OmniQO: An Adaptive Framework for Integrating ML and Traditional Query Optimizers

Query optimization is essential for database performance and is primarily managed by the query optimizer. Over the years, various optimizer have been developed, each with unique strengths and limitations. These limitations, such as high tail latency and poor performance on complex queries, have hinder their applicability in production environments. To leverage the strengths of various optimizers while mitigating their weakness, we propose a lightweight, extensible, and adaptive framework called OmniQO (Omni Query Optimizer). OmniQO is designed to seamlessly integrate multiple query optimizers by dynamically analyzing query characteristics and intelligently selecting the optimizer best suited for the current query at negligible cost. By combining the strengths of different approaches and addressing their limitations, OmniQO enhances end-to-end query performance, prevents model degradation, and significantly improves system stability. Extensive experimental results confirm that OmniQO outperforms state-of-the-art methods in query latency, stability, and efficiency.