This commit is contained in:
Andrey Mironov 2015-04-08 19:38:38 +03:00
parent c67d578e3d
commit 1e794acb87
4 changed files with 183 additions and 16 deletions

View File

@ -5,6 +5,8 @@
#include <DB/Core/Row.h>
#include <DB/Core/ColumnNumbers.h>
#include <DB/DataStreams/MergingSortedBlockInputStream.h>
#include <statdaemons/ext/range.hpp>
#include <DB/Storages/MergeTree/PKCondition.h>
namespace DB
@ -60,6 +62,14 @@ private:
Names column_names_to_sum; /// Если задано - преобразуется в column_numbers_to_sum при инициализации.
ColumnNumbers column_numbers_to_sum;
struct map_description
{
std::size_t key_col_num;
std::size_t val_col_num;
};
std::vector<map_description> maps_to_sum;
Row current_key; /// Текущий первичный ключ.
Row next_key; /// Первичный ключ следующей строки.
@ -97,12 +107,96 @@ private:
bool operator() (Array & x) const { throw Exception("Cannot sum Arrays", ErrorCodes::LOGICAL_ERROR); }
};
using map_merge_t = std::map<Field, Field>;
class MapSumVisitor : public StaticVisitor<void>
{
public:
map_merge_t & map;
const Field & key;
public:
MapSumVisitor(map_merge_t & map, const Field & key) : map(map), key(key) {}
void operator()(const UInt64 val) const
{
const auto it = map.find(key);
if (it == std::end(map))
map.emplace(key, Field{val});
else
it->second.get<UInt64>() += val;
}
void operator()(const Int64 val) const
{
const auto it = map.find(key);
if (it == std::end(map))
map.emplace(key, Field{val});
else
it->second.get<Int64>() += val;
}
void operator()(const Float64 val) const
{
const auto it = map.find(key);
if (it == std::end(map))
map.emplace(key, Field{val});
else
it->second.get<Float64>() += val;
}
void operator() (Null) const { throw Exception("Cannot merge Nulls", ErrorCodes::LOGICAL_ERROR); }
void operator() (String) const { throw Exception("Cannot merge Strings", ErrorCodes::LOGICAL_ERROR); }
void operator() (Array) const { throw Exception("Cannot merge Arrays", ErrorCodes::LOGICAL_ERROR); }
};
/** Прибавить строчку под курсором к row.
* Возвращает false, если результат получился нулевым.
*/
template<class TSortCursor>
bool addRow(Row & row, TSortCursor & cursor)
{
/// merge maps
for (const auto & map : maps_to_sum)
{
auto & key_array = row[map.key_col_num].get<Array>();
auto & val_array = row[map.val_col_num].get<Array>();
if (key_array.size() != val_array.size())
throw Exception{"Nested arrays have different sizes", ErrorCodes::LOGICAL_ERROR};
const auto key_field_rhs = (*cursor->all_columns[map.key_col_num])[cursor->pos];
const auto val_field_rhs = (*cursor->all_columns[map.val_col_num])[cursor->pos];
const auto & key_array_rhs = key_field.get<Array>();
const auto & val_array_rhs = val_field.get<Array>();
if (key_array_rhs.size() != val_array_rhs.size())
throw Exception{"Nested arrays have different sizes", ErrorCodes::LOGICAL_ERROR};
map_merge_t result;
/// merge accumulator-row and current row from cursor using std::map
for (const auto i : ext::range(0, key_array.size()))
apply_visitor(MapSumVisitor{result, key_array[i]}, val_array[i]);
for (const auto i : ext::range(0, key_array_rhs.size()))
apply_visitor(MapSumVisitor{result, key_array_rhs[i]}, val_array_rhs[i]);
Array key_array_result;
Array val_array_result;
/// skip items with value == 0
for (const auto & pair : result)
{
if (!apply_visitor(FieldVisitorAccurateEquals{}, pair.second, nearestFieldType(0)))
{
key_array_result.emplace_back(pair.first);
val_array_result.emplace_back(pair.second);
}
}
key_array = std::move(key_array_result);
val_array = std::move(val_array_result);
}
bool res = false; /// Есть ли хотя бы одно ненулевое число.
for (size_t i = 0, size = column_numbers_to_sum.size(); i < size; ++i)

View File

@ -1,4 +1,6 @@
#include <DB/DataStreams/SummingSortedBlockInputStream.h>
#include <DB/DataTypes/DataTypeNested.h>
#include <DB/DataTypes/DataTypeArray.h>
namespace DB
@ -12,6 +14,24 @@ void SummingSortedBlockInputStream::insertCurrentRow(ColumnPlainPtrs & merged_co
}
namespace
{
bool endsWith(const std::string & s, const std::string & suffix)
{
return s.size() >= suffix.size() && 0 == strncmp(s.data() + s.size() - suffix.size(), suffix.data(), suffix.size());
}
bool isInPrimaryKey(const SortDescription & description, const std::string & name, const std::size_t number)
{
for (auto & desc : description)
if (desc.column_name == name || (desc.column_name.empty() && desc.column_number == number))
return true;
return false;
}
}
Block SummingSortedBlockInputStream::readImpl()
{
if (finished)
@ -34,6 +54,7 @@ Block SummingSortedBlockInputStream::readImpl()
current_key.resize(description.size());
next_key.resize(description.size());
std::unordered_map<std::string, std::vector<std::size_t>> discovered_maps;
/** Заполним номера столбцов, которые должны быть просуммированы.
* Это могут быть только числовые столбцы, не входящие в ключ сортировки.
* Если задан непустой список column_names_to_sum, то берём только эти столбцы.
@ -43,24 +64,62 @@ Block SummingSortedBlockInputStream::readImpl()
{
ColumnWithNameAndType & column = merged_block.getByPosition(i);
/// Оставляем только числовые типы. При чём, даты и даты-со-временем здесь такими не считаются.
if (!column.type->isNumeric() || column.type->getName() == "Date" || column.type->getName() == "DateTime")
continue;
/// Входят ли в PK?
SortDescription::const_iterator it = description.begin();
for (; it != description.end(); ++it)
if (it->column_name == column.name || (it->column_name.empty() && it->column_number == i))
break;
if (it != description.end())
continue;
if (column_names_to_sum.empty()
|| column_names_to_sum.end() != std::find(column_names_to_sum.begin(), column_names_to_sum.end(), column.name))
if (const auto array_type = typeid_cast<const DataTypeArray *>(column.type.get()))
{
column_numbers_to_sum.push_back(i);
const auto map_name = DataTypeNested::extractNestedTableName(column.name);
if (map_name == column.name || !endsWith(map_name, "Map"))
continue;
discovered_maps[map_name].emplace_back(i);
}
else
{
/// Оставляем только числовые типы. При чём, даты и даты-со-временем здесь такими не считаются.
if (!column.type->isNumeric() || column.type->getName() == "Date" ||
column.type->getName() == "DateTime")
continue;
/// Входят ли в PK?
if (isInPrimaryKey(description, column.name, i))
continue;
if (column_names_to_sum.empty()
|| column_names_to_sum.end() !=
std::find(column_names_to_sum.begin(), column_names_to_sum.end(), column.name))
{
column_numbers_to_sum.push_back(i);
}
}
}
for (const auto & map : discovered_maps)
{
/// map can only contain a pair of elements (key -> value)
if (map.second.size() > 2)
continue;
/// check types of key and value
const auto key_num = map.second.front();
auto & key_col = merged_block.getByPosition(key_num);
if (isInPrimaryKey(description, key_col.name, key_num))
continue;
auto & key_nested_type = static_cast<const DataTypeArray *>(key_col.type.get())->getNestedType();
/// key can only be integral
if (!key_nested_type->isNumeric() || key_nested_type->getName() == "Float32" || key_nested_type->getName() == "Float64")
continue;
const auto value_num = map.second.back();
auto & value_col = merged_block.getByPosition(value_num);
if (isInPrimaryKey(description, value_col.name, value_num))
continue;
auto & value_nested_type = static_cast<const DataTypeArray *>(value_col.type.get())->getNestedType();
/// value can be any arithmetic type except date and datetime
if (!value_nested_type->isNumeric() || value_nested_type->getName() == "Date" || value_nested_type->getName() == "DateTime")
continue;
maps_to_sum.push_back({ key_num, value_num });
}
}

View File

@ -0,0 +1,4 @@
[1,2] [100,150]
[1] [250]
[1,2] [250,150]
[2] [150]

View File

@ -0,0 +1,10 @@
drop table if exists test.nested_map;
create table test.nested_map (d default today(), k UInt64, payload default rand(), SomeMap Nested(ID UInt32, Num Int64)) engine=SummingMergeTree(d, k, 8192);
insert into test.nested_map (k, `SomeMap.ID`, `SomeMap.Num`) values (0,[1],[100]),(1,[1],[100]),(2,[1],[100]),(3,[1,2],[100,150]);
insert into test.nested_map (k, `SomeMap.ID`, `SomeMap.Num`) values (0,[2],[150]),(1,[1],[150]),(2,[1,2],[150,150]),(3,[1],[-100]);
optimize table test.nested_map;
select `SomeMap.ID`, `SomeMap.Num` from test.nested_map;
drop table test.nested_map;