Skip to content
New issue

Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.

By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.

Already on GitHub? Sign in to your account

[DF] Add TypeCoercion optimizer rule #723

Merged
merged 2 commits into from
Aug 25, 2022
Merged
Show file tree
Hide file tree
Changes from all commits
Commits
File filter

Filter by extension

Filter by extension

Conversations
Failed to load comments.
Loading
Jump to
Jump to file
Failed to load files.
Loading
Diff view
Diff view
1 change: 1 addition & 0 deletions dask_planner/src/sql.rs
Original file line number Diff line number Diff line change
Expand Up @@ -4,6 +4,7 @@ pub mod function;
pub mod logical;
pub mod optimizer;
pub mod parser_utils;
pub mod rules;
pub mod schema;
pub mod statement;
pub mod table;
Expand Down
2 changes: 2 additions & 0 deletions dask_planner/src/sql/optimizer.rs
Original file line number Diff line number Diff line change
@@ -1,3 +1,4 @@
use crate::sql::rules::type_coercion::TypeCoercion;
use datafusion_common::DataFusionError;
use datafusion_expr::LogicalPlan;
use datafusion_optimizer::decorrelate_scalar_subquery::DecorrelateScalarSubquery;
Expand Down Expand Up @@ -29,6 +30,7 @@ impl DaskSqlOptimizer {
Box::new(EliminateLimit::new()),
Box::new(FilterNullJoinKeys::default()),
Box::new(FilterPushDown::new()),
Box::new(TypeCoercion::new()),
Box::new(LimitPushDown::new()),
Box::new(ProjectionPushDown::new()),
Box::new(SingleDistinctToGroupBy::new()),
Expand Down
1 change: 1 addition & 0 deletions dask_planner/src/sql/rules/mod.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1 @@
pub(crate) mod type_coercion;
277 changes: 277 additions & 0 deletions dask_planner/src/sql/rules/type_coercion.rs
Original file line number Diff line number Diff line change
@@ -0,0 +1,277 @@
// 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.

//! Optimizer rule for type validation and coercion

use arrow::datatypes::DataType;
use datafusion_common::{DFSchema, DFSchemaRef, Result};
use datafusion_expr::binary_rule::coerce_types;
use datafusion_expr::expr_rewriter::{ExprRewritable, ExprRewriter, RewriteRecursion};
use datafusion_expr::logical_plan::builder::build_join_schema;
use datafusion_expr::logical_plan::JoinType;
use datafusion_expr::type_coercion::data_types;
use datafusion_expr::utils::from_plan;
use datafusion_expr::{Expr, ExprSchemable, LogicalPlan, Signature};
use datafusion_optimizer::{OptimizerConfig, OptimizerRule};

#[derive(Default)]
pub struct TypeCoercion {}

impl TypeCoercion {
pub fn new() -> Self {
Self {}
}
}

impl OptimizerRule for TypeCoercion {
fn name(&self) -> &str {
"TypeCoercion"
}

fn optimize(
&self,
plan: &LogicalPlan,
optimizer_config: &mut OptimizerConfig,
) -> Result<LogicalPlan> {
// optimize child plans first
let new_inputs = plan
.inputs()
.iter()
.map(|p| self.optimize(p, optimizer_config))
.collect::<Result<Vec<_>>>()?;

let schema = match new_inputs.len() {
1 => new_inputs[0].schema().clone(),
2 => DFSchemaRef::new(build_join_schema(
new_inputs[0].schema(),
new_inputs[1].schema(),
&JoinType::Inner,
)?),
_ => DFSchemaRef::new(DFSchema::empty()),
};

let mut expr_rewrite = TypeCoercionRewriter { schema };

let new_expr = plan
.expressions()
.into_iter()
.map(|expr| expr.rewrite(&mut expr_rewrite))
.collect::<Result<Vec<_>>>()?;

from_plan(plan, &new_expr, &new_inputs)
}
}

struct TypeCoercionRewriter {
schema: DFSchemaRef,
}

impl ExprRewriter for TypeCoercionRewriter {
fn pre_visit(&mut self, _expr: &Expr) -> Result<RewriteRecursion> {
Ok(RewriteRecursion::Continue)
}

fn mutate(&mut self, expr: Expr) -> Result<Expr> {
match &expr {
Expr::BinaryExpr { left, op, right } => {
let left_type = left.get_type(&self.schema)?;
let right_type = right.get_type(&self.schema)?;
match right_type {
DataType::Interval(_) => {
// we don't want to cast intervals because that breaks
// the logic in the physical planner
Ok(expr)
}
_ => {
let coerced_type = coerce_types(&left_type, op, &right_type)?;
let left =
cast_if_needed(left.as_ref().clone(), &coerced_type, &self.schema)?;
let right =
cast_if_needed(right.as_ref().clone(), &coerced_type, &self.schema)?;
match (&left, &right) {
(Expr::Cast { .. }, _) | (_, Expr::Cast { .. }) => {
Ok(Expr::BinaryExpr {
left: Box::new(left),
op: *op,
right: Box::new(right),
})
}
_ => {
// no cast was added so we return the original expression
Ok(expr)
}
}
}
}
}
Expr::ScalarUDF { fun, args } => {
let new_expr =
coerce_arguments_for_signature(args.as_slice(), &self.schema, &fun.signature)?;
Ok(Expr::ScalarUDF {
fun: fun.clone(),
args: new_expr,
})
}
_ => Ok(expr),
}
}
}

/// Returns `expressions` coerced to types compatible with
/// `signature`, if possible.
///
/// See the module level documentation for more detail on coercion.
pub fn coerce_arguments_for_signature(
expressions: &[Expr],
schema: &DFSchema,
signature: &Signature,
) -> Result<Vec<Expr>> {
if expressions.is_empty() {
return Ok(vec![]);
}

let current_types = expressions
.iter()
.map(|e| e.get_type(schema))
.collect::<Result<Vec<_>>>()?;

let new_types = data_types(&current_types, signature)?;

expressions
.iter()
.enumerate()
.map(|(i, expr)| expr.clone().cast_to(&new_types[i], schema))
.collect::<Result<Vec<_>>>()
}

/// Create a cast expression
pub fn cast_if_needed(expr: Expr, data_type: &DataType, input_schema: &DFSchema) -> Result<Expr> {
let t = expr.get_type(input_schema)?;
if &t == data_type {
Ok(expr)
} else {
Ok(Expr::Cast {
expr: Box::new(expr),
data_type: data_type.clone(),
})
}
}

#[cfg(test)]
mod test {
use crate::sql::rules::type_coercion::TypeCoercion;
use arrow::datatypes::DataType;
use datafusion_common::{DFSchema, Result};
use datafusion_expr::logical_plan::{EmptyRelation, Projection};
use datafusion_expr::{
lit, Expr, LogicalPlan, ReturnTypeFunction, ScalarFunctionImplementation, ScalarUDF,
Signature, Volatility,
};
use datafusion_optimizer::OptimizerConfig;
use datafusion_optimizer::OptimizerRule;
use std::sync::Arc;

#[test]
fn binary_expr_simple_case() -> Result<()> {
let expr = lit(1.2_f64).lt(lit(2_u32));
let empty = empty();
let plan = LogicalPlan::Projection(Projection::try_new(vec![expr], empty, None)?);
let rule = TypeCoercion::new();
let mut config = OptimizerConfig::default();
let plan = rule.optimize(&plan, &mut config)?;
assert_eq!(
"Projection: Float64(1.2) < CAST(UInt32(2) AS Float64)\n EmptyRelation",
&format!("{:?}", plan)
);
Ok(())
}

#[test]
fn binary_expr_nested_case() -> Result<()> {
let expr = lit(1.2_f64).lt(lit(2_u32));
let empty = empty();
let plan = LogicalPlan::Projection(Projection::try_new(
vec![expr.clone().or(expr)],
empty,
None,
)?);
let rule = TypeCoercion::new();
let mut config = OptimizerConfig::default();
let plan = rule.optimize(&plan, &mut config)?;
assert_eq!("Projection: Float64(1.2) < CAST(UInt32(2) AS Float64) OR Float64(1.2) < CAST(UInt32(2) AS Float64)\
\n EmptyRelation", &format!("{:?}", plan));
Ok(())
}

#[test]
fn scalar_udf() -> Result<()> {
let empty = empty();
let return_type: ReturnTypeFunction = Arc::new(move |_| Ok(Arc::new(DataType::Utf8)));
let fun: ScalarFunctionImplementation = Arc::new(move |_| unimplemented!());
let udf = Expr::ScalarUDF {
fun: Arc::new(ScalarUDF::new(
"TestScalarUDF",
&Signature::uniform(1, vec![DataType::Float32], Volatility::Stable),
&return_type,
&fun,
)),
args: vec![lit(123_i32)],
};
let plan = LogicalPlan::Projection(Projection::try_new(vec![udf], empty, None)?);
let rule = TypeCoercion::new();
let mut config = OptimizerConfig::default();
let plan = rule.optimize(&plan, &mut config)?;
assert_eq!(
"Projection: TestScalarUDF(CAST(Int32(123) AS Float32))\n EmptyRelation",
&format!("{:?}", plan)
);
Ok(())
}

#[test]
fn scalar_udf_invalid_input() -> Result<()> {
let empty = empty();
let return_type: ReturnTypeFunction = Arc::new(move |_| Ok(Arc::new(DataType::Utf8)));
let fun: ScalarFunctionImplementation = Arc::new(move |_| unimplemented!());
let udf = Expr::ScalarUDF {
fun: Arc::new(ScalarUDF::new(
"TestScalarUDF",
&Signature::uniform(1, vec![DataType::Int32], Volatility::Stable),
&return_type,
&fun,
)),
args: vec![lit("Apple")],
};
let plan = LogicalPlan::Projection(Projection::try_new(vec![udf], empty, None)?);
let rule = TypeCoercion::new();
let mut config = OptimizerConfig::default();
let plan = rule.optimize(&plan, &mut config).err().unwrap();
assert_eq!(
"Plan(\"Coercion from [Utf8] to the signature Uniform(1, [Int32]) failed.\")",
&format!("{:?}", plan)
);
Ok(())
}

fn empty() -> Arc<LogicalPlan> {
let empty = Arc::new(LogicalPlan::EmptyRelation(EmptyRelation {
produce_one_row: false,
schema: Arc::new(DFSchema::empty()),
}));
empty
}
}