mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-09-20 08:40:50 +00:00
Make ORDER BY tuple almost as fast as ORDER BY columns
We have special optimizations for multiple column ORDER BY: https://github.com/ClickHouse/ClickHouse/pull/10831 . It's beneficial to also apply to tuple columns. Before: select * from numbers(300000000) order by (1 - number , number + 1 , number) limit 10; 2.613 sec. After: select * from numbers(300000000) order by (1 - number , number + 1 , number) limit 10; 0.755 sec No tuple: select * from numbers(300000000) order by 1 - number , number + 1 , number limit 10; 0.755 sec
This commit is contained in:
parent
b9fb741c79
commit
faee95b897
@ -4,6 +4,7 @@
|
|||||||
#include <Columns/ColumnConst.h>
|
#include <Columns/ColumnConst.h>
|
||||||
#include <Columns/ColumnNullable.h>
|
#include <Columns/ColumnNullable.h>
|
||||||
#include <Columns/ColumnLowCardinality.h>
|
#include <Columns/ColumnLowCardinality.h>
|
||||||
|
#include <Columns/ColumnTuple.h>
|
||||||
#include <Common/typeid_cast.h>
|
#include <Common/typeid_cast.h>
|
||||||
#include <Functions/FunctionHelpers.h>
|
#include <Functions/FunctionHelpers.h>
|
||||||
|
|
||||||
@ -22,6 +23,28 @@ static bool isCollationRequired(const SortColumnDescription & description)
|
|||||||
return description.collator != nullptr;
|
return description.collator != nullptr;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
/// Column with description for sort
|
||||||
|
struct ColumnWithSortDescription
|
||||||
|
{
|
||||||
|
const IColumn * column = nullptr;
|
||||||
|
const SortColumnDescription * description = nullptr;
|
||||||
|
|
||||||
|
/// It means, that this column is ColumnConst
|
||||||
|
bool column_const = false;
|
||||||
|
};
|
||||||
|
using ColumnsWithSortDescriptions = std::vector<ColumnWithSortDescription>;
|
||||||
|
|
||||||
|
void flattenTupleColumnRecursively(
|
||||||
|
ColumnsWithSortDescriptions & res, const ColumnTuple * tuple, const SortColumnDescription * description, bool is_constant)
|
||||||
|
{
|
||||||
|
for (const auto & column : tuple->getColumns())
|
||||||
|
{
|
||||||
|
if (const auto * subtuple = typeid_cast<const ColumnTuple *>(column.get()))
|
||||||
|
flattenTupleColumnRecursively(res, subtuple, description, is_constant);
|
||||||
|
else
|
||||||
|
res.emplace_back(ColumnWithSortDescription{column.get(), description, is_constant});
|
||||||
|
}
|
||||||
|
}
|
||||||
|
|
||||||
ColumnsWithSortDescriptions getColumnsWithSortDescription(const Block & block, const SortDescription & description)
|
ColumnsWithSortDescriptions getColumnsWithSortDescription(const Block & block, const SortDescription & description)
|
||||||
{
|
{
|
||||||
@ -35,13 +58,14 @@ ColumnsWithSortDescriptions getColumnsWithSortDescription(const Block & block, c
|
|||||||
? block.getByName(description[i].column_name).column.get()
|
? block.getByName(description[i].column_name).column.get()
|
||||||
: block.safeGetByPosition(description[i].column_number).column.get();
|
: block.safeGetByPosition(description[i].column_number).column.get();
|
||||||
|
|
||||||
res.emplace_back(ColumnWithSortDescription{column, description[i], isColumnConst(*column)});
|
if (const auto * tuple = typeid_cast<const ColumnTuple *>(column))
|
||||||
|
flattenTupleColumnRecursively(res, tuple, &description[i], isColumnConst(*column));
|
||||||
|
else
|
||||||
|
res.emplace_back(ColumnWithSortDescription{column, &description[i], isColumnConst(*column)});
|
||||||
}
|
}
|
||||||
|
|
||||||
return res;
|
return res;
|
||||||
}
|
}
|
||||||
|
|
||||||
|
|
||||||
struct PartialSortingLess
|
struct PartialSortingLess
|
||||||
{
|
{
|
||||||
const ColumnsWithSortDescriptions & columns;
|
const ColumnsWithSortDescriptions & columns;
|
||||||
@ -56,7 +80,7 @@ struct PartialSortingLess
|
|||||||
if (elem.column_const)
|
if (elem.column_const)
|
||||||
res = 0;
|
res = 0;
|
||||||
else
|
else
|
||||||
res = elem.description.direction * elem.column->compareAt(a, b, *elem.column, elem.description.nulls_direction);
|
res = elem.description->direction * elem.column->compareAt(a, b, *elem.column, elem.description->nulls_direction);
|
||||||
if (res < 0)
|
if (res < 0)
|
||||||
return true;
|
return true;
|
||||||
else if (res > 0)
|
else if (res > 0)
|
||||||
@ -86,13 +110,13 @@ struct PartialSortingLessWithCollation
|
|||||||
{
|
{
|
||||||
res = 0;
|
res = 0;
|
||||||
}
|
}
|
||||||
else if (isCollationRequired(elem.description))
|
else if (isCollationRequired(*elem.description))
|
||||||
{
|
{
|
||||||
res = elem.column->compareAtWithCollation(a, b, *elem.column, elem.description.nulls_direction, *elem.description.collator);
|
res = elem.column->compareAtWithCollation(a, b, *elem.column, elem.description->nulls_direction, *elem.description->collator);
|
||||||
}
|
}
|
||||||
else
|
else
|
||||||
res = elem.column->compareAt(a, b, *elem.column, elem.description.nulls_direction);
|
res = elem.column->compareAt(a, b, *elem.column, elem.description->nulls_direction);
|
||||||
res *= elem.description.direction;
|
res *= elem.description->direction;
|
||||||
if (res < 0)
|
if (res < 0)
|
||||||
return true;
|
return true;
|
||||||
else if (res > 0)
|
else if (res > 0)
|
||||||
@ -107,47 +131,35 @@ void sortBlock(Block & block, const SortDescription & description, UInt64 limit)
|
|||||||
if (!block)
|
if (!block)
|
||||||
return;
|
return;
|
||||||
|
|
||||||
|
ColumnsWithSortDescriptions columns_with_sort_desc = getColumnsWithSortDescription(block, description);
|
||||||
|
IColumn::Permutation perm;
|
||||||
/// If only one column to sort by
|
/// If only one column to sort by
|
||||||
if (description.size() == 1)
|
if (columns_with_sort_desc.size() == 1)
|
||||||
{
|
{
|
||||||
IColumn::Permutation perm;
|
bool reverse = columns_with_sort_desc[0].description->direction == -1;
|
||||||
bool reverse = description[0].direction == -1;
|
const IColumn * column = columns_with_sort_desc[0].column;
|
||||||
|
bool is_column_const = columns_with_sort_desc[0].column_const;
|
||||||
const IColumn * column = !description[0].column_name.empty()
|
|
||||||
? block.getByName(description[0].column_name).column.get()
|
|
||||||
: block.safeGetByPosition(description[0].column_number).column.get();
|
|
||||||
|
|
||||||
bool is_column_const = false;
|
|
||||||
if (isCollationRequired(description[0]))
|
if (isCollationRequired(description[0]))
|
||||||
{
|
{
|
||||||
if (!column->isCollationSupported())
|
if (!column->isCollationSupported())
|
||||||
throw Exception("Collations could be specified only for String, LowCardinality(String), Nullable(String) or for Array or Tuple, containing them.", ErrorCodes::BAD_COLLATION);
|
throw Exception(
|
||||||
|
"Collations could be specified only for String, LowCardinality(String), Nullable(String) or for Array or Tuple, "
|
||||||
|
"containing them.",
|
||||||
|
ErrorCodes::BAD_COLLATION);
|
||||||
|
|
||||||
if (isColumnConst(*column))
|
if (!is_column_const)
|
||||||
is_column_const = true;
|
|
||||||
else
|
|
||||||
column->getPermutationWithCollation(*description[0].collator, reverse, limit, description[0].nulls_direction, perm);
|
column->getPermutationWithCollation(*description[0].collator, reverse, limit, description[0].nulls_direction, perm);
|
||||||
}
|
}
|
||||||
else if (!isColumnConst(*column))
|
else if (!is_column_const)
|
||||||
{
|
{
|
||||||
int nan_direction_hint = description[0].nulls_direction;
|
int nan_direction_hint = columns_with_sort_desc[0].description->nulls_direction;
|
||||||
column->getPermutation(reverse, limit, nan_direction_hint, perm);
|
column->getPermutation(reverse, limit, nan_direction_hint, perm);
|
||||||
}
|
}
|
||||||
else
|
|
||||||
/// we don't need to do anything with const column
|
|
||||||
is_column_const = true;
|
|
||||||
|
|
||||||
size_t columns = block.columns();
|
|
||||||
for (size_t i = 0; i < columns; ++i)
|
|
||||||
{
|
|
||||||
if (!is_column_const)
|
|
||||||
block.getByPosition(i).column = block.getByPosition(i).column->permute(perm, limit);
|
|
||||||
}
|
|
||||||
}
|
}
|
||||||
else
|
else
|
||||||
{
|
{
|
||||||
size_t size = block.rows();
|
size_t size = block.rows();
|
||||||
IColumn::Permutation perm(size);
|
perm.resize(size);
|
||||||
for (size_t i = 0; i < size; ++i)
|
for (size_t i = 0; i < size; ++i)
|
||||||
perm[i] = i;
|
perm[i] = i;
|
||||||
|
|
||||||
@ -155,16 +167,15 @@ void sortBlock(Block & block, const SortDescription & description, UInt64 limit)
|
|||||||
limit = 0;
|
limit = 0;
|
||||||
|
|
||||||
bool need_collation = false;
|
bool need_collation = false;
|
||||||
ColumnsWithSortDescriptions columns_with_sort_desc = getColumnsWithSortDescription(block, description);
|
for (const auto & column : columns_with_sort_desc)
|
||||||
|
|
||||||
for (size_t i = 0, num_sort_columns = description.size(); i < num_sort_columns; ++i)
|
|
||||||
{
|
{
|
||||||
const IColumn * column = columns_with_sort_desc[i].column;
|
if (isCollationRequired(*column.description))
|
||||||
if (isCollationRequired(description[i]))
|
|
||||||
{
|
{
|
||||||
if (!column->isCollationSupported())
|
if (!column.column->isCollationSupported())
|
||||||
throw Exception("Collations could be specified only for String, LowCardinality(String), Nullable(String) or for Array or Tuple, containing them.", ErrorCodes::BAD_COLLATION);
|
throw Exception(
|
||||||
|
"Collations could be specified only for String, LowCardinality(String), Nullable(String) or for Array or Tuple, "
|
||||||
|
"containing them.",
|
||||||
|
ErrorCodes::BAD_COLLATION);
|
||||||
need_collation = true;
|
need_collation = true;
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
@ -184,15 +195,20 @@ void sortBlock(Block & block, const SortDescription & description, UInt64 limit)
|
|||||||
if (column.column_const)
|
if (column.column_const)
|
||||||
continue;
|
continue;
|
||||||
|
|
||||||
if (isCollationRequired(column.description))
|
if (isCollationRequired(*column.description))
|
||||||
{
|
{
|
||||||
column.column->updatePermutationWithCollation(
|
column.column->updatePermutationWithCollation(
|
||||||
*column.description.collator, column.description.direction < 0, limit, column.description.nulls_direction, perm, ranges);
|
*column.description->collator,
|
||||||
|
column.description->direction < 0,
|
||||||
|
limit,
|
||||||
|
column.description->nulls_direction,
|
||||||
|
perm,
|
||||||
|
ranges);
|
||||||
}
|
}
|
||||||
else
|
else
|
||||||
{
|
{
|
||||||
column.column->updatePermutation(
|
column.column->updatePermutation(
|
||||||
column.description.direction < 0, limit, column.description.nulls_direction, perm, ranges);
|
column.description->direction < 0, limit, column.description->nulls_direction, perm, ranges);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
@ -209,12 +225,15 @@ void sortBlock(Block & block, const SortDescription & description, UInt64 limit)
|
|||||||
break;
|
break;
|
||||||
|
|
||||||
column.column->updatePermutation(
|
column.column->updatePermutation(
|
||||||
column.description.direction < 0, limit, column.description.nulls_direction, perm, ranges);
|
column.description->direction < 0, limit, column.description->nulls_direction, perm, ranges);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
}
|
||||||
|
|
||||||
size_t columns = block.columns();
|
size_t columns = block.columns();
|
||||||
for (size_t i = 0; i < columns; ++i)
|
for (size_t i = 0; i < columns; ++i)
|
||||||
|
{
|
||||||
|
if (!isColumnConst(*block.getByPosition(i).column))
|
||||||
block.getByPosition(i).column = block.getByPosition(i).column->permute(perm, limit);
|
block.getByPosition(i).column = block.getByPosition(i).column->permute(perm, limit);
|
||||||
}
|
}
|
||||||
}
|
}
|
||||||
|
@ -29,18 +29,4 @@ void stableGetPermutation(const Block & block, const SortDescription & descripti
|
|||||||
*/
|
*/
|
||||||
bool isAlreadySorted(const Block & block, const SortDescription & description);
|
bool isAlreadySorted(const Block & block, const SortDescription & description);
|
||||||
|
|
||||||
/// Column with description for sort
|
|
||||||
struct ColumnWithSortDescription
|
|
||||||
{
|
|
||||||
const IColumn * column = nullptr;
|
|
||||||
SortColumnDescription description;
|
|
||||||
|
|
||||||
/// It means, that this column is ColumnConst
|
|
||||||
bool column_const = false;
|
|
||||||
};
|
|
||||||
|
|
||||||
using ColumnsWithSortDescriptions = std::vector<ColumnWithSortDescription>;
|
|
||||||
|
|
||||||
ColumnsWithSortDescriptions getColumnsWithSortDescription(const Block & block, const SortDescription & description);
|
|
||||||
|
|
||||||
}
|
}
|
||||||
|
8
tests/performance/order_by_tuple.xml
Normal file
8
tests/performance/order_by_tuple.xml
Normal file
@ -0,0 +1,8 @@
|
|||||||
|
<test>
|
||||||
|
<tags>
|
||||||
|
<tag>sorting</tag>
|
||||||
|
<tag>comparison</tag>
|
||||||
|
</tags>
|
||||||
|
|
||||||
|
<query>select * from numbers(300000000) order by (1 - number , number + 1 , number) limit 10;</query>
|
||||||
|
</test>
|
Loading…
Reference in New Issue
Block a user