database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-11-05 04:50:35 +03:00
Code Activity

feat(rbo,rules,QA): the rule now uses target table to consider interesting keys available.

Browse Source
This commit is contained in:
drrtuy
2025-07-23 15:14:21 +00:00
parent ebaa6cce5c
commit a1ba7932cd
2 changed files with 89 additions and 68 deletions
Download Patch File Download Diff File Expand all files Collapse all files

50
dbcon/mysql/rbo_apply_parallel_ces.cpp
View File

@@ -38,6 +38,13 @@ void applyParallelCES_exists(execplan::CalpontSelectExecutionPlan& csep, const s
static const std::string RewrittenSubTableAliasPrefix = "$added_sub_"; static const std::string RewrittenSubTableAliasPrefix = "$added_sub_";
static const size_t MaxParallelFactor = 16; static const size_t MaxParallelFactor = 16;
bool tableAliasEqual(const execplan::CalpontSystemCatalog::TableAliasName& lhs,
const execplan::CalpontSystemCatalog::TableAliasName& rhs)
{
return (lhs.schema == rhs.schema && lhs.table == rhs.table && lhs.alias == rhs.alias &&
lhs.fisColumnStore == rhs.fisColumnStore);
}
bool tableIsInUnion(const execplan::CalpontSystemCatalog::TableAliasName& table, bool tableIsInUnion(const execplan::CalpontSystemCatalog::TableAliasName& table,
execplan::CalpontSelectExecutionPlan& csep) execplan::CalpontSelectExecutionPlan& csep)
{ {
@@ -116,16 +123,25 @@ execplan::ParseTree* filtersWithNewRangeAddedIfNeeded(execplan::SCSEP& csep, exe
// Looking for a projected column that comes first in an available index and has EI statistics // Looking for a projected column that comes first in an available index and has EI statistics
// INV nullptr signifies that no suitable column was found // INV nullptr signifies that no suitable column was found
execplan::SimpleColumn* findSuitableKeyColumn(execplan::CalpontSelectExecutionPlan& csep, execplan::SimpleColumn* findSuitableKeyColumn(execplan::CalpontSelectExecutionPlan& csep,
execplan::CalpontSystemCatalog::TableAliasName& targetTable,
optimizer::RBOptimizerContext& ctx) optimizer::RBOptimizerContext& ctx)
{ {
// TODO supply a list of suitable columns from a higher level
for (auto& rc : csep.returnedCols()) for (auto& rc : csep.returnedCols())
{ {
// TODO extract SC from RC // TODO extract SC from RC
auto* simpleColumn = dynamic_cast<execplan::SimpleColumn*>(rc.get()); auto* simpleColumn = dynamic_cast<execplan::SimpleColumn*>(rc.get());
if (simpleColumn) if (simpleColumn)
{ {
execplan::CalpontSystemCatalog::TableAliasName rcTable(
simpleColumn->schemaName(), simpleColumn->tableName(), simpleColumn->tableAlias(), "", false);
if (!tableAliasEqual(targetTable, rcTable))
{
continue;
}
cal_impl_if::SchemaAndTableName schemaAndTableName = {simpleColumn->schemaName(), cal_impl_if::SchemaAndTableName schemaAndTableName = {simpleColumn->schemaName(),
simpleColumn->tableName()}; simpleColumn->tableName()};
auto columnStatistics = ctx.gwi.findStatisticsForATable(schemaAndTableName); auto columnStatistics = ctx.gwi.findStatisticsForATable(schemaAndTableName);
if (!columnStatistics) if (!columnStatistics)
{ {
@@ -144,18 +160,22 @@ execplan::SimpleColumn* findSuitableKeyColumn(execplan::CalpontSelectExecutionPl
// TODO char and other numerical types support // TODO char and other numerical types support
execplan::CalpontSelectExecutionPlan::SelectList makeUnionFromTable( execplan::CalpontSelectExecutionPlan::SelectList makeUnionFromTable(
execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx) execplan::CalpontSelectExecutionPlan& csep, execplan::CalpontSystemCatalog::TableAliasName& table,
optimizer::RBOptimizerContext& ctx)
{ {
execplan::CalpontSelectExecutionPlan::SelectList unionVec; execplan::CalpontSelectExecutionPlan::SelectList unionVec;
// SC type controls an integral type used to produce suitable filters. The continuation of this function // SC type controls an integral type used to produce suitable filters. The continuation of this function
// should become a template function based on SC type. // should become a template function based on SC type.
execplan::SimpleColumn* keyColumn = findSuitableKeyColumn(csep, ctx); execplan::SimpleColumn* keyColumn = findSuitableKeyColumn(csep, table, ctx);
if (!keyColumn) if (!keyColumn)
{ {
return unionVec; return unionVec;
} }
// TODO char and other numerical types support
std::vector<std::pair<uint64_t, uint64_t>> bounds;
{
cal_impl_if::SchemaAndTableName schemaAndTableName = {keyColumn->schemaName(), keyColumn->tableName()}; cal_impl_if::SchemaAndTableName schemaAndTableName = {keyColumn->schemaName(), keyColumn->tableName()};
auto tableColumnsStatisticsIt = ctx.gwi.tableStatisticsMap.find(schemaAndTableName); auto tableColumnsStatisticsIt = ctx.gwi.tableStatisticsMap.find(schemaAndTableName);
if (tableColumnsStatisticsIt == ctx.gwi.tableStatisticsMap.end()) if (tableColumnsStatisticsIt == ctx.gwi.tableStatisticsMap.end())
@@ -176,9 +196,6 @@ execplan::CalpontSelectExecutionPlan::SelectList makeUnionFromTable(
size_t numberOfUnionUnits = std::min(columnStatistics.get_json_histogram().size(), MaxParallelFactor); size_t numberOfUnionUnits = std::min(columnStatistics.get_json_histogram().size(), MaxParallelFactor);
size_t numberOfBucketsPerUnionUnit = columnStatistics.get_json_histogram().size() / numberOfUnionUnits; size_t numberOfBucketsPerUnionUnit = columnStatistics.get_json_histogram().size() / numberOfUnionUnits;
// TODO char and other numerical types support
std::vector<std::pair<uint64_t, uint64_t>> bounds;
// Loop over buckets to produce filter ranges // Loop over buckets to produce filter ranges
for (size_t i = 0; i < numberOfUnionUnits - 1; ++i) for (size_t i = 0; i < numberOfUnionUnits - 1; ++i)
{ {
@@ -191,11 +208,12 @@ execplan::CalpontSelectExecutionPlan::SelectList makeUnionFromTable(
// Add last range // Add last range
// NB despite the fact that currently Histogram_json_hb has the last bucket that has end as its start // NB despite the fact that currently Histogram_json_hb has the last bucket that has end as its start
auto lastBucket = auto lastBucket = columnStatistics.get_json_histogram().begin() +
columnStatistics.get_json_histogram().begin() + (numberOfUnionUnits - 1) * numberOfBucketsPerUnionUnit; (numberOfUnionUnits - 1) * numberOfBucketsPerUnionUnit;
uint64_t currentLowerBound = *(uint32_t*)lastBucket->start_value.data(); uint64_t currentLowerBound = *(uint32_t*)lastBucket->start_value.data();
uint64_t currentUpperBound = *(uint32_t*)columnStatistics.get_last_bucket_end_endp().data(); uint64_t currentUpperBound = *(uint32_t*)columnStatistics.get_last_bucket_end_endp().data();
bounds.push_back({currentLowerBound, currentUpperBound}); bounds.push_back({currentLowerBound, currentUpperBound});
}
for (auto& bound : bounds) for (auto& bound : bounds)
{ {
@@ -229,13 +247,13 @@ bool applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerCon
std::string tableAlias = RewrittenSubTableAliasPrefix + table.schema + "_" + table.table + "_" + std::string tableAlias = RewrittenSubTableAliasPrefix + table.schema + "_" + table.table + "_" +
std::to_string(ctx.uniqueId); std::to_string(ctx.uniqueId);
// TODO add original alias to support multiple same name tables // TODO add original alias to support multiple same name tables
tableAliasMap.insert({table, tableAlias}); tableAliasMap.insert({table, {tableAlias, 0}});
derivedSCEP->location(execplan::CalpontSelectExecutionPlan::FROM); derivedSCEP->location(execplan::CalpontSelectExecutionPlan::FROM);
derivedSCEP->subType(execplan::CalpontSelectExecutionPlan::FROM_SUBS); derivedSCEP->subType(execplan::CalpontSelectExecutionPlan::FROM_SUBS);
derivedSCEP->derivedTbAlias(tableAlias); derivedSCEP->derivedTbAlias(tableAlias);
// Create a copy of the current leaf CSEP with additional filters to partition the key column // Create a copy of the current leaf CSEP with additional filters to partition the key column
auto additionalUnionVec = makeUnionFromTable(csep, ctx); auto additionalUnionVec = makeUnionFromTable(csep, table, ctx);
derivedSCEP->unionVec().insert(derivedSCEP->unionVec().end(), additionalUnionVec.begin(), derivedSCEP->unionVec().insert(derivedSCEP->unionVec().end(), additionalUnionVec.begin(),
additionalUnionVec.end()); additionalUnionVec.end());
@@ -250,7 +268,7 @@ bool applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerCon
} }
execplan::CalpontSelectExecutionPlan::ReturnedColumnList newReturnedColumns; execplan::CalpontSelectExecutionPlan::ReturnedColumnList newReturnedColumns;
[[maybe_unused]] size_t colPosition = 0; // [[maybe_unused]] size_t colPosition = 0;
// replace parent CSEP RCs with derived table RCs using ScheamAndTableName -> tableAlias map // replace parent CSEP RCs with derived table RCs using ScheamAndTableName -> tableAlias map
for (auto& rc : csep.returnedCols()) for (auto& rc : csep.returnedCols())
{ {
@@ -268,17 +286,19 @@ bool applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, RBOptimizerCon
// auto sc = scs[0]; // auto sc = scs[0];
std::cout << "Processing RC schema " << sc->schemaName() << " table " << sc->tableName() << " alias " std::cout << "Processing RC schema " << sc->schemaName() << " table " << sc->tableName() << " alias "
<< sc->tableAlias() << std::endl; << sc->tableAlias() << std::endl;
auto newTableAlias = tableAliasMap.find( auto newTableAliasAndColPositionCounter =
{sc->schemaName(), sc->tableName(), sc->tableAlias(), "", false}); tableAliasMap.find({sc->schemaName(), sc->tableName(), sc->tableAlias(), "", false});
if (newTableAlias == tableAliasMap.end()) if (newTableAliasAndColPositionCounter == tableAliasMap.end())
{ {
std::cout << "The RC doesn't belong to any of the derived tables, so leave it intact" << std::endl; std::cout << "The RC doesn't belong to any of the derived tables, so leave it intact" << std::endl;
continue; continue;
} }
sc->tableName(""); sc->tableName("");
sc->schemaName(""); sc->schemaName("");
sc->tableAlias(newTableAlias->second); auto& [newTableAlias, colPosition] = newTableAliasAndColPositionCounter->second;
sc->isColumnStore(true); sc->tableAlias(newTableAlias);
// WIP Not needed according with CSEP output
// sc->isColumnStore(true);
sc->colPosition(colPosition++); sc->colPosition(colPosition++);
// rcCloned->colPosition(colPosition++); // rcCloned->colPosition(colPosition++);
// rcCloned->resultType(rc->resultType()); // rcCloned->resultType(rc->resultType());

13
dbcon/mysql/rbo_apply_parallel_ces.h
View File

@@ -24,8 +24,9 @@
#include "execplan/calpontselectexecutionplan.h" #include "execplan/calpontselectexecutionplan.h"
#include "rulebased_optimizer.h" #include "rulebased_optimizer.h"
namespace optimizer { namespace optimizer
struct LessThan {
struct TableAliasLessThan
{ {
bool operator()(const execplan::CalpontSystemCatalog::TableAliasName& lhs, bool operator()(const execplan::CalpontSystemCatalog::TableAliasName& lhs,
const execplan::CalpontSystemCatalog::TableAliasName& rhs) const const execplan::CalpontSystemCatalog::TableAliasName& rhs) const
@@ -52,11 +53,11 @@ namespace optimizer {
return false; return false;
} }
}; };
using TableAliasMap = std::map<execplan::CalpontSystemCatalog::TableAliasName, std::string,
LessThan>;
bool matchParallelCES(execplan::CalpontSelectExecutionPlan& csep); using NewTableAliasAndColumnPosCounter = std::pair<string, size_t>;
void applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx); using TableAliasMap = std::map<execplan::CalpontSystemCatalog::TableAliasName,
NewTableAliasAndColumnPosCounter, TableAliasLessThan>;
bool parallelCESFilter(execplan::CalpontSelectExecutionPlan& csep); bool parallelCESFilter(execplan::CalpontSelectExecutionPlan& csep);
bool applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx); bool applyParallelCES(execplan::CalpontSelectExecutionPlan& csep, optimizer::RBOptimizerContext& ctx);
} }