alamb commented on code in PR #6084:
URL: https://github.com/apache/arrow-datafusion/pull/6084#discussion_r1176640402
##########
datafusion/expr/src/logical_plan/plan.rs:
##########
@@ -484,6 +484,97 @@ impl LogicalPlan {
_ => Ok(self),
}
}
+
+ /// Returns the maximum number of rows that this plan can output.
+ pub fn max_rows(self: &LogicalPlan) -> Option<usize> {
+ match self {
+ LogicalPlan::Projection(Projection { input, .. }) =>
input.max_rows(),
+ LogicalPlan::Filter(Filter { input, .. }) => input.max_rows(),
+ LogicalPlan::Window(Window { input, .. }) => input.max_rows(),
+ LogicalPlan::Aggregate(Aggregate {
+ input, group_expr, ..
+ }) => {
+ if group_expr.is_empty() {
+ Some(1)
+ } else {
+ input.max_rows()
+ }
+ }
+ LogicalPlan::Sort(Sort { input, fetch, .. }) => {
+ match (fetch, input.max_rows()) {
+ (Some(fetch_limit), Some(input_max)) => {
+ Some(input_max.min(*fetch_limit))
+ }
+ (Some(fetch_limit), None) => Some(*fetch_limit),
+ (None, Some(input_max)) => Some(input_max),
+ (None, None) => None,
+ }
+ }
+ LogicalPlan::Join(Join {
+ left,
+ right,
+ join_type,
+ ..
+ }) => match join_type {
+ JoinType::Inner | JoinType::Left | JoinType::Right |
JoinType::Full => {
+ match (left.max_rows(), right.max_rows()) {
+ (Some(left_max), Some(right_max)) => {
+ let min_rows = match join_type {
+ JoinType::Left => left_max,
+ JoinType::Right => right_max,
+ JoinType::Full => left_max + right_max,
+ _ => 0,
Review Comment:
I think the minimum number of rows for an inner join is `None` (for
unknown) in the case of an Inner join, rather than 0.
##########
datafusion/expr/src/logical_plan/plan.rs:
##########
@@ -484,6 +484,97 @@ impl LogicalPlan {
_ => Ok(self),
}
}
+
+ /// Returns the maximum number of rows that this plan can output.
Review Comment:
```suggestion
/// Returns the maximum number of rows that this plan can output, if
known.
///
/// If `None`, the plan can return any number of rows.
/// If `Some(n)` then the plan can return at most `n` rows
/// but may return fewer.
```
##########
datafusion/optimizer/src/analyzer/subquery.rs:
##########
@@ -0,0 +1,340 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements. See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership. The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License. You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied. See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+use crate::analyzer::check_plan;
+use crate::utils::{collect_subquery_cols, split_conjunction};
+use datafusion_common::tree_node::{TreeNode, VisitRecursion};
+use datafusion_common::{DataFusionError, Result};
+use datafusion_expr::expr_rewriter::strip_outer_reference;
+use datafusion_expr::{
+ Aggregate, BinaryExpr, Cast, Expr, Filter, Join, JoinType, LogicalPlan,
Operator,
+ Window,
+};
+use std::ops::Deref;
+
+/// Do necessary check on subquery expressions and fail the invalid plan
+/// 1) Check whether the outer plan is in the allowed outer plans list to use
subquery expressions,
+/// the allowed while list: [Projection, Filter, Window, Aggregate, Join].
+/// 2) Check whether the inner plan is in the allowed inner plans list to use
correlated(outer) expressions.
+/// 3) Check and validate unsupported cases to use the correlated(outer)
expressions inside the subquery(inner) plans/inner expressions.
+/// For example, we do not want to support to use correlated expressions as
the Join conditions in the subquery plan when the Join
+/// is a Full Out Join
+pub fn check_subquery_expr(
+ outer_plan: &LogicalPlan,
+ inner_plan: &LogicalPlan,
+ expr: &Expr,
+) -> Result<()> {
+ check_plan(inner_plan)?;
+ if let Expr::ScalarSubquery(subquery) = expr {
+ // Scalar subquery should only return one column
+ if subquery.subquery.schema().fields().len() > 1 {
+ return Err(datafusion_common::DataFusionError::Plan(
+ "Scalar subquery should only return one column".to_string(),
Review Comment:
It would be nice to return the output schema / columns in this message to
help users -- maybe something like this (untested)
```suggestion
format!("Scalar subquery should only return one column,
found {}: {}",
subquery.subquery.schema().fields().len(),
subquery.subquery.schema().field_names()
)
```
##########
datafusion/optimizer/src/decorrelate_where_in.rs:
##########
@@ -602,7 +596,7 @@ mod tests {
.build()?;
let expected = "Projection: customer.c_custkey [c_custkey:Int64]\
- \n LeftSemi Join: Filter: customer.c_custkey =
__correlated_sq_1.o_custkey AND customer.c_custkey = customer.c_custkey
[c_custkey:Int64, c_name:Utf8]\
Review Comment:
this seems like an improvement to me
##########
datafusion/optimizer/src/test/mod.rs:
##########
@@ -171,6 +190,23 @@ pub fn assert_optimized_plan_eq_display_indent(
assert_eq!(formatted_plan, expected);
}
+pub fn assert_multi_rules_optimized_plan_eq_display_indent(
+ rules: Vec<Arc<dyn OptimizerRule + Send + Sync>>,
+ plan: &LogicalPlan,
+ expected: &str,
+) {
+ let optimizer = Optimizer::with_rules(rules);
+ let mut optimized_plan = plan.clone();
+ for rule in &optimizer.rules {
+ optimized_plan = optimizer
+ .optimize_recursively(rule, &optimized_plan,
&OptimizerContext::new())
+ .expect("failed to optimize plan")
+ .unwrap_or_else(|| optimized_plan.clone());
+ }
+ let formatted_plan = format!("{}", optimized_plan.display_indent_schema());
Review Comment:
```suggestion
let formatted_plan = optimized_plan.display_indent_schema().to_string();
```
##########
datafusion/optimizer/src/test/mod.rs:
##########
@@ -134,6 +134,25 @@ pub fn assert_analyzed_plan_eq_display_indent(
Ok(())
}
+
+pub fn assert_analyzer_check_err(
+ rules: Vec<Arc<dyn AnalyzerRule + Send + Sync>>,
+ plan: &LogicalPlan,
+ expected: &str,
+) {
+ let options = ConfigOptions::default();
+ let analyzed_plan =
+ Analyzer::with_rules(rules).execute_and_check(plan, &options, |_, _|
{});
+ match analyzed_plan {
+ Ok(plan) => assert_eq!(format!("{}", plan.display_indent()), "An
error"),
+ Err(ref e) => {
+ let actual = format!("{e}");
+ if expected.is_empty() || !actual.contains(expected) {
+ assert_eq!(actual, expected)
+ }
+ }
Review Comment:
```suggestion
Err(e) => {assert_contains!(e.to_string(), expected);},
```
##########
datafusion/optimizer/src/decorrelate_where_in.rs:
##########
@@ -557,7 +548,7 @@ mod tests {
let sq = Arc::new(
LogicalPlanBuilder::from(scan_tpch_table("orders"))
.filter(
- col("customer.c_custkey")
+ out_ref_col(DataType::Int64, "customer.c_custkey")
Review Comment:
I do like the idea that when creating expressions directly outer references
must be annotated explicitly
##########
datafusion/expr/src/logical_plan/plan.rs:
##########
@@ -484,6 +484,97 @@ impl LogicalPlan {
_ => Ok(self),
}
}
+
+ /// Returns the maximum number of rows that this plan can output.
+ pub fn max_rows(self: &LogicalPlan) -> Option<usize> {
+ match self {
+ LogicalPlan::Projection(Projection { input, .. }) =>
input.max_rows(),
+ LogicalPlan::Filter(Filter { input, .. }) => input.max_rows(),
+ LogicalPlan::Window(Window { input, .. }) => input.max_rows(),
+ LogicalPlan::Aggregate(Aggregate {
+ input, group_expr, ..
+ }) => {
+ if group_expr.is_empty() {
+ Some(1)
+ } else {
+ input.max_rows()
+ }
+ }
+ LogicalPlan::Sort(Sort { input, fetch, .. }) => {
+ match (fetch, input.max_rows()) {
+ (Some(fetch_limit), Some(input_max)) => {
+ Some(input_max.min(*fetch_limit))
+ }
+ (Some(fetch_limit), None) => Some(*fetch_limit),
+ (None, Some(input_max)) => Some(input_max),
+ (None, None) => None,
+ }
+ }
+ LogicalPlan::Join(Join {
+ left,
+ right,
+ join_type,
+ ..
+ }) => match join_type {
+ JoinType::Inner | JoinType::Left | JoinType::Right |
JoinType::Full => {
+ match (left.max_rows(), right.max_rows()) {
+ (Some(left_max), Some(right_max)) => {
+ let min_rows = match join_type {
+ JoinType::Left => left_max,
+ JoinType::Right => right_max,
+ JoinType::Full => left_max + right_max,
+ _ => 0,
+ };
+ Some((left_max * right_max).max(min_rows))
+ }
+ _ => None,
+ }
+ }
+ JoinType::LeftSemi | JoinType::LeftAnti => left.max_rows(),
+ JoinType::RightSemi | JoinType::RightAnti => right.max_rows(),
+ },
+ LogicalPlan::CrossJoin(CrossJoin { left, right, .. }) => {
+ match (left.max_rows(), right.max_rows()) {
+ (Some(left_max), Some(right_max)) => Some(left_max *
right_max),
+ _ => None,
+ }
+ }
+ LogicalPlan::Repartition(Repartition { input, .. }) =>
input.max_rows(),
+ LogicalPlan::Union(Union { inputs, .. }) => inputs
+ .iter()
+ .map(|plan| plan.max_rows())
+ .try_fold(0usize, |mut acc, input_max| {
+ if let Some(i_max) = input_max {
+ acc += i_max;
+ Some(acc)
+ } else {
+ None
+ }
+ }),
+ LogicalPlan::TableScan(TableScan { fetch, .. }) => *fetch,
+ LogicalPlan::EmptyRelation(_) => Some(0),
+ LogicalPlan::Subquery(_) => None,
+ LogicalPlan::SubqueryAlias(SubqueryAlias { input, .. }) =>
input.max_rows(),
+ LogicalPlan::Limit(Limit { fetch, .. }) => *fetch,
+ LogicalPlan::Distinct(Distinct { input }) => input.max_rows(),
+ LogicalPlan::Unnest(_) => None,
+ LogicalPlan::CreateMemoryTable(_)
+ | LogicalPlan::CreateExternalTable(_)
+ | LogicalPlan::CreateView(_)
+ | LogicalPlan::CreateCatalogSchema(_)
+ | LogicalPlan::CreateCatalog(_)
+ | LogicalPlan::DropTable(_)
+ | LogicalPlan::DropView(_)
+ | LogicalPlan::Explain(_)
+ | LogicalPlan::Analyze(_)
+ | LogicalPlan::Dml(_)
+ | LogicalPlan::DescribeTable(_)
+ | LogicalPlan::Prepare(_)
+ | LogicalPlan::Statement(_)
+ | LogicalPlan::Values(_)
Review Comment:
The number of rows that comes out of `Values` is known. Something like:
```
| LogicalPlan::Values(v) => Some(v.values.len())
```
##########
datafusion/optimizer/src/scalar_subquery_to_join.rs:
##########
@@ -511,16 +500,17 @@ mod tests {
// it will optimize, but fail for the same reason the unoptimized
query would
let expected = "Projection: customer.c_custkey [c_custkey:Int64]\
- \n Filter: customer.c_custkey = __scalar_sq_1.__value
[c_custkey:Int64, c_name:Utf8, __value:Int64;N]\
- \n CrossJoin: [c_custkey:Int64, c_name:Utf8, __value:Int64;N]\
- \n TableScan: customer [c_custkey:Int64, c_name:Utf8]\
- \n SubqueryAlias: __scalar_sq_1 [__value:Int64;N]\
- \n Projection: MAX(orders.o_custkey) AS __value
[__value:Int64;N]\
- \n Aggregate: groupBy=[[]], aggr=[[MAX(orders.o_custkey)]]
[MAX(orders.o_custkey):Int64;N]\
- \n Filter: customer.c_custkey = customer.c_custkey
[o_orderkey:Int64, o_custkey:Int64, o_orderstatus:Utf8, o_totalprice:Float64;N]\
- \n TableScan: orders [o_orderkey:Int64, o_custkey:Int64,
o_orderstatus:Utf8, o_totalprice:Float64;N]";
- assert_optimized_plan_eq_display_indent(
- Arc::new(ScalarSubqueryToJoin::new()),
+ \n Inner Join: customer.c_custkey = __scalar_sq_1.__value
[c_custkey:Int64, c_name:Utf8, __value:Int64;N]\
Review Comment:
that looks like a better plan to me
##########
datafusion/optimizer/src/analyzer/subquery.rs:
##########
@@ -0,0 +1,340 @@
+// Licensed to the Apache Software Foundation (ASF) under one
+// or more contributor license agreements. See the NOTICE file
+// distributed with this work for additional information
+// regarding copyright ownership. The ASF licenses this file
+// to you under the Apache License, Version 2.0 (the
+// "License"); you may not use this file except in compliance
+// with the License. You may obtain a copy of the License at
+//
+// http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing,
+// software distributed under the License is distributed on an
+// "AS IS" BASIS, WITHOUT WARRANTIES OR CONDITIONS OF ANY
+// KIND, either express or implied. See the License for the
+// specific language governing permissions and limitations
+// under the License.
+
+use crate::analyzer::check_plan;
+use crate::utils::{collect_subquery_cols, split_conjunction};
+use datafusion_common::tree_node::{TreeNode, VisitRecursion};
+use datafusion_common::{DataFusionError, Result};
+use datafusion_expr::expr_rewriter::strip_outer_reference;
+use datafusion_expr::{
+ Aggregate, BinaryExpr, Cast, Expr, Filter, Join, JoinType, LogicalPlan,
Operator,
+ Window,
+};
+use std::ops::Deref;
+
+/// Do necessary check on subquery expressions and fail the invalid plan
+/// 1) Check whether the outer plan is in the allowed outer plans list to use
subquery expressions,
+/// the allowed while list: [Projection, Filter, Window, Aggregate, Join].
+/// 2) Check whether the inner plan is in the allowed inner plans list to use
correlated(outer) expressions.
+/// 3) Check and validate unsupported cases to use the correlated(outer)
expressions inside the subquery(inner) plans/inner expressions.
+/// For example, we do not want to support to use correlated expressions as
the Join conditions in the subquery plan when the Join
+/// is a Full Out Join
+pub fn check_subquery_expr(
+ outer_plan: &LogicalPlan,
+ inner_plan: &LogicalPlan,
+ expr: &Expr,
+) -> Result<()> {
+ check_plan(inner_plan)?;
+ if let Expr::ScalarSubquery(subquery) = expr {
+ // Scalar subquery should only return one column
+ if subquery.subquery.schema().fields().len() > 1 {
+ return Err(datafusion_common::DataFusionError::Plan(
+ "Scalar subquery should only return one column".to_string(),
+ ));
+ }
+ // Correlated scalar subquery must be aggregated to return at most one
row
+ if !subquery.outer_ref_columns.is_empty() {
+ match strip_inner_query(inner_plan) {
+ LogicalPlan::Aggregate(agg) => {
+ check_aggregation_in_scalar_subquery(inner_plan, agg)
+ }
+ LogicalPlan::Filter(Filter { input, .. })
+ if matches!(input.as_ref(), LogicalPlan::Aggregate(_)) =>
+ {
+ if let LogicalPlan::Aggregate(agg) = input.as_ref() {
+ check_aggregation_in_scalar_subquery(inner_plan, agg)
+ } else {
+ Ok(())
+ }
+ }
+ _ => {
+ if inner_plan
+ .max_rows()
+ .filter(|max_row| *max_row <= 1)
+ .is_some()
+ {
+ Ok(())
+ } else {
+ Err(DataFusionError::Plan(
+ "Correlated scalar subquery must be aggregated to
return at most one row"
+ .to_string(),
+ ))
+ }
+ }
+ }?;
+ match outer_plan {
+ LogicalPlan::Projection(_)
+ | LogicalPlan::Filter(_) => Ok(()),
+ LogicalPlan::Aggregate(Aggregate {group_expr, aggr_expr,..})
=> {
+ if group_expr.contains(expr) && !aggr_expr.contains(expr) {
+ // TODO revisit this validation logic
+ Err(DataFusionError::Plan(
+ "Correlated scalar subquery in the GROUP BY clause
must also be in the aggregate expressions"
+ .to_string(),
+ ))
+ } else {
+ Ok(())
+ }
+ },
+ _ => Err(DataFusionError::Plan(
+ "Correlated scalar subquery can only be used in
Projection, Filter, Aggregate plan nodes"
+ .to_string(),
+ ))
+ }?;
+ }
+ check_correlations_in_subquery(inner_plan, true)
+ } else {
+ match outer_plan {
+ LogicalPlan::Projection(_)
+ | LogicalPlan::Filter(_)
+ | LogicalPlan::Window(_)
+ | LogicalPlan::Aggregate(_)
+ | LogicalPlan::Join(_) => Ok(()),
+ _ => Err(DataFusionError::Plan(
+ "In/Exist subquery can only be used in \
+ Projection, Filter, Window functions, Aggregate and Join plan
nodes"
+ .to_string(),
+ )),
+ }?;
+ check_correlations_in_subquery(inner_plan, false)
+ }
+}
+
+// Recursively check the unsupported outer references in the sub query plan.
+fn check_correlations_in_subquery(
+ inner_plan: &LogicalPlan,
+ is_scalar: bool,
+) -> Result<()> {
+ check_inner_plan(inner_plan, is_scalar, false, true)
+}
+
+// Recursively check the unsupported outer references in the sub query plan.
+fn check_inner_plan(
+ inner_plan: &LogicalPlan,
+ is_scalar: bool,
+ is_aggregate: bool,
+ can_contain_outer_ref: bool,
+) -> Result<()> {
+ if !can_contain_outer_ref && contains_outer_reference(inner_plan) {
+ return Err(DataFusionError::Plan(
+ "Accessing outer reference columns is not allowed in the
plan".to_string(),
+ ));
+ }
+ // We want to support as many operators as possible inside the correlated
subquery
+ match inner_plan {
+ LogicalPlan::Aggregate(_) => {
+ inner_plan.apply_children(&mut |plan| {
+ check_inner_plan(plan, is_scalar, true,
can_contain_outer_ref)?;
+ Ok(VisitRecursion::Continue)
+ })?;
+ Ok(())
+ }
+ LogicalPlan::Filter(Filter {
+ predicate, input, ..
+ }) => {
+ let (correlated, _): (Vec<_>, Vec<_>) =
split_conjunction(predicate)
+ .into_iter()
+ .partition(|e| e.contains_outer());
+ let maybe_unsupport = correlated
+ .into_iter()
+ .filter(|expr| !can_pullup_over_aggregation(expr))
+ .collect::<Vec<_>>();
+ if is_aggregate && is_scalar && !maybe_unsupport.is_empty() {
+ return Err(DataFusionError::Plan(format!(
+ "Correlated column is not allowed in predicate: {:?}",
+ predicate
+ )));
+ }
+ check_inner_plan(input, is_scalar, is_aggregate,
can_contain_outer_ref)
+ }
+ LogicalPlan::Window(window) => {
+ check_mixed_out_refer_in_window(window)?;
+ inner_plan.apply_children(&mut |plan| {
+ check_inner_plan(plan, is_scalar, is_aggregate,
can_contain_outer_ref)?;
+ Ok(VisitRecursion::Continue)
+ })?;
+ Ok(())
+ }
+ LogicalPlan::Projection(_)
+ | LogicalPlan::Distinct(_)
+ | LogicalPlan::Sort(_)
+ | LogicalPlan::CrossJoin(_)
+ | LogicalPlan::Union(_)
+ | LogicalPlan::TableScan(_)
+ | LogicalPlan::EmptyRelation(_)
+ | LogicalPlan::Limit(_)
+ | LogicalPlan::Values(_)
+ | LogicalPlan::Subquery(_)
+ | LogicalPlan::SubqueryAlias(_) => {
+ inner_plan.apply_children(&mut |plan| {
+ check_inner_plan(plan, is_scalar, is_aggregate,
can_contain_outer_ref)?;
+ Ok(VisitRecursion::Continue)
+ })?;
+ Ok(())
+ }
+ LogicalPlan::Join(Join {
+ left,
+ right,
+ join_type,
+ ..
+ }) => match join_type {
+ JoinType::Inner => {
+ inner_plan.apply_children(&mut |plan| {
+ check_inner_plan(
+ plan,
+ is_scalar,
+ is_aggregate,
+ can_contain_outer_ref,
+ )?;
+ Ok(VisitRecursion::Continue)
+ })?;
+ Ok(())
+ }
+ JoinType::Left | JoinType::LeftSemi | JoinType::LeftAnti => {
+ check_inner_plan(left, is_scalar, is_aggregate,
can_contain_outer_ref)?;
+ check_inner_plan(right, is_scalar, is_aggregate, false)
+ }
+ JoinType::Right | JoinType::RightSemi | JoinType::RightAnti => {
+ check_inner_plan(left, is_scalar, is_aggregate, false)?;
+ check_inner_plan(right, is_scalar, is_aggregate,
can_contain_outer_ref)
+ }
+ JoinType::Full => {
+ inner_plan.apply_children(&mut |plan| {
+ check_inner_plan(plan, is_scalar, is_aggregate, false)?;
+ Ok(VisitRecursion::Continue)
+ })?;
+ Ok(())
+ }
+ },
+ _ => Err(DataFusionError::Plan(
+ "Unsupported operator in the subquery plan.".to_string(),
+ )),
+ }
+}
+
+fn contains_outer_reference(inner_plan: &LogicalPlan) -> bool {
+ inner_plan
+ .expressions()
+ .iter()
+ .any(|expr| expr.contains_outer())
+}
+
+fn check_aggregation_in_scalar_subquery(
+ inner_plan: &LogicalPlan,
+ agg: &Aggregate,
+) -> Result<()> {
+ if agg.aggr_expr.is_empty() {
Review Comment:
it seems like you might be able to include this logic in
`LogicalPlan::max_rows()` and avoid having a special check for aggregates
--
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]
