Worked on the apache/doris repository to enhance the reliability and observability of workload metrics in the executor path. Developed a real elapsed time-based metrics refresh mechanism using C++ that replaced the previous fixed-interval approach, ensuring CPU and IO rates accurately reflect actual workload even during delays or configuration changes. Introduced a division-by-zero safeguard for sub-second refresh intervals to prevent misleading metric values, and created unit tests to validate the new timing logic and edge cases. This work improved the accuracy of performance dashboards and supported more effective capacity planning, demonstrating a focus on performance optimization and robust engineering practices.