Cardinality and cost estimation are critical components of query optimization, as they directly influence the construction of efficient physical execution plans. While machine learning-based estimators have achieved notable success, they face several challenges: (1) Training data derived from rigid, template-driven benchmarks exhibits significant distributional divergence from real-world query workloads, a challenge further compounded by the manual template design in exhaustively representing the full spectrum of query patterns. (2) These methods demonstrate limited generalization capabilities, especially in scenarios involving sub-plan estimation or queries that significantly deviate from the training query templates. Furthermore, the inherent inefficiency of operator-level cardinality estimation frequently undermines its applicability for accurate cost estimation. (3) These approaches frequently fail to leverage the rich semantic information and dynamic dependencies between operators.
To solve the problems, a research team led by Hong CHEN (Renmin University of China) published their new research on 15 September 2026 in Frontiers of Computer Science co-published by Higher Education Press and Springer Nature.
The team proposed propose OCACO, an Operator-level generalized CArdinality and COst joint estimator.
To address Challenge 1, they introduce a query synthesis method to automatically generate high-quality and diverse training data for pre-training OCACO. This enables the model to achieve better generalization. The pre-trained OCACO can be applied in two ways: (1) zero-shot inference for unseen queries without additional training, or (2) fine-tuning on specific workloads to enhance estimation accuracy.
For Challenge 2, They design a novel operator-level training and prediction framework. This framework can simultaneously predict the cardinality and cost of each sub-plan within a query plan, using the cardinality estimation results as features for cost estimation. Specifically, They propose an operator-level plan encoding model, named Tree-GAU (Tree-aware Gated Attention Unit), which integrates gate and attention mechanisms. Additionally, They introduce a novel loss function that enables the OCACO to jointly CAE and COE at the operator level.
To resolve Challenge 3, They develop a novel feature encoding method that leverages a pre-trained language model to capture the semantic characteristics of each operator in a plan.
Extensive experiments demonstrate the superiority of OCACO in CAE and COE tasks, achieving accurate estimations across multiple datasets compared to state-of-the-art methods. Additionally, the pre-trained version exhibits exceptional performance in zero-shot and fine-tuning setting.
DOI:10.1007/s11704-025-50348-6