Add support for boolean columns in pruning logic #500
Conversation
alamb
changed the title
| @@ -324,42 +417,20 @@ impl<'a> PruningExpressionBuilder<'a> { | |||
| self.scalar_expr | |||
| } | |||
|
|
|||
| fn is_stat_column_missing(&self, statistics_type: StatisticsType) -> bool { | |||
There was a problem hiding this comment.
This logic was just refactored into RequiredStatColumns so I could reuse it
| use crate::logical_plan; | ||
| let field = schema.field_with_name(column_name).ok()?; | ||
|
|
||
| if matches!(field.data_type(), &DataType::Boolean) { |
There was a problem hiding this comment.
Here is the actual logic / rules
| let result = p.prune(&statistics).unwrap_err(); | ||
| assert!( | ||
| result.to_string().contains( | ||
| "Data type Boolean not supported for scalar operation on dyn array" |
There was a problem hiding this comment.
these aren't great messages, but they are what happens on master today, and I figured I would document them for posterity (and maybe inspire people to help fix them)
| // predicate expression can only be a binary expression | ||
| let (left, op, right) = match expr { | ||
| Expr::BinaryExpr { left, op, right } => (left, *op, right), | ||
| Expr::Column(name) => { | ||
| if let Some(expr) = | ||
| build_single_column_expr(&name, schema, required_columns, false) |
There was a problem hiding this comment.
This kind of pattern probably can be written a bit shorter with some combinators. Something like:
build_single_column_expr(&name, schema, required_columns).ok().or(Ok(unhandled))
There was a problem hiding this comment.
An excellent idea -- I will do so
Codecov Report
@@ Coverage Diff @@
## master #500 +/- ##
==========================================
+ Coverage 75.92% 76.02% +0.09%
==========================================
Files 154 154
Lines 26195 26421 +226
==========================================
+ Hits 19889 20087 +198
- Misses 6306 6334 +28
Continue to review full report at Codecov.
|
|
I rebased this PR and added a few more tests. The code is unchanged |
Closes #490
This PR adds support for pruning of boolean predicates such as
flag_col, andnot flag_colso that they can be used to prune row groups from parquet files and other predicatesIt does not add code to handle
flag_col = trueandflag_col != false(which currently error and continue to do so) as those are simplified in the ConstantEvaluation pass.This ended up being a larger change than I wanted because the logic to create
col_minandcol_maxreferences was intertwined inPruningExpressionBuilderRationale for this change
See #490
What changes are included in this PR?
Major changes:
stat_column_reqinto a newRequiredStatColumnsstructStatisticsColumnsAre there any user-facing changes?
Additional predicates can be used to prune