Left-Deep Tree Join Enumeration for Subquery Optimization

The patent segments the query optimization process into distinct phases: building a query optimization graph for each query block, generating access and join methods separately, and determining optimal plans through systematic evaluation. This segmentation allows memory-efficient processing by handling subqueries as discrete units rather than transforming entire query structures, thus improving query optimization capability without proportionally increasing memory requirements.

The patent applies partial action by focusing optimization efforts on specific query blocks and subqueries individually through the query optimization graph, rather than attempting to optimize the entire query simultaneously. This approach enables effective query optimization for each subquery while avoiding the exponential memory growth that would result from comprehensive query transformation, thus resolving the contradiction between optimization capability and memory usage.

The patent divides the search space into manageable segments by creating separate query optimization graphs for each query block and subquery. Each graph represents a discrete optimization problem with its own set of plan nodes, access methods, and join methods. This segmentation reduces search space complexity by preventing the combinatorial explosion that would occur if all possible query transformations were considered simultaneously, while still achieving comprehensive optimization through systematic evaluation of each segment.

The patent performs preliminary action by pre-generating sets of access methods and join methods for each plan node before determining the optimal access plan. This preliminary enumeration of possibilities allows the optimizer to evaluate and compare methods systematically without exploring the entire search space exhaustively during query execution, thus improving query execution efficiency while keeping search space complexity manageable through advance preparation.