All posts published on this site.
Everything published here, newest first.
DataFusion implements ASOF join as a forward-only merge over two sorted inputs, keeping exactly one candidate right row per equality group and replacing it as better matches …
A dictionary column is a pre-built hash cache: DataFusion hashes each distinct dictionary value once, scatters those hashes to the keys, and memoizes value-index to group-slot so …
Constant folding happens in the logical optimizer via a dummy-batch evaluator, algebraic rules fire in a fixpoint loop, and casts are unwrapped into the literal when the conversion …
DataFusion can fold a projection into the operator below it, so a hash join only builds the output columns that are actually used — removing both a plan node and a per-column take.
WindowAggExec asks the planner for hash-partitioned, sorted input, then walks each partition’s frame with a sliding accumulator that adds rows entering the frame and retracts rows …