database/mariadb-columnstore-engine
1
0
Fork 0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-04-18 21:44:02 +03:00
Code
mariadb-columnstore-engine/dbcon/execplan/windowfunctioncolumn.cpp
Gagan Goel 973e5024d8 MCOL-4957 Fix performance slowdown for processing TIMESTAMP columns. 
Part 1:
 As part of MCOL-3776 to address synchronization issue while accessing
 the fTimeZone member of the Func class, mutex locks were added to the
 accessor and mutator methods. However, this slows down processing
 of TIMESTAMP columns in PrimProc significantly as all threads across
 all concurrently running queries would serialize on the mutex. This
 is because PrimProc only has a single global object for the functor
 class (class derived from Func in utils/funcexp/functor.h) for a given
 function name. To fix this problem:

   (1) We remove the fTimeZone as a member of the Func derived classes
   (hence removing the mutexes) and instead use the fOperationType
   member of the FunctionColumn class to propagate the timezone values
   down to the individual functor processing functions such as
   FunctionColumn::getStrVal(), FunctionColumn::getIntVal(), etc.

   (2) To achieve (1), a timezone member is added to the
   execplan::CalpontSystemCatalog::ColType class.

Part 2:
 Several functors in the Funcexp code call dataconvert::gmtSecToMySQLTime()
 and dataconvert::mySQLTimeToGmtSec() functions for conversion between seconds
 since unix epoch and broken-down representation. These functions in turn call
 the C library function localtime_r() which currently has a known bug of holding
 a global lock via a call to __tz_convert. This significantly reduces performance
 in multi-threaded applications where multiple threads concurrently call
 localtime_r(). More details on the bug:
   https://sourceware.org/bugzilla/show_bug.cgi?id=16145

 This bug in localtime_r() caused processing of the Functors in PrimProc to
 slowdown significantly since a query execution causes Functors code to be
 processed in a multi-threaded manner.

 As a fix, we remove the calls to localtime_r() from gmtSecToMySQLTime()
 and mySQLTimeToGmtSec() by performing the timezone-to-offset conversion
 (done in dataconvert::timeZoneToOffset()) during the execution plan
 creation in the plugin. Note that localtime_r() is only called when the
 time_zone system variable is set to "SYSTEM".

 This fix also required changing the timezone type from a std::string to
 a long across the system.
2022-02-14 14:12:27 -05:00

728 lines
18 KiB
C++
Raw Blame History

/* Copyright (C) 2014 InfiniDB, Inc.
Copyright (C) 2019 MariaDB Corporation
This program is free software; you can redistribute it and/or
modify it under the terms of the GNU General Public License
as published by the Free Software Foundation; version 2 of
the License.
This program is distributed in the hope that it will be useful,
but WITHOUT ANY WARRANTY; without even the implied warranty of
MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
GNU General Public License for more details.
You should have received a copy of the GNU General Public License
along with this program; if not, write to the Free Software
Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston,
MA 02110-1301, USA. */
/***********************************************************************
* $Id: windowfunctioncolumn.cpp 9679 2013-07-11 22:32:03Z zzhu $
*
*
***********************************************************************/
#include <string>
#include <iostream>
#include <sstream>
using namespace std;
#include <boost/tokenizer.hpp>
#include <boost/algorithm/string.hpp>
using namespace boost;
#include "bytestream.h"
#include "windowfunctioncolumn.h"
#include "constantcolumn.h"
#include "arithmeticcolumn.h"
#include "simplecolumn.h"
#include "objectreader.h"
#include "calpontselectexecutionplan.h"
#include "simplefilter.h"
#include "aggregatecolumn.h"
#include "functioncolumn.h"
#include "funcexp.h"
#include "functor_export.h"
using namespace funcexp;
#include "rowgroup.h"
using namespace rowgroup;
#include "joblisttypes.h"
using namespace joblist;
#ifdef _MSC_VER
#define strcasecmp stricmp
#endif
namespace execplan
{
void getWindowFunctionCols(execplan::ParseTree* n, void* obj)
{
vector<WindowFunctionColumn*>* list = reinterpret_cast<vector<WindowFunctionColumn*>*>(obj);
TreeNode* tn = n->data();
WindowFunctionColumn* afc = dynamic_cast<WindowFunctionColumn*>(tn);
ArithmeticColumn* ac = dynamic_cast<ArithmeticColumn*>(tn);
FunctionColumn* fc = dynamic_cast<FunctionColumn*>(tn);
SimpleFilter* sf = dynamic_cast<SimpleFilter*>(tn);
if (afc)
list->push_back(afc);
else if (ac)
{
if (ac->hasWindowFunc()) // adds window functions to list
list->insert(list->end(), ac->windowfunctionColumnList().begin(), ac->windowfunctionColumnList().end());
}
else if (fc)
{
if (fc->hasWindowFunc())
list->insert(list->end(), fc->windowfunctionColumnList().begin(), fc->windowfunctionColumnList().end());
}
else if (sf)
{
list->insert(list->end(), sf->windowfunctionColumnList().begin(), sf->windowfunctionColumnList().end());
}
}
/**
* WF_Boundary class methods definition
*/
const std::string WF_Boundary::toString() const
{
ostringstream output;
if (fVal)
{
output << "val: ";
output << fVal->toString() << endl;
}
if (fBound)
{
output << "bound exp: ";
output << fBound->toString() << endl;
}
switch (fFrame)
{
case WF_PRECEDING: output << "PRECEDING"; break;
case WF_FOLLOWING: output << "FOLLOWING"; break;
case WF_UNBOUNDED_PRECEDING: output << "UNBOUNDED PRECEDING"; break;
case WF_UNBOUNDED_FOLLOWING: output << "UNBOUNDED FOLLOWING"; break;
case WF_CURRENT_ROW: output << "CURRENT ROW"; break;
default: output << "UNKNOWN";
}
return output.str();
}
void WF_Boundary::serialize(messageqcpp::ByteStream& b) const
{
b << (uint8_t)fFrame;
if (fVal)
fVal->serialize(b);
else
b << (uint8_t)ObjectReader::NULL_CLASS;
if (fBound)
fBound->serialize(b);
else
b << (uint8_t)ObjectReader::NULL_CLASS;
}
void WF_Boundary::unserialize(messageqcpp::ByteStream& b)
{
b >> (uint8_t&)fFrame;
fVal.reset(dynamic_cast<ReturnedColumn*>(ObjectReader::createTreeNode(b)));
fBound.reset(dynamic_cast<ReturnedColumn*>(ObjectReader::createTreeNode(b)));
}
/**
* WF_Frame class methods definition
*/
const string WF_Frame::toString() const
{
ostringstream output;
output << "WindowFrame:" << endl;
output << "Start:" << endl;
output << fStart.toString() << endl;
output << "End:" << endl;
output << fEnd.toString() << endl;
return output.str();
}
void WF_Frame::serialize(messageqcpp::ByteStream& b) const
{
fStart.serialize(b);
fEnd.serialize(b);
b << (uint8_t)fIsRange;
}
void WF_Frame::unserialize(messageqcpp::ByteStream& b)
{
fStart.unserialize(b);
fEnd.unserialize(b);
b >> (uint8_t&)fIsRange;
}
/**
* WF_OrderBy class methods definition
*/
const string WF_OrderBy::toString() const
{
ostringstream output;
output << "order by: " << endl;
for (uint32_t i = 0; i < fOrders.size(); i++)
output << fOrders[i]->toString() << endl;
output << fFrame.toString();
return output.str();
}
void WF_OrderBy::serialize(messageqcpp::ByteStream& b) const
{
b << (uint32_t)fOrders.size();
for (uint32_t i = 0; i < fOrders.size(); i++)
fOrders[i]->serialize(b);
fFrame.serialize(b);
}
void WF_OrderBy::unserialize(messageqcpp::ByteStream& b)
{
uint32_t size;
b >> (uint32_t&)size;
SRCP srcp;
for (uint32_t i = 0; i < size; i++)
{
srcp.reset(dynamic_cast<ReturnedColumn*>(ObjectReader::createTreeNode(b)));
fOrders.push_back(srcp);
}
fFrame.unserialize(b);
}
/**
* WindowFunctionColumn class definition
*/
WindowFunctionColumn::WindowFunctionColumn()
{
}
WindowFunctionColumn::WindowFunctionColumn(const string& functionName, const uint32_t sessionID)
: ReturnedColumn(sessionID), fFunctionName(functionName)
{
}
WindowFunctionColumn::WindowFunctionColumn(const WindowFunctionColumn& rhs, const uint32_t sessionID)
: ReturnedColumn(rhs, sessionID)
, fFunctionName(rhs.functionName())
, fFunctionParms(rhs.functionParms())
, fPartitions(rhs.partitions())
, fOrderBy(rhs.orderBy())
, udafContext(rhs.getUDAFContext())
, fTimeZone(rhs.timeZone())
{
}
const string WindowFunctionColumn::toString() const
{
ostringstream output;
output << "WindowFunctionColumn: " << fFunctionName;
if (distinct())
output << " DISTINCT";
output << endl;
output << "expressionId=" << fExpressionId << endl;
output << "resultType=" << colDataTypeToString(fResultType.colDataType) << "|" << fResultType.colWidth
<< endl;
output << "operationType=" << colDataTypeToString(fOperationType.colDataType) << endl;
output << "function parm: " << endl;
for (uint32_t i = 0; i < fFunctionParms.size(); i++)
output << fFunctionParms[i]->toString() << endl;
output << "partition by: " << endl;
for (uint32_t i = 0; i < fPartitions.size(); i++)
output << fPartitions[i]->toString() << endl;
output << fOrderBy.toString() << endl;
output << "getColumnList():" << endl;
vector<SRCP> colList = getColumnList();
for (uint32_t i = 0; i < colList.size(); i++)
output << colList[i]->toString() << endl;
return output.str();
}
void WindowFunctionColumn::serialize(messageqcpp::ByteStream& b) const
{
b << (ObjectReader::id_t)ObjectReader::WINDOWFUNCTIONCOLUMN;
ReturnedColumn::serialize(b);
b << fFunctionName;
b << (uint32_t)fFunctionParms.size();
for (uint32_t i = 0; i < fFunctionParms.size(); i++)
fFunctionParms[i]->serialize(b);
b << (uint32_t)fPartitions.size();
for (uint32_t i = 0; i < fPartitions.size(); i++)
fPartitions[i]->serialize(b);
fOrderBy.serialize(b);
udafContext.serialize(b);
messageqcpp::ByteStream::octbyte timeZone = fTimeZone;
b << timeZone;
}
void WindowFunctionColumn::unserialize(messageqcpp::ByteStream& b)
{
ObjectReader::checkType(b, ObjectReader::WINDOWFUNCTIONCOLUMN);
ReturnedColumn::unserialize(b);
uint32_t size;
SRCP srcp;
fFunctionParms.clear();
fPartitions.clear();
b >> fFunctionName;
b >> (uint32_t&)size;
for (uint32_t i = 0; i < size; i++)
{
srcp.reset(dynamic_cast<ReturnedColumn*>(ObjectReader::createTreeNode(b)));
fFunctionParms.push_back(srcp);
}
b >> (uint32_t&)size;
for (uint32_t i = 0; i < size; i++)
{
srcp.reset(dynamic_cast<ReturnedColumn*>(ObjectReader::createTreeNode(b)));
fPartitions.push_back(srcp);
}
fOrderBy.unserialize(b);
udafContext.unserialize(b);
messageqcpp::ByteStream::octbyte timeZone;
b >> timeZone;
fTimeZone = timeZone;
}
void WindowFunctionColumn::addToPartition(vector<SRCP>& groupByList)
{
fPartitions.insert(fPartitions.end(), groupByList.begin(), groupByList.end());
}
vector<SRCP> WindowFunctionColumn::getColumnList() const
{
vector<SRCP> columnList;
columnList.insert(columnList.end(), fFunctionParms.begin(), fFunctionParms.end());
columnList.insert(columnList.end(), fPartitions.begin(), fPartitions.end());
columnList.insert(columnList.end(), fOrderBy.fOrders.begin(), fOrderBy.fOrders.end());
if (fOrderBy.fFrame.fStart.fVal)
columnList.push_back(fOrderBy.fFrame.fStart.fVal);
if (fOrderBy.fFrame.fStart.fBound)
columnList.push_back(fOrderBy.fFrame.fStart.fBound);
if (fOrderBy.fFrame.fEnd.fVal)
columnList.push_back(fOrderBy.fFrame.fEnd.fVal);
if (fOrderBy.fFrame.fEnd.fBound)
columnList.push_back(fOrderBy.fFrame.fEnd.fBound);
return columnList;
}
bool WindowFunctionColumn::hasWindowFunc()
{
fWindowFunctionColumnList.push_back(this);
return true;
}
void WindowFunctionColumn::adjustResultType()
{
if ((fResultType.colDataType == CalpontSystemCatalog::DECIMAL ||
fResultType.colDataType == CalpontSystemCatalog::UDECIMAL) &&
!boost::iequals(fFunctionName, "COUNT") && !boost::iequals(fFunctionName, "COUNT(*)") &&
!boost::iequals(fFunctionName, "ROW_NUMBER") && !boost::iequals(fFunctionName, "RANK") &&
!boost::iequals(fFunctionName, "PERCENT_RANK") && !boost::iequals(fFunctionName, "DENSE_RANK") &&
!boost::iequals(fFunctionName, "CUME_DIST") && !boost::iequals(fFunctionName, "NTILE") &&
!boost::iequals(fFunctionName, "PERCENTILE") && !fFunctionParms.empty() &&
fFunctionParms[0]->resultType().colDataType == CalpontSystemCatalog::DOUBLE)
fResultType = fFunctionParms[0]->resultType();
if ((boost::iequals(fFunctionName, "LEAD") || boost::iequals(fFunctionName, "LAG") ||
boost::iequals(fFunctionName, "MIN") || boost::iequals(fFunctionName, "MAX") ||
boost::iequals(fFunctionName, "FIRST_VALUE") || boost::iequals(fFunctionName, "LAST_VALUE") ||
boost::iequals(fFunctionName, "NTH_VALUE")) &&
!fFunctionParms.empty())
fResultType = fFunctionParms[0]->resultType();
if (boost::iequals(fFunctionName, "SUM") || boost::iequals(fFunctionName, "AVG") ||
boost::iequals(fFunctionName, "AVG_DISTINCT") || boost::iequals(fFunctionName, "PERCENTILE"))
{
if (fFunctionParms[0]->resultType().colDataType == CalpontSystemCatalog::DECIMAL ||
fFunctionParms[0]->resultType().colDataType == CalpontSystemCatalog::UDECIMAL)
{
fResultType.colWidth = datatypes::MAXDECIMALWIDTH;
}
else
{
fResultType.colDataType = CalpontSystemCatalog::LONGDOUBLE;
fResultType.colWidth = sizeof(long double);
fResultType.precision = -1;
}
}
}
void WindowFunctionColumn::evaluate(Row& row, bool& isNull)
{
switch (fResultType.colDataType)
{
case CalpontSystemCatalog::DATE:
{
if (row.equals<4>(DATENULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getUintField<4>(fInputIndex);
break;
}
case CalpontSystemCatalog::DATETIME:
{
if (row.equals<8>(DATETIMENULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getUintField<8>(fInputIndex);
break;
}
case CalpontSystemCatalog::TIMESTAMP:
{
if (row.equals<8>(TIMESTAMPNULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getUintField<8>(fInputIndex);
break;
}
case CalpontSystemCatalog::TIME:
{
if (row.equals<8>(TIMENULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getIntField<8>(fInputIndex);
break;
}
case CalpontSystemCatalog::CHAR:
case CalpontSystemCatalog::VARCHAR:
case CalpontSystemCatalog::STRINT:
{
switch (row.getColumnWidth(fInputIndex))
{
case 1:
if (row.equals<1>(CHAR1NULL, fInputIndex))
isNull = true;
else
fResult.origIntVal = row.getUintField<1>(fInputIndex);
break;
case 2:
if (row.equals<2>(CHAR2NULL, fInputIndex))
isNull = true;
else
fResult.origIntVal = row.getUintField<2>(fInputIndex);
break;
case 3:
case 4:
if (row.equals<4>(CHAR4NULL, fInputIndex))
isNull = true;
else
fResult.origIntVal = row.getUintField<4>(fInputIndex);
break;
case 5:
case 6:
case 7:
case 8:
if (row.equals<8>(CHAR8NULL, fInputIndex))
isNull = true;
else
fResult.origIntVal = row.getUintField<8>(fInputIndex);
break;
// TODO MCOL-641
case 16:
// fallthrough
default:
{
const auto str = row.getConstString(fInputIndex);
if (str.eq(utils::ConstString(CPNULLSTRMARK)))
isNull = true;
else
fResult.strVal = str.toString();
// stringColVal is padded with '\0' to colWidth so can't use str.length()
if (strlen(fResult.strVal.c_str()) == 0)
isNull = true;
break;
}
}
if (fResultType.colDataType == CalpontSystemCatalog::STRINT)
fResult.intVal = uint64ToStr(fResult.origIntVal);
else
fResult.intVal = atoll((char*)&fResult.origIntVal);
break;
}
case CalpontSystemCatalog::BIGINT:
{
if (row.equals<8>(BIGINTNULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getIntField<8>(fInputIndex);
break;
}
case CalpontSystemCatalog::UBIGINT:
{
if (row.equals<8>(UBIGINTNULL, fInputIndex))
isNull = true;
else
fResult.uintVal = row.getUintField<8>(fInputIndex);
break;
}
case CalpontSystemCatalog::INT:
case CalpontSystemCatalog::MEDINT:
{
if (row.equals<4>(INTNULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getIntField<4>(fInputIndex);
break;
}
case CalpontSystemCatalog::UINT:
case CalpontSystemCatalog::UMEDINT:
{
if (row.equals<4>(UINTNULL, fInputIndex))
isNull = true;
else
fResult.uintVal = row.getUintField<4>(fInputIndex);
break;
}
case CalpontSystemCatalog::SMALLINT:
{
if (row.equals<2>(SMALLINTNULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getIntField<2>(fInputIndex);
break;
}
case CalpontSystemCatalog::USMALLINT:
{
if (row.equals<2>(USMALLINTNULL, fInputIndex))
isNull = true;
else
fResult.uintVal = row.getUintField<2>(fInputIndex);
break;
}
case CalpontSystemCatalog::TINYINT:
{
if (row.equals<1>(TINYINTNULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getIntField<1>(fInputIndex);
break;
}
case CalpontSystemCatalog::UTINYINT:
{
if (row.equals<1>(UTINYINTNULL, fInputIndex))
isNull = true;
else
fResult.uintVal = row.getUintField<1>(fInputIndex);
break;
}
// In this case, we're trying to load a double output column with float data. This is the
// case when you do sum(floatcol), e.g.
case CalpontSystemCatalog::FLOAT:
case CalpontSystemCatalog::UFLOAT:
{
if (row.equals<4>(FLOATNULL, fInputIndex))
isNull = true;
else
fResult.floatVal = row.getFloatField(fInputIndex);
break;
}
case CalpontSystemCatalog::DOUBLE:
case CalpontSystemCatalog::UDOUBLE:
{
if (row.equals<8>(DOUBLENULL, fInputIndex))
isNull = true;
else
fResult.doubleVal = row.getDoubleField(fInputIndex);
break;
}
case CalpontSystemCatalog::LONGDOUBLE:
{
if (row.equals(LONGDOUBLENULL, fInputIndex))
isNull = true;
else
fResult.longDoubleVal = row.getLongDoubleField(fInputIndex);
break;
}
case CalpontSystemCatalog::DECIMAL:
case CalpontSystemCatalog::UDECIMAL:
{
switch (fResultType.colWidth)
{
case 1:
{
if (row.equals<1>(TINYINTNULL, fInputIndex))
isNull = true;
else
{
fResult.decimalVal =
IDB_Decimal(row.getIntField<1>(fInputIndex), fResultType.scale, fResultType.precision);
}
break;
}
case 2:
{
if (row.equals<2>(SMALLINTNULL, fInputIndex))
isNull = true;
else
{
fResult.decimalVal =
IDB_Decimal(row.getIntField<2>(fInputIndex), fResultType.scale, fResultType.precision);
}
break;
}
case 4:
{
if (row.equals<4>(INTNULL, fInputIndex))
isNull = true;
else
{
fResult.decimalVal =
IDB_Decimal(row.getIntField<4>(fInputIndex), fResultType.scale, fResultType.precision);
}
break;
}
case 8:
{
if (row.equals<8>(BIGINTNULL, fInputIndex))
isNull = true;
else
{
fResult.decimalVal =
IDB_Decimal(row.getIntField<8>(fInputIndex), fResultType.scale, fResultType.precision);
}
break;
}
case 16:
{
datatypes::TSInt128 val = row.getTSInt128Field(fInputIndex);
if (val.isNull())
{
isNull = true;
}
else
{
fResult.decimalVal = IDB_Decimal(val, fResultType.scale, fResultType.precision);
}
break;
}
default:
// Should log error
break;
}
break;
}
case CalpontSystemCatalog::VARBINARY: isNull = true; break;
default: // treat as int64
{
if (row.equals<8>(BIGINTNULL, fInputIndex))
isNull = true;
else
fResult.intVal = row.getUintField<8>(fInputIndex);
break;
}
}
}
ostream& operator<<(ostream& output, const WindowFunctionColumn& rhs)
{
output << rhs.toString();
return output;
}
} // namespace execplan
View Git Blame Copy Permalink