mirror of
https://github.com/ClickHouse/ClickHouse.git
synced 2024-12-17 11:52:27 +00:00
Normal Projections analysis using query plan [In progress]
This commit is contained in:
parent
e6e486c4a1
commit
522a39f93f
@ -93,7 +93,8 @@ using Stack = std::vector<Frame>;
|
||||
void optimizePrimaryKeyCondition(const Stack & stack);
|
||||
void optimizeReadInOrder(QueryPlan::Node & node, QueryPlan::Nodes & nodes);
|
||||
void optimizeAggregationInOrder(QueryPlan::Node & node, QueryPlan::Nodes &);
|
||||
bool optimizeUseProjections(QueryPlan::Node & node, QueryPlan::Nodes &);
|
||||
bool optimizeUseAggProjections(QueryPlan::Node & node, QueryPlan::Nodes & nodes);
|
||||
bool optimizeUseNormalProjections(Stack & stack, QueryPlan::Nodes & nodes);
|
||||
|
||||
/// Enable memory bound merging of aggregation states for remote queries
|
||||
/// in case it was enabled for local plan
|
||||
|
@ -122,7 +122,7 @@ void optimizeTreeSecondPass(const QueryPlanOptimizationSettings & optimization_s
|
||||
optimizeReadInOrder(*frame.node, nodes);
|
||||
|
||||
if (optimization_settings.optimize_projection)
|
||||
applied_projection |= optimizeUseProjections(*frame.node, nodes);
|
||||
applied_projection |= optimizeUseAggProjections(*frame.node, nodes);
|
||||
|
||||
if (optimization_settings.aggregation_in_order)
|
||||
optimizeAggregationInOrder(*frame.node, nodes);
|
||||
@ -140,6 +140,14 @@ void optimizeTreeSecondPass(const QueryPlanOptimizationSettings & optimization_s
|
||||
continue;
|
||||
}
|
||||
|
||||
if (optimization_settings.optimize_projection)
|
||||
{
|
||||
bool applied = optimizeUseNormalProjections(stack, nodes);
|
||||
applied_projection |= applied;
|
||||
if (applied && stack.back().next_child == 0)
|
||||
continue;
|
||||
}
|
||||
|
||||
optimizePrimaryKeyCondition(stack);
|
||||
enableMemoryBoundMerging(*frame.node, nodes);
|
||||
|
||||
|
@ -12,6 +12,7 @@
|
||||
#include <Storages/MergeTree/MergeTreeDataSelectExecutor.h>
|
||||
#include <Storages/StorageReplicatedMergeTree.h>
|
||||
#include <Common/logger_useful.h>
|
||||
#include <Processors/QueryPlan/UnionStep.h>
|
||||
#include <AggregateFunctions/AggregateFunctionCount.h>
|
||||
#include <Processors/QueryPlan/ReadFromPreparedSource.h>
|
||||
#include <Processors/Sources/NullSource.h>
|
||||
@ -167,6 +168,16 @@ struct AggregateProjectionCandidate
|
||||
size_t sum_marks = 0;
|
||||
};
|
||||
|
||||
struct NormalProjectionCandidate
|
||||
{
|
||||
const ProjectionDescription * projection;
|
||||
|
||||
MergeTreeDataSelectAnalysisResultPtr merge_tree_projection_select_result_ptr;
|
||||
MergeTreeDataSelectAnalysisResultPtr merge_tree_normal_select_result_ptr;
|
||||
|
||||
size_t sum_marks = 0;
|
||||
};
|
||||
|
||||
ActionsDAGPtr analyzeAggregateProjection(
|
||||
const AggregateProjectionInfo & info,
|
||||
ActionsDAG & query_dag,
|
||||
@ -430,7 +441,7 @@ ActionsDAGPtr analyzeAggregateProjection(
|
||||
return proj_dag;
|
||||
}
|
||||
|
||||
bool optimizeUseProjections(QueryPlan::Node & node, QueryPlan::Nodes & nodes)
|
||||
bool optimizeUseAggProjections(QueryPlan::Node & node, QueryPlan::Nodes & nodes)
|
||||
{
|
||||
if (node.children.size() != 1)
|
||||
return false;
|
||||
@ -724,4 +735,226 @@ bool optimizeUseProjections(QueryPlan::Node & node, QueryPlan::Nodes & nodes)
|
||||
return true;
|
||||
}
|
||||
|
||||
|
||||
bool optimizeUseNormalProjections(Stack & stack, QueryPlan::Nodes & nodes)
|
||||
{
|
||||
const auto & frame = stack.back();
|
||||
|
||||
auto * reading = typeid_cast<ReadFromMergeTree *>(frame.node->step.get());
|
||||
if (!reading)
|
||||
return false;
|
||||
|
||||
auto iter = stack.rbegin();
|
||||
while (iter != stack.rend())
|
||||
{
|
||||
auto next = std::next(iter);
|
||||
|
||||
if (!typeid_cast<FilterStep *>(next->node->step.get()) &&
|
||||
!typeid_cast<ExpressionStep *>(next->node->step.get()))
|
||||
break;
|
||||
|
||||
iter = next;
|
||||
}
|
||||
|
||||
if (iter == stack.rbegin())
|
||||
return false;
|
||||
|
||||
const auto metadata = reading->getStorageMetadata();
|
||||
const auto & projections = metadata->projections;
|
||||
|
||||
std::vector<const ProjectionDescription *> normal_projections;
|
||||
for (const auto & projection : projections)
|
||||
if (projection.type == ProjectionDescription::Type::Normal)
|
||||
normal_projections.push_back(&projection);
|
||||
|
||||
if (normal_projections.empty())
|
||||
return false;
|
||||
|
||||
ActionsDAGPtr dag;
|
||||
ActionsDAG::NodeRawConstPtrs filter_nodes;
|
||||
if (!buildAggregatingDAG(*iter->node->children.front(), dag, filter_nodes))
|
||||
return false;
|
||||
|
||||
LOG_TRACE(&Poco::Logger::get("optimizeUseProjections"), "Query DAG: {}", dag->dumpDAG());
|
||||
|
||||
const ActionsDAG::Node * filter_node = nullptr;
|
||||
if (!filter_nodes.empty())
|
||||
{
|
||||
filter_node = filter_nodes.front();
|
||||
if (filter_nodes.size() > 1)
|
||||
{
|
||||
FunctionOverloadResolverPtr func_builder_and =
|
||||
std::make_unique<FunctionToOverloadResolverAdaptor>(
|
||||
std::make_shared<FunctionAnd>());
|
||||
|
||||
filter_node = &dag->addFunction(func_builder_and, std::move(filter_nodes), {});
|
||||
}
|
||||
|
||||
dag->getOutputs().push_back(filter_node);
|
||||
}
|
||||
|
||||
std::list<NormalProjectionCandidate> candidates;
|
||||
NormalProjectionCandidate * best_candidate = nullptr;
|
||||
|
||||
const Block & header = frame.node->step->getOutputStream().header;
|
||||
const Names & required_columns = reading->getRealColumnNames();
|
||||
const auto & parts = reading->getParts();
|
||||
const auto & query_info = reading->getQueryInfo();
|
||||
ContextPtr context = reading->getContext();
|
||||
MergeTreeDataSelectExecutor reader(reading->getMergeTreeData());
|
||||
|
||||
std::shared_ptr<PartitionIdToMaxBlock> max_added_blocks;
|
||||
if (context->getSettingsRef().select_sequential_consistency)
|
||||
{
|
||||
if (const StorageReplicatedMergeTree * replicated = dynamic_cast<const StorageReplicatedMergeTree *>(&reading->getMergeTreeData()))
|
||||
max_added_blocks = std::make_shared<PartitionIdToMaxBlock>(replicated->getMaxAddedBlocks());
|
||||
}
|
||||
|
||||
for (const auto * projection : normal_projections)
|
||||
{
|
||||
bool has_all_columns = true;
|
||||
for (const auto & col : required_columns)
|
||||
{
|
||||
if (!projection->sample_block.has(col))
|
||||
{
|
||||
has_all_columns = false;
|
||||
break;
|
||||
}
|
||||
}
|
||||
|
||||
if (!has_all_columns)
|
||||
continue;
|
||||
|
||||
MergeTreeData::DataPartsVector projection_parts;
|
||||
MergeTreeData::DataPartsVector normal_parts;
|
||||
for (const auto & part : parts)
|
||||
{
|
||||
const auto & created_projections = part->getProjectionParts();
|
||||
auto it = created_projections.find(projection->name);
|
||||
if (it != created_projections.end())
|
||||
projection_parts.push_back(it->second);
|
||||
else
|
||||
normal_parts.push_back(part);
|
||||
}
|
||||
|
||||
if (projection_parts.empty())
|
||||
continue;
|
||||
|
||||
ActionDAGNodes added_filter_nodes;
|
||||
if (filter_node)
|
||||
added_filter_nodes.nodes.push_back(filter_node);
|
||||
|
||||
auto projection_result_ptr = reader.estimateNumMarksToRead(
|
||||
std::move(projection_parts),
|
||||
nullptr,
|
||||
header.getNames(),
|
||||
metadata,
|
||||
projection->metadata,
|
||||
query_info, /// How it is actually used? I hope that for index we need only added_filter_nodes
|
||||
added_filter_nodes,
|
||||
context,
|
||||
context->getSettingsRef().max_threads,
|
||||
max_added_blocks);
|
||||
|
||||
if (projection_result_ptr->error())
|
||||
continue;
|
||||
|
||||
auto & candidate = candidates.emplace_back();
|
||||
candidate.merge_tree_projection_select_result_ptr = std::move(projection_result_ptr);
|
||||
candidate.sum_marks += candidate.merge_tree_projection_select_result_ptr->marks();
|
||||
|
||||
if (!normal_parts.empty())
|
||||
{
|
||||
auto normal_result_ptr = reading->selectRangesToRead(std::move(normal_parts));
|
||||
|
||||
if (normal_result_ptr->error())
|
||||
continue;
|
||||
|
||||
if (normal_result_ptr->marks() != 0)
|
||||
{
|
||||
candidate.sum_marks += normal_result_ptr->marks();
|
||||
candidate.merge_tree_normal_select_result_ptr = std::move(normal_result_ptr);
|
||||
}
|
||||
}
|
||||
|
||||
if (best_candidate == nullptr || best_candidate->sum_marks > candidate.sum_marks)
|
||||
best_candidate = &candidate;
|
||||
}
|
||||
|
||||
if (!best_candidate)
|
||||
return false;
|
||||
|
||||
auto storage_snapshot = reading->getStorageSnapshot();
|
||||
auto proj_snapshot = std::make_shared<StorageSnapshot>(
|
||||
storage_snapshot->storage, storage_snapshot->metadata, storage_snapshot->object_columns); //, storage_snapshot->data);
|
||||
proj_snapshot->addProjection(best_candidate->projection);
|
||||
|
||||
LOG_TRACE(&Poco::Logger::get("optimizeUseProjections"), "Proj snapshot {}", proj_snapshot->getColumns(GetColumnsOptions::Kind::All).toString());
|
||||
|
||||
auto projection_reading = reader.readFromParts(
|
||||
{},
|
||||
header.getNames(),
|
||||
proj_snapshot,
|
||||
query_info,
|
||||
context,
|
||||
reading->getMaxBlockSize(),
|
||||
reading->getNumStreams(),
|
||||
max_added_blocks,
|
||||
best_candidate->merge_tree_projection_select_result_ptr,
|
||||
reading->isParallelReadingEnabled());
|
||||
|
||||
if (!projection_reading)
|
||||
{
|
||||
Pipe pipe(std::make_shared<NullSource>(proj_snapshot->getSampleBlockForColumns(header.getNames())));
|
||||
projection_reading = std::make_unique<ReadFromPreparedSource>(std::move(pipe));
|
||||
}
|
||||
|
||||
bool has_nornal_parts = best_candidate->merge_tree_normal_select_result_ptr != nullptr;
|
||||
if (has_nornal_parts)
|
||||
reading->setAnalyzedResult(std::move(best_candidate->merge_tree_normal_select_result_ptr));
|
||||
|
||||
LOG_TRACE(&Poco::Logger::get("optimizeUseProjections"), "Projection reading header {}", projection_reading->getOutputStream().header.dumpStructure());
|
||||
|
||||
projection_reading->setStepDescription(best_candidate->projection->name);
|
||||
|
||||
auto & projection_reading_node = nodes.emplace_back(QueryPlan::Node{.step = std::move(projection_reading)});
|
||||
auto & expr_or_filter_node = nodes.emplace_back();
|
||||
|
||||
if (filter_node)
|
||||
{
|
||||
expr_or_filter_node.step = std::make_unique<FilterStep>(
|
||||
projection_reading_node.step->getOutputStream(),
|
||||
dag,
|
||||
dag->getOutputs().front()->result_name,
|
||||
true);
|
||||
}
|
||||
else
|
||||
expr_or_filter_node.step = std::make_unique<ExpressionStep>(
|
||||
projection_reading_node.step->getOutputStream(),
|
||||
dag);
|
||||
|
||||
expr_or_filter_node.children.push_back(&projection_reading_node);
|
||||
|
||||
if (!has_nornal_parts)
|
||||
{
|
||||
/// All parts are taken from projection
|
||||
iter->node->children.front() = &expr_or_filter_node;
|
||||
|
||||
//optimizeAggregationInOrder(node, nodes);
|
||||
}
|
||||
else
|
||||
{
|
||||
auto & union_node = nodes.emplace_back();
|
||||
DataStreams input_streams = {iter->node->children.front()->step->getOutputStream(), expr_or_filter_node.step->getOutputStream()};
|
||||
union_node.step = std::make_unique<UnionStep>(std::move(input_streams));
|
||||
union_node.children = {iter->node->children.front(), &expr_or_filter_node};
|
||||
iter->node->children.front() = &union_node;
|
||||
|
||||
iter->next_child = 0;
|
||||
stack.resize(iter.base() - stack.begin() + 1);
|
||||
}
|
||||
|
||||
return true;
|
||||
}
|
||||
|
||||
}
|
||||
|
Loading…
Reference in New Issue
Block a user