xudong963 commented on PR #22343: URL: https://github.com/apache/datafusion/pull/22343#issuecomment-4609938359
Haven't reviewed the PR, but one related implementation in my mind is DuckDB's predicate reordering. DuckDB implements this as a dedicated `REORDER_FILTER` optimizer pass: https://github.com/duckdb/duckdb/blob/9ddf2e47203038eadab10735d0b5f301893c724e/src/optimizer/optimizer.cpp#L395-L399 The optimizer visits logical filters and nested conjunctions: https://github.com/duckdb/duckdb/blob/9ddf2e47203038eadab10735d0b5f301893c724e/src/optimizer/expression_heuristics.cpp#L17-L33 A notable safety choice is that DuckDB skips reordering entirely if any predicate can throw: https://github.com/duckdb/duckdb/blob/9ddf2e47203038eadab10735d0b5f301893c724e/src/optimizer/expression_heuristics.cpp#L49-L53 It also uses a numeric heuristic cost model https://github.com/duckdb/duckdb/blob/9ddf2e47203038eadab10735d0b5f301893c724e/src/optimizer/expression_heuristics.cpp#L89-L167 DuckDB then goes one step further at **execution time**: filter order can be adaptively adjusted based on observed runtime( This is cool): https://github.com/duckdb/duckdb/blob/9ddf2e47203038eadab10735d0b5f301893c724e/src/execution/adaptive_filter.cpp#L109-L166 So the DuckDB design has three useful ideas: 1. keep predicate reordering as a separate optimizer concern; 2. avoid reordering fallible predicates; 3. use static heuristic ordering as an initial order, then let runtime feedback improve it. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
