mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-11-16 20:53:27 +00:00
312 lines
10 KiB
C++
312 lines
10 KiB
C++
#pragma once
|
|
|
|
#include <DB/Storages/MergeTree/MergeTreeData.h>
|
|
#include <DB/Parsers/ASTSelectQuery.h>
|
|
#include <DB/Parsers/ASTFunction.h>
|
|
#include <DB/Parsers/ASTIdentifier.h>
|
|
#include <DB/Parsers/ASTLiteral.h>
|
|
#include <DB/Parsers/ASTExpressionList.h>
|
|
#include <DB/Parsers/ASTSubquery.h>
|
|
#include <DB/Parsers/formatAST.h>
|
|
#include <DB/Common/escapeForFileName.h>
|
|
#include <statdaemons/ext/memory.hpp>
|
|
#include <unordered_map>
|
|
#include <map>
|
|
#include <limits>
|
|
#include <cstddef>
|
|
|
|
namespace DB
|
|
{
|
|
|
|
|
|
/** Identifies WHERE expressions that can be placed in PREWHERE by calculating respective
|
|
* sizes of columns used in particular expression and identifying "good" conditions of
|
|
* form "column_name = constant", where "constant" is outside some `threshold` specified in advance.
|
|
*
|
|
* If there are "good" conditions present in WHERE, the one with minimal summary column size is
|
|
* transferred to PREWHERE.
|
|
* Otherwise any condition with minimal summary column size can be transferred to PREWHERE, if only
|
|
* its relative size (summary column size divided by query column size) is less than `max_columns_relative_size`.
|
|
*/
|
|
class MergeTreeWhereOptimizer
|
|
{
|
|
static constexpr auto threshold = 10;
|
|
static constexpr auto max_columns_relative_size = 0.25f;
|
|
static constexpr auto and_function_name = "and";
|
|
static constexpr auto equals_function_name = "equals";
|
|
|
|
public:
|
|
MergeTreeWhereOptimizer(const MergeTreeWhereOptimizer&) = delete;
|
|
MergeTreeWhereOptimizer& operator=(const MergeTreeWhereOptimizer&) = delete;
|
|
|
|
MergeTreeWhereOptimizer(
|
|
ASTSelectQuery & select, const MergeTreeData & data,
|
|
const Names & column_names, Logger * log)
|
|
: primary_key_columns{toUnorderedSet(data.getPrimaryExpression()->getRequiredColumnsWithTypes())},
|
|
table_columns{toUnorderedSet(data.getColumnsList())}, log{log}
|
|
{
|
|
calculateColumnSizes(data, column_names);
|
|
optimize(select);
|
|
}
|
|
|
|
private:
|
|
void optimize(ASTSelectQuery & select) const
|
|
{
|
|
if (!select.where_expression || select.prewhere_expression)
|
|
return;
|
|
|
|
const auto function = typeid_cast<ASTFunction *>(select.where_expression.get());
|
|
if (function && function->name == and_function_name)
|
|
optimizeConjunction(select, function);
|
|
else
|
|
optimizeArbitrary(select);
|
|
}
|
|
|
|
void calculateColumnSizes(const MergeTreeData & data, const Names & column_names)
|
|
{
|
|
for (const auto & column_name : column_names)
|
|
{
|
|
const auto column_size = data.getColumnSize(column_name);
|
|
|
|
column_sizes[column_name] = column_size;
|
|
total_column_size += column_size;
|
|
}
|
|
}
|
|
|
|
void optimizeConjunction(ASTSelectQuery & select, ASTFunction * const fun) const
|
|
{
|
|
/// used as max possible size and indicator that appropriate condition has not been found
|
|
const auto no_such_condition = std::numeric_limits<std::size_t>::max();
|
|
|
|
/// { first: condition index, second: summary column size }
|
|
std::pair<std::size_t, std::size_t> lightest_good_condition{no_such_condition, no_such_condition};
|
|
std::pair<std::size_t, std::size_t> lightest_viable_condition{no_such_condition, no_such_condition};
|
|
|
|
auto & conditions = fun->arguments->children;
|
|
|
|
/// remove condition by swapping it with the last one and calling ::pop_back()
|
|
const auto remove_condition_at_index = [&conditions] (const std::size_t idx) {
|
|
if (idx < conditions.size() - 1)
|
|
std::swap(conditions[idx], conditions.back());
|
|
conditions.pop_back();
|
|
};
|
|
|
|
/// linearize conjunction and divide conditions into "good" and not-"good" ones
|
|
for (std::size_t idx = 0; idx < conditions.size();)
|
|
{
|
|
const auto condition = conditions[idx].get();
|
|
|
|
IdentifierNameSet identifiers{};
|
|
collectIdentifiersNoSubqueries(condition, identifiers);
|
|
|
|
/// do not take into consideration the conditions consisting only of primary key columns
|
|
if (hasNonPrimaryKeyColumns(identifiers) && isSubsetOfTableColumns(identifiers))
|
|
{
|
|
/// linearize sub-conjunctions
|
|
if (const auto function = typeid_cast<ASTFunction *>(condition))
|
|
{
|
|
if (function->name == and_function_name)
|
|
{
|
|
for (auto & child : function->arguments->children)
|
|
conditions.emplace_back(std::move(child));
|
|
|
|
/// remove the condition corresponding to conjunction
|
|
remove_condition_at_index(idx);
|
|
|
|
/// continue iterating without increment to ensure the just added conditions are processed
|
|
continue;
|
|
}
|
|
}
|
|
|
|
/// calculate size of columns involved in condition
|
|
const auto cond_columns_size = getIdentifiersColumnSize(identifiers);
|
|
|
|
/// place condition either in good or viable conditions set
|
|
auto & good_or_viable_condition = isConditionGood(condition) ? lightest_good_condition : lightest_viable_condition;
|
|
if (good_or_viable_condition.second > cond_columns_size)
|
|
{
|
|
good_or_viable_condition.first = idx;
|
|
good_or_viable_condition.second = cond_columns_size;
|
|
}
|
|
}
|
|
|
|
++idx;
|
|
}
|
|
|
|
const auto move_condition_to_prewhere = [&] (const std::size_t idx) {
|
|
select.prewhere_expression = conditions[idx];
|
|
select.children.push_back(select.prewhere_expression);
|
|
LOG_DEBUG(log, "MergeTreeWhereOptimizer: condition `" << select.prewhere_expression << "` moved to PREWHERE");
|
|
|
|
/** Replace conjunction with the only remaining argument if only two conditions were present,
|
|
* remove selected condition from conjunction otherwise. */
|
|
if (conditions.size() == 2)
|
|
{
|
|
/// find old where_expression in children of select
|
|
const auto it = std::find(std::begin(select.children), std::end(select.children), select.where_expression);
|
|
/// replace where_expression with the remaining argument
|
|
select.where_expression = std::move(conditions[idx == 0 ? 1 : 0]);
|
|
/// overwrite child entry with the new where_expression
|
|
*it = select.where_expression;
|
|
}
|
|
else
|
|
remove_condition_at_index(idx);
|
|
};
|
|
|
|
/// if there is a "good" condition - move it to PREWHERE
|
|
if (lightest_good_condition.first != no_such_condition)
|
|
{
|
|
move_condition_to_prewhere(lightest_good_condition.first);
|
|
}
|
|
else if (lightest_viable_condition.first != no_such_condition)
|
|
{
|
|
/// check that the relative column size is less than max
|
|
if (total_column_size != 0)
|
|
{
|
|
/// calculate relative size of condition's columns
|
|
const auto cond_columns_size = lightest_viable_condition.second;
|
|
const auto columns_relative_size = static_cast<float>(cond_columns_size) / total_column_size;
|
|
|
|
/// do nothing if it exceeds max relative size
|
|
if (columns_relative_size > max_columns_relative_size)
|
|
return;
|
|
}
|
|
|
|
move_condition_to_prewhere(lightest_viable_condition.first);
|
|
}
|
|
}
|
|
|
|
void optimizeArbitrary(ASTSelectQuery & select) const
|
|
{
|
|
auto & condition = select.where_expression;
|
|
|
|
IdentifierNameSet identifiers{};
|
|
collectIdentifiersNoSubqueries(condition, identifiers);
|
|
|
|
if (!hasNonPrimaryKeyColumns(identifiers) || !isSubsetOfTableColumns(identifiers))
|
|
return;
|
|
|
|
/// if condition is not "good" - check that it can be moved
|
|
if (!isConditionGood(condition.get()) && total_column_size != 0)
|
|
{
|
|
const auto cond_columns_size = getIdentifiersColumnSize(identifiers);
|
|
const auto columns_relative_size = static_cast<float>(cond_columns_size) / total_column_size;
|
|
|
|
if (columns_relative_size > max_columns_relative_size)
|
|
return;
|
|
}
|
|
|
|
/// add the condition to PREWHERE, remove it from WHERE
|
|
std::swap(select.prewhere_expression, condition);
|
|
LOG_DEBUG(log, "MergeTreeWhereOptimizer: condition `" << select.prewhere_expression << "` moved to PREWHERE");
|
|
}
|
|
|
|
std::size_t getIdentifiersColumnSize(const IdentifierNameSet & identifiers) const
|
|
{
|
|
std::size_t size{};
|
|
|
|
for (const auto & identifier : identifiers)
|
|
if (column_sizes.count(identifier))
|
|
size += column_sizes.find(identifier)->second;
|
|
|
|
return size;
|
|
}
|
|
|
|
bool isConditionGood(const IAST * condition) const
|
|
{
|
|
const auto function = typeid_cast<const ASTFunction *>(condition);
|
|
if (!function)
|
|
return false;
|
|
|
|
/** we are only considering conditions of form `equals(one, another)` or `one = another`,
|
|
* especially if either `one` or `another` is ASTIdentifier */
|
|
if (function->name != equals_function_name)
|
|
return false;
|
|
|
|
auto left_arg = function->arguments->children.front().get();
|
|
auto right_arg = function->arguments->children.back().get();
|
|
|
|
/// try to ensure left_arg points to ASTIdentifier
|
|
if (!typeid_cast<const ASTIdentifier *>(left_arg) && typeid_cast<const ASTIdentifier *>(right_arg))
|
|
std::swap(left_arg, right_arg);
|
|
|
|
if (typeid_cast<const ASTIdentifier *>(left_arg))
|
|
{
|
|
/// condition may be "good" if only right_arg is a constant and its value is outside the threshold
|
|
if (const auto literal = typeid_cast<const ASTLiteral *>(right_arg))
|
|
{
|
|
const auto & field = literal->value;
|
|
const auto type = field.getType();
|
|
|
|
/// check the value with respect to threshold
|
|
if (type == Field::Types::UInt64)
|
|
{
|
|
const auto value = field.get<UInt64>();
|
|
return value > threshold;
|
|
}
|
|
else if (type == Field::Types::Int64)
|
|
{
|
|
const auto value = field.get<Int64>();
|
|
return value < -threshold || threshold < value;
|
|
}
|
|
else if (type == Field::Types::Float64)
|
|
{
|
|
const auto value = field.get<Float64>();
|
|
return value < threshold || threshold < value;
|
|
}
|
|
}
|
|
}
|
|
|
|
return false;
|
|
}
|
|
|
|
static void collectIdentifiersNoSubqueries(const IAST * const ast, IdentifierNameSet & set)
|
|
{
|
|
if (const auto identifier = typeid_cast<const ASTIdentifier *>(ast))
|
|
return (void) set.insert(identifier->name);
|
|
|
|
if (typeid_cast<const ASTSubquery *>(ast))
|
|
return;
|
|
|
|
for (const auto & child : ast->children)
|
|
collectIdentifiersNoSubqueries(child.get(), set);
|
|
}
|
|
|
|
using string_set_t = std::unordered_set<std::string>;
|
|
|
|
static string_set_t toUnorderedSet(const NamesAndTypesList & columns)
|
|
{
|
|
string_set_t result{};
|
|
|
|
for (const auto column : columns)
|
|
result.insert(column.name);
|
|
|
|
return result;
|
|
}
|
|
|
|
bool hasNonPrimaryKeyColumns(const IdentifierNameSet & identifiers) const {
|
|
for (const auto & identifier : identifiers)
|
|
if (primary_key_columns.count(identifier) == 0)
|
|
return true;
|
|
|
|
return false;
|
|
}
|
|
|
|
bool isSubsetOfTableColumns(const IdentifierNameSet & identifiers) const {
|
|
for (const auto & identifier : identifiers)
|
|
if (table_columns.count(identifier) == 0)
|
|
return false;
|
|
|
|
return true;
|
|
}
|
|
|
|
string_set_t primary_key_columns{};
|
|
string_set_t table_columns{};
|
|
Logger * log;
|
|
std::unordered_map<std::string, std::size_t> column_sizes{};
|
|
std::size_t total_column_size{};
|
|
};
|
|
|
|
|
|
}
|