized view refresh query plans, reducing the risk of stale data in analytics workloads. Leveraging Java and SQL for back end development, the work demonstrates a targeted approach to enhancing reliability and performance in complex data processing environments, with careful attention to the nuances of partition predicate handling in distributed systems.