alamb commented on code in PR #12269:
URL: https://github.com/apache/datafusion/pull/12269#discussion_r1773892921
##########
datafusion/physical-plan/src/aggregates/group_values/mod.rs:
##########
@@ -92,5 +96,36 @@ pub fn new_group_values(schema: SchemaRef) -> Result<Box<dyn
GroupValues>> {
}
}
- Ok(Box::new(GroupValuesRows::try_new(schema)?))
+ if schema
+ .fields()
+ .iter()
+ .map(|f| f.data_type())
+ .all(has_row_like_feature)
+ {
+ Ok(Box::new(GroupValuesColumn::try_new(schema)?))
+ } else {
+ Ok(Box::new(GroupValuesRows::try_new(schema)?))
+ }
+}
+
+fn has_row_like_feature(data_type: &DataType) -> bool {
Review Comment:
Minor nit -- but since this list of types needs to remain in sync with
`GroupValuesColumn` it might make sense to move it into that module too
Like `GroupValuesColumn::is_supported` or something
##########
datafusion/physical-plan/src/aggregates/group_values/group_value_row.rs:
##########
@@ -0,0 +1,456 @@
+// 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 arrow::array::BooleanBufferBuilder;
+use arrow::array::BufferBuilder;
+use arrow::array::GenericBinaryArray;
+use arrow::array::GenericStringArray;
+use arrow::array::OffsetSizeTrait;
+use arrow::array::PrimitiveArray;
+use arrow::array::{Array, ArrayRef, ArrowPrimitiveType, AsArray};
+use arrow::buffer::NullBuffer;
+use arrow::buffer::OffsetBuffer;
+use arrow::buffer::ScalarBuffer;
+use arrow::datatypes::ArrowNativeType;
+use arrow::datatypes::ByteArrayType;
+use arrow::datatypes::DataType;
+use arrow::datatypes::GenericBinaryType;
+use arrow::datatypes::GenericStringType;
+use datafusion_common::utils::proxy::VecAllocExt;
+
+use std::sync::Arc;
+use std::vec;
+
+use datafusion_physical_expr_common::binary_map::{OutputType,
INITIAL_BUFFER_CAPACITY};
+
+/// Trait for group values column-wise row comparison
+///
+/// Implementations of this trait store a in-progress collection of group
values
+/// (similar to various builders in Arrow-rs) that allow for quick comparison
to
+/// incoming rows.
+///
+pub trait ArrayRowEq: Send + Sync {
+ /// Returns equal if the row stored in this builder at `lhs_row` is equal
to
+ /// the row in `array` at `rhs_row`
+ fn equal_to(&self, lhs_row: usize, array: &ArrayRef, rhs_row: usize) ->
bool;
+ /// Appends the row at `row` in `array` to this builder
+ fn append_val(&mut self, array: &ArrayRef, row: usize);
+ /// Returns the number of rows stored in this builder
+ fn len(&self) -> usize;
+ /// Returns the number of bytes used by this [`ArrayRowEq`]
+ fn size(&self) -> usize;
+ /// Builds a new array from all of the stored rows
+ fn build(self: Box<Self>) -> ArrayRef;
+ /// Builds a new array from the first `n` stored rows, shifting the
+ /// remaining rows to the start of the builder
+ fn take_n(&mut self, n: usize) -> ArrayRef;
+}
+
+pub struct PrimitiveGroupValueBuilder<T: ArrowPrimitiveType> {
+ group_values: Vec<T::Native>,
+ nulls: Vec<bool>,
+ // whether the array contains at least one null, for fast non-null path
+ has_null: bool,
+ nullable: bool,
+}
+
+impl<T> PrimitiveGroupValueBuilder<T>
+where
+ T: ArrowPrimitiveType,
+{
+ pub fn new(nullable: bool) -> Self {
+ Self {
+ group_values: vec![],
+ nulls: vec![],
+ has_null: false,
+ nullable,
+ }
+ }
+}
+
+impl<T: ArrowPrimitiveType> ArrayRowEq for PrimitiveGroupValueBuilder<T> {
+ fn equal_to(&self, lhs_row: usize, array: &ArrayRef, rhs_row: usize) ->
bool {
+ // non-null fast path
+ // both non-null
+ if !self.nullable {
+ return self.group_values[lhs_row]
+ == array.as_primitive::<T>().value(rhs_row);
+ }
+
+ // lhs is non-null
+ if self.nulls[lhs_row] {
+ if array.is_null(rhs_row) {
+ return false;
+ }
+
+ return self.group_values[lhs_row]
+ == array.as_primitive::<T>().value(rhs_row);
+ }
+
+ array.is_null(rhs_row)
+ }
+
+ fn append_val(&mut self, array: &ArrayRef, row: usize) {
+ if self.nullable && array.is_null(row) {
+ self.group_values.push(T::default_value());
+ self.nulls.push(false);
+ self.has_null = true;
+ } else {
+ let elem = array.as_primitive::<T>().value(row);
+ self.group_values.push(elem);
+ self.nulls.push(true);
+ }
+ }
+
+ fn len(&self) -> usize {
+ self.group_values.len()
+ }
+
+ fn size(&self) -> usize {
+ self.group_values.allocated_size() + self.nulls.allocated_size()
+ }
+
+ fn build(self: Box<Self>) -> ArrayRef {
+ if self.has_null {
+ Arc::new(PrimitiveArray::<T>::new(
+ ScalarBuffer::from(self.group_values),
+ Some(NullBuffer::from(self.nulls)),
+ ))
+ } else {
+ Arc::new(PrimitiveArray::<T>::new(
+ ScalarBuffer::from(self.group_values),
+ None,
+ ))
+ }
+ }
+
+ fn take_n(&mut self, n: usize) -> ArrayRef {
+ if self.has_null {
+ let first_n = self.group_values.drain(0..n).collect::<Vec<_>>();
+ let first_n_nulls = self.nulls.drain(0..n).collect::<Vec<_>>();
+ Arc::new(PrimitiveArray::<T>::new(
+ ScalarBuffer::from(first_n),
+ Some(NullBuffer::from(first_n_nulls)),
+ ))
+ } else {
+ let first_n = self.group_values.drain(0..n).collect::<Vec<_>>();
+ self.nulls.truncate(self.nulls.len() - n);
+ Arc::new(PrimitiveArray::<T>::new(ScalarBuffer::from(first_n),
None))
+ }
+ }
+}
+
+pub struct ByteGroupValueBuilder<O>
+where
+ O: OffsetSizeTrait,
+{
+ output_type: OutputType,
+ buffer: BufferBuilder<u8>,
+ /// Offsets into `buffer` for each distinct value. These offsets as used
+ /// directly to create the final `GenericBinaryArray`. The `i`th string is
+ /// stored in the range `offsets[i]..offsets[i+1]` in `buffer`. Null values
+ /// are stored as a zero length string.
+ offsets: Vec<O>,
+ /// Null indexes in offsets, if `i` is in nulls, `offsets[i]` should be
equals to `offsets[i+1]`
+ nulls: Vec<usize>,
Review Comment:
Is there a reason to use Vec<usize> rather than `Vec<bool>` as used in
`PrimitiveGroupValueBuilder`?
Maybe as a follow on PR we can explore using
[`BooleanBufferBuilder`](https://docs.rs/datafusion/latest/datafusion/common/arrow/array/struct.BooleanBufferBuilder.html)
from arrow-rs directly
##########
datafusion/physical-plan/src/aggregates/group_values/column_wise.rs:
##########
@@ -0,0 +1,314 @@
+// 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::aggregates::group_values::group_value_row::{
+ ArrayRowEq, ByteGroupValueBuilder, PrimitiveGroupValueBuilder,
+};
+use crate::aggregates::group_values::GroupValues;
+use ahash::RandomState;
+use arrow::compute::cast;
+use arrow::datatypes::{
+ Date32Type, Date64Type, Float32Type, Float64Type, Int16Type, Int32Type,
Int64Type,
+ Int8Type, UInt16Type, UInt32Type, UInt64Type, UInt8Type,
+};
+use arrow::record_batch::RecordBatch;
+use arrow_array::{Array, ArrayRef};
+use arrow_schema::{DataType, SchemaRef};
+use datafusion_common::hash_utils::create_hashes;
+use datafusion_common::{DataFusionError, Result};
+use datafusion_execution::memory_pool::proxy::{RawTableAllocExt, VecAllocExt};
+use datafusion_expr::EmitTo;
+use datafusion_physical_expr::binary_map::OutputType;
+
+use hashbrown::raw::RawTable;
+
+/// Compare GroupValue Rows column by column
+pub struct GroupValuesColumn {
+ /// The output schema
+ schema: SchemaRef,
+
+ /// Logically maps group values to a group_index in
+ /// [`Self::group_values`] and in each accumulator
+ ///
+ /// Uses the raw API of hashbrown to avoid actually storing the
+ /// keys (group values) in the table
+ ///
+ /// keys: u64 hashes of the GroupValue
+ /// values: (hash, group_index)
+ map: RawTable<(u64, usize)>,
+
+ /// The size of `map` in bytes
+ map_size: usize,
+
+ /// The actual group by values, stored column-wise. Compare from
+ /// the left to right, each column is stored as `ArrayRowEq`.
+ /// This is shown faster than the row format
+ group_values: Vec<Box<dyn ArrayRowEq>>,
+
+ /// reused buffer to store hashes
+ hashes_buffer: Vec<u64>,
+
+ /// Random state for creating hashes
+ random_state: RandomState,
+}
+
+impl GroupValuesColumn {
+ pub fn try_new(schema: SchemaRef) -> Result<Self> {
+ let map = RawTable::with_capacity(0);
+ Ok(Self {
+ schema,
+ map,
+ map_size: 0,
+ group_values: vec![],
+ hashes_buffer: Default::default(),
+ random_state: Default::default(),
+ })
+ }
+}
+
+impl GroupValues for GroupValuesColumn {
+ fn intern(&mut self, cols: &[ArrayRef], groups: &mut Vec<usize>) ->
Result<()> {
+ let n_rows = cols[0].len();
+
+ if self.group_values.is_empty() {
+ let mut v = Vec::with_capacity(cols.len());
+
+ for f in self.schema.fields().iter() {
+ let nullable = f.is_nullable();
+ match f.data_type() {
+ &DataType::Int8 => {
+ let b =
PrimitiveGroupValueBuilder::<Int8Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Int16 => {
+ let b =
PrimitiveGroupValueBuilder::<Int16Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Int32 => {
+ let b =
PrimitiveGroupValueBuilder::<Int32Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Int64 => {
+ let b =
PrimitiveGroupValueBuilder::<Int64Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::UInt8 => {
+ let b =
PrimitiveGroupValueBuilder::<UInt8Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::UInt16 => {
+ let b =
PrimitiveGroupValueBuilder::<UInt16Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::UInt32 => {
+ let b =
PrimitiveGroupValueBuilder::<UInt32Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::UInt64 => {
+ let b =
PrimitiveGroupValueBuilder::<UInt64Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Float32 => {
+ let b =
PrimitiveGroupValueBuilder::<Float32Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Float64 => {
+ let b =
PrimitiveGroupValueBuilder::<Float64Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Date32 => {
+ let b =
PrimitiveGroupValueBuilder::<Date32Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Date64 => {
+ let b =
PrimitiveGroupValueBuilder::<Date64Type>::new(nullable);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Utf8 => {
+ let b =
ByteGroupValueBuilder::<i32>::new(OutputType::Utf8);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::LargeUtf8 => {
+ let b =
ByteGroupValueBuilder::<i64>::new(OutputType::Utf8);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::Binary => {
+ let b =
ByteGroupValueBuilder::<i32>::new(OutputType::Binary);
+ v.push(Box::new(b) as _)
+ }
+ &DataType::LargeBinary => {
+ let b =
ByteGroupValueBuilder::<i64>::new(OutputType::Binary);
+ v.push(Box::new(b) as _)
+ }
+ dt => todo!("{dt} not impl"),
Review Comment:
It would be nice if this was an internal error rather than a panic
##########
datafusion/sqllogictest/test_files/group_by.slt:
##########
@@ -5148,3 +5148,42 @@ NULL
statement ok
drop table test_case_expr
+
+statement ok
+drop table t;
+
+# test multi group by for binary type with nulls
+statement ok
+create table t(a int, b bytea) as values (1, 0xa), (1, 0xa), (2, null), (null,
0xb), (null, 0xb);
+
+query I?I
+select a, b, count(*) from t group by grouping sets ((a, b), (a), (b));
Review Comment:
That appears to be added above by @jayzhan211
--
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]
