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/primitives/primproc/batchprimitiveprocessor.cpp
david.hall 3b6449842f Merge branch 'develop' into MCOL-4841 
# Conflicts:
#	exemgr/main.cpp
#	oam/etc/Columnstore.xml.singleserver
#	primitives/primproc/primproc.cpp
2022-06-09 10:07:26 -05:00

2821 lines
84 KiB
C++
Raw Blame History

/* Copyright (C) 2014 InfiniDB, Inc.
Copyright (C) 2016-2020 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: batchprimitiveprocessor.cpp 2136 2013-07-24 21:04:30Z pleblanc $
// C++ Implementation: batchprimitiveprocessor
//
// Description:
//
//
// Author: Patrick LeBlanc <pleblanc@calpont.com>, (C) 2008
//
// Copyright: See COPYING file that comes with this distribution
//
//
#include <stdexcept>
#include <unistd.h>
#include <cstring>
//#define NDEBUG
#include <cassert>
#include <string>
#include <sstream>
#include <set>
#include <stdlib.h>
using namespace std;
#include <boost/thread.hpp>
#include <boost/shared_ptr.hpp>
using namespace boost;
#include "bpp.h"
#include "primitiveserver.h"
#include "errorcodes.h"
#include "exceptclasses.h"
#include "pp_logger.h"
#include "funcexpwrapper.h"
#include "fixedallocator.h"
#include "blockcacheclient.h"
#include "MonitorProcMem.h"
#include "threadnaming.h"
#include "vlarray.h"
#include "widedecimalutils.h"
#define MAX64 0x7fffffffffffffffLL
#define MIN64 0x8000000000000000LL
using namespace messageqcpp;
using namespace joiner;
using namespace std::tr1;
using namespace rowgroup;
using namespace funcexp;
using namespace logging;
using namespace utils;
using namespace joblist;
namespace primitiveprocessor
{
#ifdef PRIMPROC_STOPWATCH
#include "poormanprofiler.inc"
#endif
// these are config parms defined in primitiveserver.cpp, initialized by PrimProc main().
extern uint32_t blocksReadAhead;
extern uint32_t dictBufferSize;
extern uint32_t defaultBufferSize;
extern int fCacheCount;
extern uint32_t connectionsPerUM;
extern int noVB;
// copied from https://graphics.stanford.edu/~seander/bithacks.html#RoundUpPowerOf2
uint nextPowOf2(uint x)
{
x--;
x |= x >> 1;
x |= x >> 2;
x |= x >> 4;
x |= x >> 8;
x |= x >> 16;
x++;
return x;
}
BatchPrimitiveProcessor::BatchPrimitiveProcessor()
: ot(BPS_ELEMENT_TYPE)
, txnID(0)
, sessionID(0)
, stepID(0)
, uniqueID(0)
, count(1)
, baseRid(0)
, ridCount(0)
, needStrValues(false)
, wideColumnsWidths(0)
, filterCount(0)
, projectCount(0)
, sendRidsAtDelivery(false)
, ridMap(0)
, gotAbsRids(false)
, gotValues(false)
, hasScan(false)
, validCPData(false)
, minVal(MAX64)
, maxVal(MIN64)
, cpDataFromDictScan(false)
, lbidForCP(0)
, hasWideColumnOut(false)
, busyLoaderCount(0)
, physIO(0)
, cachedIO(0)
, touchedBlocks(0)
, LBIDTrace(false)
, fBusy(false)
, doJoin(false)
, hasFilterStep(false)
, filtOnString(false)
, prefetchThreshold(0)
, mJOINHasSkewedKeyColumn(false)
, mSmallSideRGPtr(nullptr)
, mSmallSideKeyColumnsPtr(nullptr)
, hasDictStep(false)
, sockIndex(0)
, endOfJoinerRan(false)
, processorThreads(0)
, ptMask(0)
, firstInstance(false)
, valuesLBID(0)
{
pp.setLogicalBlockMode(true);
pp.setBlockPtr((int*)blockData);
pthread_mutex_init(&objLock, NULL);
}
BatchPrimitiveProcessor::BatchPrimitiveProcessor(ByteStream& b, double prefetch,
boost::shared_ptr<BPPSendThread> bppst,
uint _processorThreads)
: ot(BPS_ELEMENT_TYPE)
, txnID(0)
, sessionID(0)
, stepID(0)
, uniqueID(0)
, count(1)
, baseRid(0)
, ridCount(0)
, needStrValues(false)
, wideColumnsWidths(0)
, filterCount(0)
, projectCount(0)
, sendRidsAtDelivery(false)
, ridMap(0)
, gotAbsRids(false)
, gotValues(false)
, hasScan(false)
, validCPData(false)
, minVal(MAX64)
, maxVal(MIN64)
, cpDataFromDictScan(false)
, lbidForCP(0)
, hasWideColumnOut(false)
, busyLoaderCount(0)
, physIO(0)
, cachedIO(0)
, touchedBlocks(0)
, LBIDTrace(false)
, fBusy(false)
, doJoin(false)
, hasFilterStep(false)
, filtOnString(false)
, prefetchThreshold(prefetch)
, mJOINHasSkewedKeyColumn(false)
, mSmallSideRGPtr(nullptr)
, mSmallSideKeyColumnsPtr(nullptr)
, hasDictStep(false)
, sockIndex(0)
, endOfJoinerRan(false)
, processorThreads(_processorThreads)
// processorThreads(32),
// ptMask(processorThreads - 1),
, firstInstance(true)
, valuesLBID(0)
{
// promote processorThreads to next power of 2. also need to change the name to bucketCount or similar
processorThreads = nextPowOf2(processorThreads);
ptMask = processorThreads - 1;
pp.setLogicalBlockMode(true);
pp.setBlockPtr((int*)blockData);
sendThread = bppst;
pthread_mutex_init(&objLock, NULL);
initBPP(b);
}
#if 0
BatchPrimitiveProcessor::BatchPrimitiveProcessor(const BatchPrimitiveProcessor& bpp)
{
throw logic_error("copy BPP deprecated");
}
#endif
BatchPrimitiveProcessor::~BatchPrimitiveProcessor()
{
// FIXME: just do a sync fetch
counterLock.lock(); // need to make sure the loader has exited
while (busyLoaderCount > 0)
{
counterLock.unlock();
usleep(100000);
counterLock.lock();
}
counterLock.unlock();
pthread_mutex_destroy(&objLock);
}
/**
* InitBPP Parses the creation messages from BatchPrimitiveProcessor-JL::createBPP()
* Refer to that fcn for message format info.
*/
void BatchPrimitiveProcessor::initBPP(ByteStream& bs)
{
uint32_t i;
uint8_t tmp8;
uint16_t tmp16;
Command::CommandType type;
bs.advance(sizeof(ISMPacketHeader)); // skip the header
bs >> tmp8;
ot = static_cast<BPSOutputType>(tmp8);
bs >> txnID;
bs >> sessionID;
bs >> stepID;
bs >> uniqueID;
bs >> versionInfo;
bs >> tmp16;
needStrValues = tmp16 & NEED_STR_VALUES;
gotAbsRids = tmp16 & GOT_ABS_RIDS;
gotValues = tmp16 & GOT_VALUES;
LBIDTrace = tmp16 & LBID_TRACE;
sendRidsAtDelivery = tmp16 & SEND_RIDS_AT_DELIVERY;
doJoin = tmp16 & HAS_JOINER;
hasRowGroup = tmp16 & HAS_ROWGROUP;
getTupleJoinRowGroupData = tmp16 & JOIN_ROWGROUP_DATA;
bool hasWideColumnsIn = tmp16 & HAS_WIDE_COLUMNS;
// This used to signify that there was input row data from previous jobsteps, and
// it never quite worked right. No need to fix it or update it; all BPP's have started
// with a scan for years. Took it out.
assert(!hasRowGroup);
if (hasWideColumnsIn)
bs >> wideColumnsWidths;
bs >> bop;
bs >> forHJ;
if (ot == ROW_GROUP)
{
bs >> outputRG;
// outputRG.setUseStringTable(true);
bs >> tmp8;
if (tmp8)
{
fe1.reset(new FuncExpWrapper());
bs >> *fe1;
bs >> fe1Input;
}
bs >> tmp8;
if (tmp8)
{
fe2.reset(new FuncExpWrapper());
bs >> *fe2;
bs >> fe2Output;
}
}
if (doJoin)
{
pthread_mutex_lock(&objLock);
if (ot == ROW_GROUP)
{
bs >> joinerCount;
// cout << "joinerCount = " << joinerCount << endl;
joinTypes.reset(new JoinType[joinerCount]);
tJoiners.reset(new boost::shared_array<boost::shared_ptr<TJoiner> >[joinerCount]);
for (uint j = 0; j < joinerCount; ++j)
tJoiners[j].reset(new boost::shared_ptr<TJoiner>[processorThreads]);
tlJoiners.reset(new boost::shared_array<boost::shared_ptr<TLJoiner> >[joinerCount]);
for (uint j = 0; j < joinerCount; ++j)
tlJoiners[j].reset(new boost::shared_ptr<TLJoiner>[processorThreads]);
addToJoinerLocks.reset(new boost::scoped_array<boost::mutex>[joinerCount]);
for (uint j = 0; j < joinerCount; ++j)
addToJoinerLocks[j].reset(new boost::mutex[processorThreads]);
smallSideDataLocks.reset(new boost::mutex[joinerCount]);
tJoinerSizes.reset(new std::atomic<uint32_t>[joinerCount]);
largeSideKeyColumns.reset(new uint32_t[joinerCount]);
tlLargeSideKeyColumns.reset(new vector<uint32_t>[joinerCount]);
tlSmallSideKeyColumns.reset(new std::vector<uint32_t>);
typelessJoin.reset(new bool[joinerCount]);
tlSmallSideKeyLengths.reset(new uint32_t[joinerCount]);
storedKeyAllocators.reset(new PoolAllocator[joinerCount]);
for (uint j = 0; j < joinerCount; ++j)
storedKeyAllocators[j].setUseLock(true);
joinNullValues.reset(new uint64_t[joinerCount]);
doMatchNulls.reset(new bool[joinerCount]);
joinFEFilters.reset(new scoped_ptr<FuncExpWrapper>[joinerCount]);
hasJoinFEFilters = false;
hasSmallOuterJoin = false;
bool smallSideRGRecvd = false;
for (i = 0; i < joinerCount; i++)
{
doMatchNulls[i] = false;
uint32_t tmp32;
bs >> tmp32;
tJoinerSizes[i] = tmp32;
// bs >> tJoinerSizes[i];
// cout << "joiner size = " << tJoinerSizes[i] << endl;
bs >> joinTypes[i];
bs >> tmp8;
typelessJoin[i] = (bool)tmp8;
if (joinTypes[i] & WITHFCNEXP)
{
hasJoinFEFilters = true;
joinFEFilters[i].reset(new FuncExpWrapper());
bs >> *joinFEFilters[i];
}
if (joinTypes[i] & SMALLOUTER)
hasSmallOuterJoin = true;
if (!typelessJoin[i])
{
bs >> joinNullValues[i];
bs >> largeSideKeyColumns[i];
// cout << "large side key is " << largeSideKeyColumns[i] << endl;
for (uint j = 0; j < processorThreads; ++j)
tJoiners[i][j].reset(new TJoiner(10, TupleJoiner::hasher()));
}
else
{
deserializeVector<uint32_t>(bs, tlLargeSideKeyColumns[i]);
bs >> tlSmallSideKeyLengths[i];
bs >> tmp8;
mJOINHasSkewedKeyColumn = (bool)tmp8;
// Deser smallSideRG if key data types are different, e.g. INT vs wide-DECIMAL.
if (mJOINHasSkewedKeyColumn && !smallSideRGRecvd)
{
smallSideRGs.emplace_back(rowgroup::RowGroup(bs));
// LargeSide key columns number equals to SmallSide key columns number.
deserializeVector<uint32_t>(bs, *tlSmallSideKeyColumns);
mSmallSideRGPtr = &smallSideRGs[0];
mSmallSideKeyColumnsPtr = &(*tlSmallSideKeyColumns);
smallSideRGRecvd = true;
}
for (uint j = 0; j < processorThreads; ++j)
{
auto tlHasher = TupleJoiner::TypelessDataHasher(&outputRG, &tlLargeSideKeyColumns[i],
mSmallSideKeyColumnsPtr, mSmallSideRGPtr);
auto tlComparator = TupleJoiner::TypelessDataComparator(&outputRG, &tlLargeSideKeyColumns[i],
mSmallSideKeyColumnsPtr, mSmallSideRGPtr);
tlJoiners[i][j].reset(new TLJoiner(10, tlHasher, tlComparator));
}
}
}
if (hasJoinFEFilters)
{
joinFERG.reset(new RowGroup());
bs >> *joinFERG;
}
if (getTupleJoinRowGroupData)
{
deserializeVector(bs, smallSideRGs);
// cout << "deserialized " << smallSideRGs.size() << " small-side
// rowgroups\n";
idbassert(smallSideRGs.size() == joinerCount);
smallSideRowLengths.reset(new uint32_t[joinerCount]);
smallSideRowData.reset(new RGData[joinerCount]);
smallNullRowData.reset(new RGData[joinerCount]);
smallNullPointers.reset(new Row::Pointer[joinerCount]);
ssrdPos.reset(new uint64_t[joinerCount]);
for (i = 0; i < joinerCount; i++)
{
smallSideRowLengths[i] = smallSideRGs[i].getRowSize();
;
smallSideRowData[i] = RGData(smallSideRGs[i], tJoinerSizes[i]);
// smallSideRowData[i].reset(new uint8_t[
// smallSideRGs[i].getEmptySize() +
// (uint64_t) smallSideRowLengths[i] * tJoinerSizes[i]]);
smallSideRGs[i].setData(&smallSideRowData[i]);
smallSideRGs[i].resetRowGroup(0);
ssrdPos[i] = smallSideRGs[i].getEmptySize();
if (joinTypes[i] & (LARGEOUTER | SEMI | ANTI))
{
Row smallRow;
smallSideRGs[i].initRow(&smallRow);
smallNullRowData[i] = RGData(smallSideRGs[i], 1);
smallSideRGs[i].setData(&smallNullRowData[i]);
smallSideRGs[i].getRow(0, &smallRow);
smallRow.initToNull();
smallNullPointers[i] = smallRow.getPointer();
smallSideRGs[i].setData(&smallSideRowData[i]);
}
}
bs >> largeSideRG;
bs >> joinedRG;
// cout << "got the joined Rowgroup: " << joinedRG.toString() << "\n";
}
}
#ifdef __FreeBSD__
pthread_mutex_unlock(&objLock);
#endif
}
bs >> filterCount;
filterSteps.resize(filterCount);
// cout << "deserializing " << filterCount << " filters\n";
hasScan = false;
hasPassThru = false;
for (i = 0; i < filterCount; ++i)
{
// cout << "deserializing step " << i << endl;
filterSteps[i] = SCommand(Command::makeCommand(bs, &type, filterSteps));
if (type == Command::COLUMN_COMMAND)
{
ColumnCommand* col = (ColumnCommand*)filterSteps[i].get();
if (col->isScan())
hasScan = true;
if (bop == BOP_OR)
col->setScan(true);
}
else if (type == Command::FILTER_COMMAND)
{
hasFilterStep = true;
if (dynamic_cast<StrFilterCmd*>(filterSteps[i].get()) != NULL)
filtOnString = true;
}
else if (type == Command::DICT_STEP || type == Command::RID_TO_STRING)
hasDictStep = true;
}
bs >> projectCount;
// cout << "deserializing " << projectCount << " projected columns\n\n";
projectSteps.resize(projectCount);
for (i = 0; i < projectCount; ++i)
{
// cout << "deserializing step " << i << endl;
projectSteps[i] = SCommand(Command::makeCommand(bs, &type, projectSteps));
if (type == Command::PASS_THRU)
hasPassThru = true;
else if (type == Command::DICT_STEP || type == Command::RID_TO_STRING)
hasDictStep = true;
}
if (ot == ROW_GROUP)
{
bs >> tmp8;
if (tmp8 > 0)
{
bs >> fAggregateRG;
fAggregator.reset(new RowAggregation);
bs >> *(fAggregator.get());
// If there's UDAF involved, set up for PM processing
for (uint64_t i = 0; i < fAggregator->getAggFunctions().size(); i++)
{
RowUDAFFunctionCol* rowUDAF =
dynamic_cast<RowUDAFFunctionCol*>(fAggregator->getAggFunctions()[i].get());
if (rowUDAF)
{
// On the PM, the aux column is not sent, but rather is output col + 1.
rowUDAF->fAuxColumnIndex = rowUDAF->fOutputColumnIndex + 1;
// Set the PM flag in case the UDAF cares.
rowUDAF->fUDAFContext.setContextFlags(rowUDAF->fUDAFContext.getContextFlags() |
mcsv1sdk::CONTEXT_IS_PM);
}
}
}
}
initProcessor();
}
/**
* resetBPP Parses the run messages from BatchPrimitiveProcessor-JL::runBPP()
* Refer to that fcn for message format info.
*/
void BatchPrimitiveProcessor::resetBPP(ByteStream& bs, const SP_UM_MUTEX& w, const SP_UM_IOSOCK& s)
{
uint32_t i;
vector<uint64_t> preloads;
pthread_mutex_lock(&objLock);
writelock = w;
sock = s;
newConnection = true;
// skip the header, sessionID, stepID, uniqueID, and priority
bs.advance(sizeof(ISMPacketHeader) + 16);
bs >> dbRoot;
bs >> count;
bs >> ridCount;
if (gotAbsRids)
{
assert(0);
memcpy(absRids.get(), bs.buf(), ridCount << 3);
bs.advance(ridCount << 3);
/* TODO: this loop isn't always necessary or sensible */
ridMap = 0;
baseRid = absRids[0] & 0xffffffffffffe000ULL;
for (uint32_t i = 0; i < ridCount; i++)
{
relRids[i] = absRids[i] - baseRid;
ridMap |= 1 << (relRids[i] >> 9);
}
}
else
{
bs >> ridMap;
bs >> baseRid;
memcpy(relRids, bs.buf(), ridCount << 1);
bs.advance(ridCount << 1);
}
if (gotValues)
{
memcpy(values, bs.buf(), ridCount << 3);
bs.advance(ridCount << 3);
}
for (i = 0; i < filterCount; ++i)
{
filterSteps[i]->resetCommand(bs);
}
for (i = 0; i < projectCount; ++i)
{
projectSteps[i]->resetCommand(bs);
}
idbassert(bs.length() == 0);
/* init vars not part of the BS */
currentBlockOffset = 0;
memset(relLBID.get(), 0, sizeof(uint64_t) * (projectCount + 1));
memset(asyncLoaded.get(), 0, sizeof(bool) * (projectCount + 1));
buildVSSCache(count);
#ifdef __FreeBSD__
pthread_mutex_unlock(&objLock);
#endif
}
// This version of addToJoiner() is multithreaded. Values are first
// hashed into thread-local vectors corresponding to the shared hash
// tables. Once the incoming values are organized locally, it grabs
// the lock for each shared table and inserts them there.
void BatchPrimitiveProcessor::addToJoiner(ByteStream& bs)
{
/* to get wall-time of hash table construction
idbassert(processorThreads != 0);
if (firstCallTime.is_not_a_date_time())
firstCallTime = boost::posix_time::microsec_clock::universal_time();
*/
uint32_t count, i, joinerNum, tlIndex, startPos, bucket;
#pragma pack(push, 1)
struct JoinerElements
{
uint64_t key;
uint32_t value;
} * arr;
#pragma pack(pop)
/* skip the header */
bs.advance(sizeof(ISMPacketHeader) + 3 * sizeof(uint32_t));
bs >> count;
bs >> startPos;
if (ot == ROW_GROUP)
{
bs >> joinerNum;
idbassert(joinerNum < joinerCount);
arr = (JoinerElements*)bs.buf();
std::atomic<uint32_t>& tJoinerSize = tJoinerSizes[joinerNum];
// XXXPAT: enormous if stmts are evil. TODO: move each block into
// properly-named functions for clarity.
if (typelessJoin[joinerNum])
{
utils::VLArray<vector<pair<TypelessData, uint32_t> > > tmpBuckets(processorThreads);
uint8_t nullFlag;
PoolAllocator& storedKeyAllocator = storedKeyAllocators[joinerNum];
// this first loop hashes incoming values into vectors that parallel the hash tables.
uint nullCount = 0;
for (i = 0; i < count; ++i)
{
bs >> nullFlag;
if (nullFlag == 0)
{
TypelessData tlSmallSideKey(bs, storedKeyAllocator);
if (mJOINHasSkewedKeyColumn)
tlSmallSideKey.setSmallSideWithSkewedData();
else
tlSmallSideKey.setSmallSide();
bs >> tlIndex;
// The bucket number corresponds with the index used later inserting TL keys into permanent JOIN
// hash map.
auto ha = tlSmallSideKey.hash(outputRG, tlLargeSideKeyColumns[joinerNum], mSmallSideKeyColumnsPtr,
mSmallSideRGPtr);
bucket = ha & ptMask;
tmpBuckets[bucket].push_back(make_pair(tlSmallSideKey, tlIndex));
}
else
++nullCount;
}
tJoinerSize -= nullCount;
bool done = false, didSomeWork;
// uint loopCounter = 0, noWorkCounter = 0;
// this loop moves the elements from each vector into its corresponding hash table.
while (!done)
{
//++loopCounter;
done = true;
didSomeWork = false;
for (i = 0; i < processorThreads; ++i)
{
if (!tmpBuckets[i].empty())
{
bool gotIt = addToJoinerLocks[joinerNum][i].try_lock();
if (!gotIt)
{
done = false; // didn't get it, don't block, try the next bucket
continue;
}
for (auto& element : tmpBuckets[i])
tlJoiners[joinerNum][i]->insert(element);
addToJoinerLocks[joinerNum][i].unlock();
tmpBuckets[i].clear();
didSomeWork = true;
}
}
// if this iteration did no useful work, everything we need is locked; wait briefly
// and try again.
if (!done && !didSomeWork)
{
::usleep(500 * processorThreads);
//++noWorkCounter;
}
}
// cout << "TL join insert. Took " << loopCounter << " loops" << endl;
}
else
{
boost::shared_array<boost::shared_ptr<TJoiner> > tJoiner = tJoiners[joinerNum];
uint64_t nullValue = joinNullValues[joinerNum];
bool& l_doMatchNulls = doMatchNulls[joinerNum];
joblist::JoinType joinType = joinTypes[joinerNum];
utils::VLArray<vector<pair<uint64_t, uint32_t> > > tmpBuckets(processorThreads);
if (joinType & MATCHNULLS)
{
// this first loop hashes incoming values into vectors that parallel the hash tables.
for (i = 0; i < count; ++i)
{
/* A minor optimization: the matchnull logic should only be used with
* the jointype specifies it and there's a null value in the small side */
if (!l_doMatchNulls && arr[i].key == nullValue)
l_doMatchNulls = true;
bucket = bucketPicker((char*)&arr[i].key, 8, bpSeed) & ptMask;
tmpBuckets[bucket].push_back(make_pair(arr[i].key, arr[i].value));
}
bool done = false, didSomeWork;
// uint loopCounter = 0, noWorkCounter = 0;
// this loop moves the elements from each vector into its corresponding hash table.
while (!done)
{
//++loopCounter;
done = true;
didSomeWork = false;
for (i = 0; i < processorThreads; ++i)
{
if (!tmpBuckets[i].empty())
{
bool gotIt = addToJoinerLocks[joinerNum][i].try_lock();
if (!gotIt)
{
done = false; // didn't get it, don't block, try the next bucket
continue;
}
for (auto& element : tmpBuckets[i])
tJoiners[joinerNum][i]->insert(element);
addToJoinerLocks[joinerNum][i].unlock();
tmpBuckets[i].clear();
didSomeWork = true;
}
}
// if this iteration did no useful work, everything we need is locked; wait briefly
// and try again.
if (!done && !didSomeWork)
{
::usleep(500 * processorThreads);
//++noWorkCounter;
}
}
// cout << "T numeric join insert. Took " << loopCounter << " loops" << endl;
}
else
{
// this first loop hashes incoming values into vectors that parallel the hash tables.
for (i = 0; i < count; ++i)
{
bucket = bucketPicker((char*)&arr[i].key, 8, bpSeed) & ptMask;
tmpBuckets[bucket].push_back(make_pair(arr[i].key, arr[i].value));
}
bool done = false;
bool didSomeWork;
// uint loopCounter = 0, noWorkCounter = 0;
// this loop moves the elements from each vector into its corresponding hash table.
while (!done)
{
//++loopCounter;
done = true;
didSomeWork = false;
for (i = 0; i < processorThreads; ++i)
{
if (!tmpBuckets[i].empty())
{
bool gotIt = addToJoinerLocks[joinerNum][i].try_lock();
if (!gotIt)
{
done = false; // didn't get it, don't block, try the next bucket
continue;
}
for (auto& element : tmpBuckets[i])
tJoiners[joinerNum][i]->insert(element);
addToJoinerLocks[joinerNum][i].unlock();
tmpBuckets[i].clear();
didSomeWork = true;
}
}
// if this iteration did no useful work, everything we need is locked; wait briefly
// and try again.
if (!done && !didSomeWork)
{
::usleep(500 * processorThreads);
//++noWorkCounter;
}
}
// cout << "T numeric join insert 2. Took " << loopCounter << " loops," <<
// " unproductive iterations = " << noWorkCounter << endl;
}
}
if (!typelessJoin[joinerNum])
bs.advance(count * sizeof(JoinerElements));
if (getTupleJoinRowGroupData)
{
RowGroup& smallSide = smallSideRGs[joinerNum];
RGData offTheWire;
// TODO: write an RGData fcn to let it interpret data within a ByteStream to avoid
// the extra copying.
offTheWire.deserialize(bs);
boost::mutex::scoped_lock lk(smallSideDataLocks[joinerNum]);
smallSide.setData(&smallSideRowData[joinerNum]);
smallSide.append(offTheWire, startPos);
/* This prints the row data
smallSideRGs[joinerNum].initRow(&r);
for (i = 0; i < (tJoinerSizes[joinerNum] * smallSideRowLengths[joinerNum]);
i+=r.getSize()) { r.setData(&smallSideRowData[joinerNum][i +
smallSideRGs[joinerNum].getEmptySize()]); cout << " got row: " << r.toString() << endl;
}
*/
}
}
idbassert(bs.length() == 0);
}
void BatchPrimitiveProcessor::doneSendingJoinerData()
{
/* to get wall-time of hash table construction
if (!firstCallTime.is_not_a_date_time() && !(sessionID & 0x80000000))
{
boost::posix_time::ptime now = boost::posix_time::microsec_clock::universal_time();
Logger logger;
ostringstream os;
os << "id " << uniqueID << ": joiner construction time = " << now-firstCallTime;
logger.logMessage(os.str());
cout << os.str() << endl;
}
*/
}
int BatchPrimitiveProcessor::endOfJoiner()
{
/* Wait for all joiner elements to be added */
uint32_t i;
size_t currentSize;
// it should be safe to run this without grabbing this lock
// boost::mutex::scoped_lock scoped(addToJoinerLock);
if (endOfJoinerRan)
return 0;
// minor hack / optimization. The instances not inserting the table data don't
// need to check that the table is complete.
if (!firstInstance)
{
endOfJoinerRan = true;
pthread_mutex_unlock(&objLock);
return 0;
}
for (i = 0; i < joinerCount; i++)
{
if (!typelessJoin[i])
{
currentSize = 0;
for (uint j = 0; j < processorThreads; ++j)
if (!tJoiners[i] || !tJoiners[i][j])
return -1;
else
currentSize += tJoiners[i][j]->size();
if (currentSize != tJoinerSizes[i])
return -1;
// if ((!tJoiners[i] || tJoiners[i]->size() != tJoinerSizes[i]))
// return -1;
}
else
{
currentSize = 0;
for (uint j = 0; j < processorThreads; ++j)
if (!tlJoiners[i] || !tlJoiners[i][j])
return -1;
else
currentSize += tlJoiners[i][j]->size();
if (currentSize != tJoinerSizes[i])
return -1;
// if ((!tJoiners[i] || tlJoiners[i]->size() != tJoinerSizes[i]))
// return -1;
}
}
endOfJoinerRan = true;
#ifndef __FreeBSD__
pthread_mutex_unlock(&objLock);
#endif
return 0;
}
void BatchPrimitiveProcessor::initProcessor()
{
uint32_t i, j;
if (gotAbsRids || needStrValues || hasRowGroup)
absRids.reset(new uint64_t[LOGICAL_BLOCK_RIDS]);
if (needStrValues)
strValues.reset(new string[LOGICAL_BLOCK_RIDS]);
outMsgSize = defaultBufferSize;
outputMsg.reset(reinterpret_cast<uint8_t*>(aligned_alloc(utils::MAXCOLUMNWIDTH, outMsgSize)));
if (ot == ROW_GROUP)
{
// calculate the projection -> outputRG mapping
projectionMap.reset(new int[projectCount]);
bool* reserved = (bool*)alloca(outputRG.getColumnCount() * sizeof(bool));
for (i = 0; i < outputRG.getColumnCount(); i++)
reserved[i] = false;
for (i = 0; i < projectCount; i++)
{
for (j = 0; j < outputRG.getColumnCount(); j++)
if (projectSteps[i]->getTupleKey() == outputRG.getKeys()[j] && !reserved[j])
{
projectionMap[i] = j;
reserved[j] = true;
break;
}
if (j == outputRG.getColumnCount())
projectionMap[i] = -1;
}
if (doJoin)
{
outputRG.initRow(&oldRow);
outputRG.initRow(&newRow);
tSmallSideMatches.reset(new MatchedData[joinerCount]);
keyColumnProj.reset(new bool[projectCount]);
for (i = 0; i < projectCount; i++)
{
keyColumnProj[i] = false;
for (j = 0; j < joinerCount; j++)
{
if (!typelessJoin[j])
{
if (projectionMap[i] == (int)largeSideKeyColumns[j])
{
keyColumnProj[i] = true;
break;
}
}
else
{
for (uint32_t k = 0; k < tlLargeSideKeyColumns[j].size(); k++)
{
if (projectionMap[i] == (int)tlLargeSideKeyColumns[j][k])
{
keyColumnProj[i] = true;
break;
}
}
}
}
}
if (hasJoinFEFilters)
{
joinFERG->initRow(&joinFERow, true);
joinFERowData.reset(new uint8_t[joinFERow.getSize()]);
joinFERow.setData(joinFERowData.get());
joinFEMappings.reset(new shared_array<int>[joinerCount + 1]);
for (i = 0; i < joinerCount; i++)
joinFEMappings[i] = makeMapping(smallSideRGs[i], *joinFERG);
joinFEMappings[joinerCount] = makeMapping(largeSideRG, *joinFERG);
}
}
/*
Calculate the FE1 -> projection mapping
Calculate the projection step -> FE1 input mapping
*/
if (fe1)
{
fe1ToProjection = makeMapping(fe1Input, outputRG);
projectForFE1.reset(new int[projectCount]);
bool* reserved = (bool*)alloca(fe1Input.getColumnCount() * sizeof(bool));
for (i = 0; i < fe1Input.getColumnCount(); i++)
reserved[i] = false;
for (i = 0; i < projectCount; i++)
{
projectForFE1[i] = -1;
for (j = 0; j < fe1Input.getColumnCount(); j++)
{
if (projectSteps[i]->getTupleKey() == fe1Input.getKeys()[j] && !reserved[j])
{
reserved[j] = true;
projectForFE1[i] = j;
break;
}
}
}
fe1Input.initRow(&fe1In);
outputRG.initRow(&fe1Out);
}
if (fe2)
{
fe2Input = (doJoin ? &joinedRG : &outputRG);
fe2Mapping = makeMapping(*fe2Input, fe2Output);
fe2Input->initRow(&fe2In);
fe2Output.initRow(&fe2Out);
}
if (getTupleJoinRowGroupData)
{
gjrgPlaceHolders.reset(new uint32_t[joinerCount]);
outputRG.initRow(&largeRow);
joinedRG.initRow(&joinedRow);
joinedRG.initRow(&baseJRow, true);
smallRows.reset(new Row[joinerCount]);
for (i = 0; i < joinerCount; i++)
smallSideRGs[i].initRow(&smallRows[i], true);
baseJRowMem.reset(new uint8_t[baseJRow.getSize()]);
baseJRow.setData(baseJRowMem.get());
gjrgMappings.reset(new shared_array<int>[joinerCount + 1]);
for (i = 0; i < joinerCount; i++)
gjrgMappings[i] = makeMapping(smallSideRGs[i], joinedRG);
gjrgMappings[joinerCount] = makeMapping(outputRG, joinedRG);
}
}
// @bug 1051
if (hasFilterStep)
{
for (uint64_t i = 0; i < 2; i++)
{
fFiltRidCount[i] = 0;
fFiltCmdRids[i].reset(new uint16_t[LOGICAL_BLOCK_RIDS]);
fFiltCmdValues[i].reset(new int64_t[LOGICAL_BLOCK_RIDS]);
if (wideColumnsWidths | datatypes::MAXDECIMALWIDTH)
fFiltCmdBinaryValues[i].reset(new int128_t[LOGICAL_BLOCK_RIDS]);
if (filtOnString)
fFiltStrValues[i].reset(new string[LOGICAL_BLOCK_RIDS]);
}
}
/* init the Commands */
if (filterCount > 0)
{
for (i = 0; i < (uint32_t)filterCount - 1; ++i)
{
// cout << "prepping filter " << i << endl;
filterSteps[i]->setBatchPrimitiveProcessor(this);
if (filterSteps[i + 1]->getCommandType() == Command::DICT_STEP)
filterSteps[i]->prep(OT_BOTH, true);
else if (filterSteps[i]->filterFeeder() != Command::NOT_FEEDER)
filterSteps[i]->prep(OT_BOTH, false);
else
filterSteps[i]->prep(OT_RID, false);
}
// cout << "prepping filter " << i << endl;
filterSteps[i]->setBatchPrimitiveProcessor(this);
filterSteps[i]->prep(OT_BOTH, false);
}
for (i = 0; i < projectCount; ++i)
{
// cout << "prepping projection " << i << endl;
projectSteps[i]->setBatchPrimitiveProcessor(this);
if (noVB)
projectSteps[i]->prep(OT_BOTH, false);
else
projectSteps[i]->prep(OT_DATAVALUE, false);
if (0 < filterCount)
{
// if there is an rtscommand with a passThru, the passThru must make its own absRids
// unless there is only one project step, then the last filter step can make absRids
RTSCommand* rts = dynamic_cast<RTSCommand*>(projectSteps[i].get());
if (rts && rts->isPassThru())
{
if (1 == projectCount)
filterSteps[filterCount - 1]->setMakeAbsRids(true);
else
rts->setAbsNull();
}
}
}
if (fAggregator.get() != NULL)
{
// fAggRowGroupData.reset(new uint8_t[fAggregateRG.getMaxDataSize()]);
fAggRowGroupData.reinit(fAggregateRG);
fAggregateRG.setData(&fAggRowGroupData);
if (doJoin)
{
fAggregator->setInputOutput(fe2 ? fe2Output : joinedRG, &fAggregateRG);
fAggregator->setJoinRowGroups(&smallSideRGs, &largeSideRG);
}
else
fAggregator->setInputOutput(fe2 ? fe2Output : outputRG, &fAggregateRG);
}
if (LIKELY(!hasWideColumnOut))
{
minVal = MAX64;
maxVal = MIN64;
}
else
{
max128Val = datatypes::Decimal::minInt128;
min128Val = datatypes::Decimal::maxInt128;
}
// @bug 1269, initialize data used by execute() for async loading blocks
// +1 for the scan filter step with no predicate, if any
relLBID.reset(new uint64_t[projectCount + 1]);
asyncLoaded.reset(new bool[projectCount + 1]);
}
/* This version does a join on projected rows */
// In order to prevent super size result sets in the case of near cartesian joins on three or more joins,
// the startRid start at 0) is used to begin the rid loop and if we cut off processing early because of
// the size of the result set, we return the next rid to start with. If we finish ridCount rids, return 0-
uint32_t BatchPrimitiveProcessor::executeTupleJoin(uint32_t startRid)
{
uint32_t newRowCount = 0, i, j;
vector<uint32_t> matches;
uint64_t largeKey;
uint64_t resultCount = 0;
uint32_t newStartRid = startRid;
outputRG.getRow(0, &oldRow);
outputRG.getRow(0, &newRow);
// cout << "before join, RG has " << outputRG.getRowCount() << " BPP ridcount= " << ridCount << endl;
// ridCount gets modified based on the number of Rids actually processed during this call.
// origRidCount is the number of rids for this thread after filter, which are the total
// number of rids to be processed from all calls to this function during this thread.
for (i = startRid; i < origRidCount && !sendThread->aborted(); i++, oldRow.nextRow())
{
/* Decide whether this large-side row belongs in the output. The breaks
* in the loop mean that it doesn't.
*
* In English the logic is:
* Reject the row if there's no match and it's not an anti or an outer join
* or if there is a match and it's an anti join with no filter.
* If there's an antijoin with a filter nothing can be eliminated at this stage.
* If there's an antijoin where the join op should match NULL values, and there
* are NULL values to match against, but there is no filter, all rows can be eliminated.
*/
// cout << "large side row: " << oldRow.toString() << endl;
for (j = 0; j < joinerCount; j++)
{
bool found;
if (UNLIKELY(joinTypes[j] & ANTI))
{
if (joinTypes[j] & WITHFCNEXP)
continue;
else if (doMatchNulls[j])
break;
}
if (LIKELY(!typelessJoin[j]))
{
// cout << "not typeless join\n";
bool isNull;
uint32_t colIndex = largeSideKeyColumns[j];
if (oldRow.isUnsigned(colIndex))
largeKey = oldRow.getUintField(colIndex);
else
largeKey = oldRow.getIntField(colIndex);
uint bucket = bucketPicker((char*)&largeKey, 8, bpSeed) & ptMask;
bool joinerIsEmpty = tJoiners[j][bucket]->empty() ? true : false;
found = (tJoiners[j][bucket]->find(largeKey) != tJoiners[j][bucket]->end());
isNull = oldRow.isNullValue(colIndex);
/* These conditions define when the row is NOT in the result set:
* - if the key is not in the small side, and the join isn't a large-outer or anti join
* - if the key is NULL, and the join isn't anti- or large-outer
* - if it's an anti-join and the key is either in the small side or it's NULL
*/
if (((!found || isNull) && !(joinTypes[j] & (LARGEOUTER | ANTI))) ||
((joinTypes[j] & ANTI) && !joinerIsEmpty &&
((isNull && (joinTypes[j] & MATCHNULLS)) || (found && !isNull))))
{
// cout << " - not in the result set\n";
break;
}
// else
// cout << " - in the result set\n";
}
else
{
// cout << " typeless join\n";
// the null values are not sent by UM in typeless case. null -> !found
TypelessData tlLargeKey(&oldRow);
uint bucket = oldRow.hashTypeless(tlLargeSideKeyColumns[j], mSmallSideKeyColumnsPtr,
mSmallSideRGPtr ? &mSmallSideRGPtr->getColWidths() : nullptr) &
ptMask;
found = tlJoiners[j][bucket]->find(tlLargeKey) != tlJoiners[j][bucket]->end();
if ((!found && !(joinTypes[j] & (LARGEOUTER | ANTI))) || (joinTypes[j] & ANTI))
{
/* Separated the ANTI join logic for readability.
*
*/
if (joinTypes[j] & ANTI)
{
if (found)
break;
else if (joinTypes[j] & MATCHNULLS)
{
bool hasNull = false;
for (uint32_t z = 0; z < tlLargeSideKeyColumns[j].size(); z++)
if (oldRow.isNullValue(tlLargeSideKeyColumns[j][z]))
{
hasNull = true;
break;
}
if (hasNull) // keys with nulls match everything
break;
else
continue; // non-null keys not in the small side
// are in the result
}
else // signifies a not-exists query
continue;
}
break;
}
}
}
if (j == joinerCount)
{
uint32_t matchCount;
for (j = 0; j < joinerCount; j++)
{
/* The result is already known if...
* -- anti-join with no fcnexp
* -- semi-join with no fcnexp and not scalar
*
* The ANTI join case isn't just a shortcut. getJoinResults() will produce results
* for a different case and generate the wrong result. Need to fix that, later.
*/
if ((joinTypes[j] & (SEMI | ANTI)) && !(joinTypes[j] & WITHFCNEXP) && !(joinTypes[j] & SCALAR))
{
tSmallSideMatches[j][newRowCount].push_back(-1);
continue;
}
getJoinResults(oldRow, j, tSmallSideMatches[j][newRowCount]);
matchCount = tSmallSideMatches[j][newRowCount].size();
if (joinTypes[j] & WITHFCNEXP)
{
vector<uint32_t> newMatches;
applyMapping(joinFEMappings[joinerCount], oldRow, &joinFERow);
for (uint32_t k = 0; k < matchCount; k++)
{
if (tSmallSideMatches[j][newRowCount][k] == (uint32_t)-1)
smallRows[j].setPointer(smallNullPointers[j]);
else
{
smallSideRGs[j].getRow(tSmallSideMatches[j][newRowCount][k], &smallRows[j]);
// uint64_t rowOffset = ((uint64_t) tSmallSideMatches[j][newRowCount][k]) *
// smallRows[j].getSize() + smallSideRGs[j].getEmptySize();
// smallRows[j].setData(&smallSideRowData[j][rowOffset]);
}
applyMapping(joinFEMappings[j], smallRows[j], &joinFERow);
if (joinFEFilters[j]->evaluate(&joinFERow))
{
/* The first match includes it in a SEMI join result and excludes it from an ANTI join
* result. If it's SEMI & SCALAR however, it needs to continue.
*/
newMatches.push_back(tSmallSideMatches[j][newRowCount][k]);
if ((joinTypes[j] & ANTI) || ((joinTypes[j] & (SEMI | SCALAR)) == SEMI))
break;
}
}
tSmallSideMatches[j][newRowCount].swap(newMatches);
matchCount = tSmallSideMatches[j][newRowCount].size();
}
if (matchCount == 0 && (joinTypes[j] & LARGEOUTER))
{
tSmallSideMatches[j][newRowCount].push_back(-1);
matchCount = 1;
}
/* Scalar check */
if ((joinTypes[j] & SCALAR) && matchCount > 1)
throw scalar_exception();
/* Reverse the result for anti-join */
if (joinTypes[j] & ANTI)
{
if (matchCount == 0)
{
tSmallSideMatches[j][newRowCount].push_back(-1);
matchCount = 1;
}
else
{
tSmallSideMatches[j][newRowCount].clear();
matchCount = 0;
}
}
/* For all join types, no matches here means it's not in the result */
if (matchCount == 0)
break;
/* Pair non-scalar semi-joins with a NULL row */
if ((joinTypes[j] & SEMI) && !(joinTypes[j] & SCALAR))
{
tSmallSideMatches[j][newRowCount].clear();
tSmallSideMatches[j][newRowCount].push_back(-1);
matchCount = 1;
}
resultCount += matchCount;
}
/* Finally, copy the row into the output */
if (j == joinerCount)
{
// We need to update 8 and 16 bytes in values and wide128Values buffers
// otherwise unrelated values will be observed in the JOIN-ed output RGData.
if (i != newRowCount)
{
values[newRowCount] = values[i];
if (mJOINHasSkewedKeyColumn)
wide128Values[newRowCount] = wide128Values[i];
relRids[newRowCount] = relRids[i];
copyRow(oldRow, &newRow);
// cout << "joined row: " << newRow.toString() << endl;
}
newRowCount++;
newRow.nextRow();
}
// else
// cout << "j != joinerCount\n";
}
// If we've accumulated more than maxResultCount -- 1048576 (2^20)_ of resultCounts, cut off processing.
// The caller will restart to continue where we left off.
if (resultCount >= maxResultCount)
{
newStartRid += newRowCount;
break;
}
}
if (resultCount < maxResultCount)
newStartRid = 0;
ridCount = newRowCount;
outputRG.setRowCount(ridCount);
/* prints out the whole result set.
if (ridCount != 0) {
cout << "RG rowcount=" << outputRG.getRowCount() << " BPP ridcount=" << ridCount << endl;
for (i = 0; i < joinerCount; i++) {
for (j = 0; j < ridCount; j++) {
cout << "joiner " << i << " has " << tSmallSideMatches[i][j].size() << "
entries" << endl; cout << "row " << j << ":"; for (uint32_t k = 0; k < tSmallSideMatches[i][j].size();
k++) cout << " " << tSmallSideMatches[i][j][k]; cout << endl;
}
cout << endl;
}
}
*/
return newStartRid;
}
#ifdef PRIMPROC_STOPWATCH
void BatchPrimitiveProcessor::execute(StopWatch* stopwatch)
#else
void BatchPrimitiveProcessor::execute()
#endif
{
uint8_t sendCount = 0;
// bool smoreRGs = false;
// uint32_t sStartRid = 0;
uint32_t i, j;
try
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("BatchPrimitiveProcessor::execute first part");
#endif
// if only one scan step which has no predicate, async load all columns
if (filterCount == 1 && hasScan)
{
ColumnCommand* col = dynamic_cast<ColumnCommand*>(filterSteps[0].get());
if ((col != NULL) && (col->getFilterCount() == 0) && (col->getLBID() != 0))
{
// stored in last pos in relLBID[] and asyncLoaded[]
uint64_t p = projectCount;
asyncLoaded[p] = asyncLoaded[p] && (relLBID[p] % blocksReadAhead != 0);
relLBID[p] += col->getWidth();
if (!asyncLoaded[p] && col->willPrefetch())
{
loadBlockAsync(col->getLBID(), versionInfo, txnID, col->getCompType(), &cachedIO, &physIO,
LBIDTrace, sessionID, &counterLock, &busyLoaderCount, sendThread, &vssCache);
asyncLoaded[p] = true;
}
asyncLoadProjectColumns();
}
}
#ifdef PRIMPROC_STOPWATCH
stopwatch->stop("BatchPrimitiveProcessor::execute first part");
stopwatch->start("BatchPrimitiveProcessor::execute second part");
#endif
// filters use relrids and values for intermediate results.
if (bop == BOP_AND)
{
for (j = 0; j < filterCount; ++j)
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("- filterSteps[j]->execute()");
filterSteps[j]->execute();
stopwatch->stop("- filterSteps[j]->execute()");
#else
filterSteps[j]->execute();
#endif
}
}
else // BOP_OR
{
/* XXXPAT: This is a hacky impl of OR logic. Each filter is configured to
be a scan operation on init. This code runs each independently and
unions their output ridlists using accumulator. At the end it turns
accumulator into a final ridlist for subsequent steps.
If there's a join or a passthru command in the projection list, the
values array has to contain values from the last filter step. In that
case, the last filter step isn't part of the "OR" filter processing.
JLF has added it to prep those operations, not to be a filter.
7/7/09 update: the multiple-table join required relocating the join op. It's
no longer necessary to add the loader columncommand to the filter array.
*/
bool accumulator[LOGICAL_BLOCK_RIDS];
// uint32_t realFilterCount = ((forHJ || hasPassThru) ? filterCount - 1 : filterCount);
uint32_t realFilterCount = filterCount;
for (i = 0; i < LOGICAL_BLOCK_RIDS; i++)
accumulator[i] = false;
if (!hasScan) // there are input rids
for (i = 0; i < ridCount; i++)
accumulator[relRids[i]] = true;
ridCount = 0;
for (i = 0; i < realFilterCount; ++i)
{
filterSteps[i]->execute();
if (!filterSteps[i]->filterFeeder())
{
for (j = 0; j < ridCount; j++)
accumulator[relRids[j]] = true;
ridCount = 0;
}
}
for (ridMap = 0, i = 0; i < LOGICAL_BLOCK_RIDS; ++i)
{
if (accumulator[i])
{
relRids[ridCount] = i;
ridMap |= 1 << (relRids[ridCount] >> 9);
++ridCount;
}
}
}
#ifdef PRIMPROC_STOPWATCH
stopwatch->stop("BatchPrimitiveProcessor::execute second part");
stopwatch->start("BatchPrimitiveProcessor::execute third part");
#endif
if (projectCount > 0 || ot == ROW_GROUP)
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("- writeProjectionPreamble");
writeProjectionPreamble();
stopwatch->stop("- writeProjectionPreamble");
#else
writeProjectionPreamble();
#endif
}
// async load blocks for project phase, if not alread loaded
if (ridCount > 0)
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("- asyncLoadProjectColumns");
asyncLoadProjectColumns();
stopwatch->stop("- asyncLoadProjectColumns");
#else
asyncLoadProjectColumns();
#endif
}
#ifdef PRIMPROC_STOPWATCH
stopwatch->stop("BatchPrimitiveProcessor::execute third part");
stopwatch->start("BatchPrimitiveProcessor::execute fourth part");
#endif
// projection commands read relrids and write output directly to a rowgroup
// or the serialized bytestream
if (ot != ROW_GROUP)
{
for (j = 0; j < projectCount; ++j)
{
projectSteps[j]->project();
}
}
else
{
/* Function & Expression group 1 processing
- project for FE1
- execute FE1 row by row
- if return value = true, map input row into the projection RG, adjust ridlist
*/
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("- if(ot != ROW_GROUP) else");
#endif
outputRG.resetRowGroup(baseRid);
if (fe1)
{
uint32_t newRidCount = 0;
fe1Input.resetRowGroup(baseRid);
fe1Input.setRowCount(ridCount);
fe1Input.getRow(0, &fe1In);
outputRG.getRow(0, &fe1Out);
for (j = 0; j < projectCount; j++)
if (projectForFE1[j] != -1)
projectSteps[j]->projectIntoRowGroup(fe1Input, projectForFE1[j]);
for (j = 0; j < ridCount; j++, fe1In.nextRow())
if (fe1->evaluate(&fe1In))
{
applyMapping(fe1ToProjection, fe1In, &fe1Out);
relRids[newRidCount] = relRids[j];
values[newRidCount++] = values[j];
fe1Out.nextRow();
}
ridCount = newRidCount;
}
outputRG.setRowCount(ridCount);
if (sendRidsAtDelivery)
{
Row r;
outputRG.initRow(&r);
outputRG.getRow(0, &r);
for (j = 0; j < ridCount; ++j)
{
r.setRid(relRids[j]);
r.nextRow();
}
}
/* 7/7/09 PL: I Changed the projection alg to reduce block touches when there's
a join. The key columns get projected first, the join is executed to further
reduce the ridlist, then the rest of the columns get projected */
if (!doJoin)
{
for (j = 0; j < projectCount; ++j)
{
// cout << "projectionMap[" << j << "] = " << projectionMap[j] << endl;
if (projectionMap[j] != -1)
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("-- projectIntoRowGroup");
projectSteps[j]->projectIntoRowGroup(outputRG, projectionMap[j]);
stopwatch->stop("-- projectIntoRowGroup");
#else
projectSteps[j]->projectIntoRowGroup(outputRG, projectionMap[j]);
#endif
}
// else
// cout << " no target found for OID " << projectSteps[j]->getOID() <<
//endl;
}
if (fe2)
{
/* functionize this -> processFE2() */
fe2Output.resetRowGroup(baseRid);
fe2Output.getRow(0, &fe2Out);
fe2Input->getRow(0, &fe2In);
// cerr << "input row: " << fe2In.toString() << endl;
for (j = 0; j < outputRG.getRowCount(); j++, fe2In.nextRow())
{
if (fe2->evaluate(&fe2In))
{
applyMapping(fe2Mapping, fe2In, &fe2Out);
// cerr << " passed. output row: " << fe2Out.toString() << endl;
fe2Out.setRid(fe2In.getRelRid());
fe2Output.incRowCount();
fe2Out.nextRow();
}
}
if (!fAggregator)
{
*serialized << (uint8_t)1; // the "count this msg" var
fe2Output.setDBRoot(dbRoot);
fe2Output.serializeRGData(*serialized);
//*serialized << fe2Output.getDataSize();
// serialized->append(fe2Output.getData(), fe2Output.getDataSize());
}
}
if (fAggregator)
{
*serialized << (uint8_t)1; // the "count this msg" var
RowGroup& toAggregate = (fe2 ? fe2Output : outputRG);
// toAggregate.convertToInlineDataInPlace();
if (fe2)
fe2Output.setDBRoot(dbRoot);
else
outputRG.setDBRoot(dbRoot);
fAggregator->addRowGroup(&toAggregate);
if ((currentBlockOffset + 1) == count) // @bug4507, 8k
{
fAggregator->loadResult(*serialized); // @bug4507, 8k
} // @bug4507, 8k
else if (utils::MonitorProcMem::isMemAvailable()) // @bug4507, 8k
{
fAggregator->loadEmptySet(*serialized); // @bug4507, 8k
} // @bug4507, 8k
else // @bug4507, 8k
{
fAggregator->loadResult(*serialized); // @bug4507, 8k
fAggregator->aggReset(); // @bug4507, 8k
} // @bug4507, 8k
}
if (!fAggregator && !fe2)
{
*serialized << (uint8_t)1; // the "count this msg" var
outputRG.setDBRoot(dbRoot);
// cerr << "serializing " << outputRG.toString() << endl;
outputRG.serializeRGData(*serialized);
//*serialized << outputRG.getDataSize();
// serialized->append(outputRG.getData(), outputRG.getDataSize());
}
#ifdef PRIMPROC_STOPWATCH
stopwatch->stop("- if(ot != ROW_GROUP) else");
#endif
}
else // Is doJoin
{
uint32_t startRid = 0;
ByteStream preamble = *serialized;
origRidCount = ridCount; // ridCount can get modified by executeTupleJoin(). We need to keep track of
// the original val.
/* project the key columns. If there's the filter IN the join, project everything.
Also need to project 'long' strings b/c executeTupleJoin may copy entire rows
using copyRow(), which will try to interpret the uninit'd string ptr.
Valgrind will legitimately complain about copying uninit'd values for the
other types but that is technically safe. */
for (j = 0; j < projectCount; j++)
{
if (keyColumnProj[j] ||
(projectionMap[j] != -1 && (hasJoinFEFilters || oldRow.isLongString(projectionMap[j]))))
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("-- projectIntoRowGroup");
projectSteps[j]->projectIntoRowGroup(outputRG, projectionMap[j]);
stopwatch->stop("-- projectIntoRowGroup");
#else
projectSteps[j]->projectIntoRowGroup(outputRG, projectionMap[j]);
#endif
}
}
do // while (startRid > 0)
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("-- executeTupleJoin()");
startRid = executeTupleJoin(startRid);
stopwatch->stop("-- executeTupleJoin()");
#else
startRid = executeTupleJoin(startRid);
// sStartRid = startRid;
#endif
/* project the non-key columns */
for (j = 0; j < projectCount; ++j)
{
if (projectionMap[j] != -1 && !keyColumnProj[j] && !hasJoinFEFilters &&
!oldRow.isLongString(projectionMap[j]))
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("-- projectIntoRowGroup");
projectSteps[j]->projectIntoRowGroup(outputRG, projectionMap[j]);
stopwatch->stop("-- projectIntoRowGroup");
#else
projectSteps[j]->projectIntoRowGroup(outputRG, projectionMap[j]);
#endif
}
}
/* The RowGroup is fully joined at this point.
* Add additional RowGroup processing here.
* TODO: Try to clean up all of the switching */
if (fe2 || fAggregator)
{
bool moreRGs = true;
initGJRG();
while (moreRGs && !sendThread->aborted())
{
/*
* generate 1 rowgroup (8192 rows max) of joined rows
* if there's an FE2, run it
* -pack results into a new rowgroup
* -if there are < 8192 rows in the new RG, continue
* if there's an agg, run it
* send the result
*/
resetGJRG();
moreRGs = generateJoinedRowGroup(baseJRow);
// smoreRGs = moreRGs;
sendCount = (uint8_t)(!moreRGs && !startRid);
// *serialized << (uint8_t)(!moreRGs && !startRid); // the "count
// this msg" var
*serialized << sendCount;
if (fe2)
{
/* functionize this -> processFE2()*/
fe2Output.resetRowGroup(baseRid);
fe2Output.setDBRoot(dbRoot);
fe2Output.getRow(0, &fe2Out);
fe2Input->getRow(0, &fe2In);
for (j = 0; j < joinedRG.getRowCount(); j++, fe2In.nextRow())
{
if (fe2->evaluate(&fe2In))
{
applyMapping(fe2Mapping, fe2In, &fe2Out);
fe2Out.setRid(fe2In.getRelRid());
fe2Output.incRowCount();
fe2Out.nextRow();
}
}
}
RowGroup& nextRG = (fe2 ? fe2Output : joinedRG);
nextRG.setDBRoot(dbRoot);
if (fAggregator)
{
fAggregator->addRowGroup(&nextRG);
if ((currentBlockOffset + 1) == count && moreRGs == false && startRid == 0) // @bug4507, 8k
{
fAggregator->loadResult(*serialized); // @bug4507, 8k
} // @bug4507, 8k
else if (utils::MonitorProcMem::isMemAvailable()) // @bug4507, 8k
{
fAggregator->loadEmptySet(*serialized); // @bug4507, 8k
} // @bug4507, 8k
else // @bug4507, 8k
{
fAggregator->loadResult(*serialized); // @bug4507, 8k
fAggregator->aggReset(); // @bug4507, 8k
} // @bug4507, 8k
}
else
{
// cerr <<" * serialzing " << nextRG.toString() << endl;
nextRG.serializeRGData(*serialized);
}
/* send the msg & reinit the BS */
if (moreRGs)
{
sendResponse();
serialized.reset(new ByteStream());
*serialized = preamble;
}
}
if (hasSmallOuterJoin)
{
// Should we happen to finish sending data rows right on the boundary of when moreRGs flips off,
// then we need to start a new buffer. I.e., it needs the count this message byte pushed.
if (serialized->length() == preamble.length())
*serialized << (uint8_t)(startRid > 0 ? 0 : 1); // the "count this msg" var
*serialized << ridCount;
for (i = 0; i < joinerCount; i++)
{
for (j = 0; j < ridCount; ++j)
{
serializeInlineVector<uint32_t>(*serialized, tSmallSideMatches[i][j]);
tSmallSideMatches[i][j].clear();
}
}
}
else
{
// We hae no more use for this allocation
for (i = 0; i < joinerCount; i++)
for (j = 0; j < ridCount; ++j)
tSmallSideMatches[i][j].clear();
}
}
else
{
*serialized << (uint8_t)(startRid > 0 ? 0 : 1); // the "count this msg" var
outputRG.setDBRoot(dbRoot);
// cerr << "serializing " << outputRG.toString() << endl;
outputRG.serializeRGData(*serialized);
//*serialized << outputRG.getDataSize();
// serialized->append(outputRG.getData(), outputRG.getDataSize());
for (i = 0; i < joinerCount; i++)
{
for (j = 0; j < ridCount; ++j)
{
serializeInlineVector<uint32_t>(*serialized, tSmallSideMatches[i][j]);
tSmallSideMatches[i][j].clear();
}
}
}
if (startRid > 0)
{
sendResponse();
serialized.reset(new ByteStream());
*serialized = preamble;
}
} while (startRid > 0);
}
#ifdef PRIMPROC_STOPWATCH
stopwatch->stop("- if(ot != ROW_GROUP) else");
#endif
}
ridCount = origRidCount; // May not be needed, but just to be safe.
// std::cout << "end of send. startRid=" << sStartRid << " moreRG=" << smoreRGs << " sendCount=" <<
// sendCount << std::endl;
if (projectCount > 0 || ot == ROW_GROUP)
{
*serialized << cachedIO;
cachedIO = 0;
*serialized << physIO;
physIO = 0;
*serialized << touchedBlocks;
touchedBlocks = 0;
// cout << "sent physIO=" << physIO << " cachedIO=" << cachedIO <<
// " touchedBlocks=" << touchedBlocks << endl;
}
#ifdef PRIMPROC_STOPWATCH
stopwatch->stop("BatchPrimitiveProcessor::execute fourth part");
#endif
}
catch (logging::QueryDataExcept& qex)
{
writeErrorMsg(qex.what(), qex.errorCode());
}
catch (logging::DictionaryBufferOverflow& db)
{
writeErrorMsg(db.what(), db.errorCode());
}
catch (scalar_exception& se)
{
writeErrorMsg(IDBErrorInfo::instance()->errorMsg(ERR_MORE_THAN_1_ROW), ERR_MORE_THAN_1_ROW, false);
}
catch (NeedToRestartJob& n)
{
#if 0
/* This block of code will flush the problematic OIDs from the
* cache. It seems to have no effect on the problem, so it's commented
* for now.
*
* This is currently thrown only on syscat queries. If we find the problem
* in user tables also, we should avoid dropping entire OIDs if possible.
*
* In local testing there was no need for flushing, because DDL flushes
* the syscat constantly. However, it can take a long time (>10 s) before
* that happens. Doing it locally should make it much more likely only
* one restart is necessary.
*/
try
{
vector<uint32_t> oids;
uint32_t oid;
for (uint32_t i = 0; i < filterCount; i++)
{
oid = filterSteps[i]->getOID();
if (oid > 0)
oids.push_back(oid);
}
for (uint32_t i = 0; i < projectCount; i++)
{
oid = projectSteps[i]->getOID();
if (oid > 0)
oids.push_back(oid);
}
#if 0
Logger logger;
ostringstream os;
os << "dropping OIDs: ";
for (int i = 0; i < oids.size(); i++)
os << oids[i] << " ";
logger.logMessage(os.str());
#endif
for (int i = 0; i < fCacheCount; i++)
{
dbbc::blockCacheClient bc(*BRPp[i]);
// bc.flushCache();
bc.flushOIDs(&oids[0], oids.size());
}
}
catch (...) { } // doesn't matter if this fails, just avoid crashing
#endif
#ifndef __FreeBSD__
pthread_mutex_unlock(&objLock);
#endif
throw n; // need to pass this through to BPPSeeder
}
catch (IDBExcept& iex)
{
writeErrorMsg(iex.what(), iex.errorCode(), true, false);
}
catch (const std::exception& ex)
{
writeErrorMsg(ex.what(), logging::batchPrimitiveProcessorErr);
}
catch (...)
{
string msg("BatchPrimitiveProcessor caught an unknown exception");
writeErrorMsg(msg, logging::batchPrimitiveProcessorErr);
}
}
void BatchPrimitiveProcessor::writeErrorMsg(const string& error, uint16_t errCode, bool logIt, bool critical)
{
ISMPacketHeader ism;
PrimitiveHeader ph;
// we don't need every field of these headers. Init'ing them anyway
// makes memory checkers happy.
void* ismp = static_cast<void*>(&ism);
void* php = static_cast<void*>(&ph);
memset(ismp, 0, sizeof(ISMPacketHeader));
memset(php, 0, sizeof(PrimitiveHeader));
ph.SessionID = sessionID;
ph.StepID = stepID;
ph.UniqueID = uniqueID;
ism.Status = errCode;
serialized.reset(new ByteStream());
serialized->append((uint8_t*)&ism, sizeof(ism));
serialized->append((uint8_t*)&ph, sizeof(ph));
*serialized << error;
if (logIt)
{
Logger log;
log.logMessage(error, critical);
}
}
void BatchPrimitiveProcessor::writeProjectionPreamble()
{
ISMPacketHeader ism;
PrimitiveHeader ph;
// we don't need every field of these headers. Init'ing them anyway
// makes memory checkers happy.
void* ismp = static_cast<void*>(&ism);
void* php = static_cast<void*>(&ph);
memset(ismp, 0, sizeof(ISMPacketHeader));
memset(php, 0, sizeof(PrimitiveHeader));
ph.SessionID = sessionID;
ph.StepID = stepID;
ph.UniqueID = uniqueID;
serialized.reset(new ByteStream());
serialized->append((uint8_t*)&ism, sizeof(ism));
serialized->append((uint8_t*)&ph, sizeof(ph));
/* add-ons */
if (hasScan)
{
if (validCPData)
{
*serialized << (uint8_t)1;
*serialized << lbidForCP;
*serialized << ((uint8_t)cpDataFromDictScan);
if (UNLIKELY(hasWideColumnOut))
{
// PSA width
*serialized << (uint8_t)wideColumnWidthOut;
*serialized << min128Val;
*serialized << max128Val;
}
else
{
*serialized << (uint8_t)utils::MAXLEGACYWIDTH; // width of min/max value
*serialized << (uint64_t)minVal;
*serialized << (uint64_t)maxVal;
}
}
else
{
*serialized << (uint8_t)0;
*serialized << lbidForCP;
}
}
// ridsOut += ridCount;
/* results */
if (ot != ROW_GROUP)
{
*serialized << ridCount;
if (sendRidsAtDelivery)
{
*serialized << baseRid;
serialized->append((uint8_t*)relRids, ridCount << 1);
}
}
}
void BatchPrimitiveProcessor::serializeElementTypes()
{
*serialized << baseRid;
*serialized << ridCount;
serialized->append((uint8_t*)relRids, ridCount << 1);
serialized->append((uint8_t*)values, ridCount << 3);
}
void BatchPrimitiveProcessor::serializeStrings()
{
*serialized << ridCount;
serialized->append((uint8_t*)absRids.get(), ridCount << 3);
for (uint32_t i = 0; i < ridCount; ++i)
*serialized << strValues[i];
}
void BatchPrimitiveProcessor::sendResponse()
{
if (sendThread->flowControlEnabled())
{
// newConnection should be set only for the first result of a batch job
// it tells sendthread it should consider it for the connection array
sendThread->sendResult(BPPSendThread::Msg_t(serialized, sock, writelock, sockIndex), newConnection);
newConnection = false;
}
else
{
boost::mutex::scoped_lock lk(*writelock);
sock->write(*serialized);
}
serialized.reset();
}
/* The output of a filter chain is either ELEMENT_TYPE or STRING_ELEMENT_TYPE */
void BatchPrimitiveProcessor::makeResponse()
{
ISMPacketHeader ism;
PrimitiveHeader ph;
// we don't need every field of these headers. Init'ing them anyway
// makes memory checkers happy.
void* ismp = static_cast<void*>(&ism);
void* php = static_cast<void*>(&ph);
memset(ismp, 0, sizeof(ISMPacketHeader));
memset(php, 0, sizeof(PrimitiveHeader));
ph.SessionID = sessionID;
ph.StepID = stepID;
ph.UniqueID = uniqueID;
serialized.reset(new ByteStream());
serialized->append((uint8_t*)&ism, sizeof(ism));
serialized->append((uint8_t*)&ph, sizeof(ph));
/* add-ons */
if (hasScan)
{
if (validCPData)
{
*serialized << (uint8_t)1;
*serialized << lbidForCP;
*serialized << ((uint8_t)cpDataFromDictScan);
if (UNLIKELY(hasWideColumnOut))
{
// PSA width
// Remove the assert for >16 bytes DTs.
assert(wideColumnWidthOut == datatypes::MAXDECIMALWIDTH);
*serialized << (uint8_t)wideColumnWidthOut;
*serialized << min128Val;
*serialized << max128Val;
}
else
{
*serialized << (uint8_t)utils::MAXLEGACYWIDTH; // width of min/max value
*serialized << (uint64_t)minVal;
*serialized << (uint64_t)maxVal;
}
}
else
{
*serialized << (uint8_t)0;
*serialized << lbidForCP;
}
}
/* results */
/* Take the rid and value arrays, munge into OutputType ot */
switch (ot)
{
case BPS_ELEMENT_TYPE: serializeElementTypes(); break;
case STRING_ELEMENT_TYPE: serializeStrings(); break;
default:
{
ostringstream oss;
oss << "BPP: makeResponse(): Bad output type: " << ot;
throw logic_error(oss.str());
}
// throw logic_error("BPP: makeResponse(): Bad output type");
}
*serialized << cachedIO;
cachedIO = 0;
*serialized << physIO;
physIO = 0;
*serialized << touchedBlocks;
touchedBlocks = 0;
// cout << "sent physIO=" << physIO << " cachedIO=" << cachedIO <<
// " touchedBlocks=" << touchedBlocks << endl;
}
int BatchPrimitiveProcessor::operator()()
{
utils::setThreadName("PPBatchPrimProc");
#ifdef PRIMPROC_STOPWATCH
const static std::string msg{"BatchPrimitiveProcessor::operator()"};
logging::StopWatch* stopwatch = profiler.getTimer();
#endif
if (currentBlockOffset == 0)
{
#ifdef PRIMPROC_STOPWATCH
stopwatch->start(msg);
#endif
idbassert(count > 0);
}
if (fAggregator && currentBlockOffset == 0) // @bug4507, 8k
fAggregator->aggReset(); // @bug4507, 8k
for (; currentBlockOffset < count; currentBlockOffset++)
{
if (!(sessionID & 0x80000000)) // can't do this with syscat queries
{
if (sendThread->aborted())
break;
if (sendThread->sizeTooBig())
{
// The send buffer is full of messages yet to be sent, so this thread would block anyway.
freeLargeBuffers();
return -1; // the reschedule error code
}
}
allocLargeBuffers();
if (LIKELY(!hasWideColumnOut))
{
minVal = MAX64;
maxVal = MIN64;
}
else
{
max128Val = datatypes::Decimal::minInt128;
min128Val = datatypes::Decimal::maxInt128;
}
validCPData = false;
cpDataFromDictScan = false;
#ifdef PRIMPROC_STOPWATCH
stopwatch->start("BPP() execute");
execute(stopwatch);
stopwatch->stop("BPP() execute");
#else
execute();
#endif
if (projectCount == 0 && ot != ROW_GROUP)
makeResponse();
try
{
sendResponse();
}
catch (std::exception& e)
{
break; // If we make this throw, be sure to do the cleanup at the end
}
// Bug 4475: Control outgoing socket so that all messages from a
// batch go out the same socket
sockIndex = (sockIndex + 1) % connectionsPerUM;
nextLBID();
/* Base RIDs are now a combination of partition#, segment#, extent#, and block#. */
uint32_t partNum;
uint16_t segNum;
uint8_t extentNum;
uint16_t blockNum;
rowgroup::getLocationFromRid(baseRid, &partNum, &segNum, &extentNum, &blockNum);
/*
cout << "baseRid=" << baseRid << " partNum=" << partNum << " segNum=" << segNum <<
" extentNum=" << (int) extentNum
<< " blockNum=" << blockNum << endl;
*/
blockNum++;
baseRid = rowgroup::convertToRid(partNum, segNum, extentNum, blockNum);
/*
cout << "-- baseRid=" << baseRid << " partNum=" << partNum << " extentNum=" << (int) extentNum
<< " blockNum=" << blockNum << endl;
*/
}
vssCache.clear();
#ifndef __FreeBSD__
pthread_mutex_unlock(&objLock);
#endif
freeLargeBuffers();
#ifdef PRIMPROC_STOPWATCH
stopwatch->stop(msg);
#endif
// cout << "sent " << count << " responses" << endl;
fBusy = false;
return 0;
}
void BatchPrimitiveProcessor::allocLargeBuffers()
{
if (ot == ROW_GROUP && !outRowGroupData)
{
// outputRG.setUseStringTable(true);
outRowGroupData.reset(new RGData(outputRG));
outputRG.setData(outRowGroupData.get());
}
if (fe1 && !fe1Data)
{
// fe1Input.setUseStringTable(true);
fe1Data.reset(new RGData(fe1Input));
// fe1Data.reset(new uint8_t[fe1Input.getMaxDataSize()]);
fe1Input.setData(fe1Data.get());
}
if (fe2 && !fe2Data)
{
// fe2Output.setUseStringTable(true);
fe2Data.reset(new RGData(fe2Output));
fe2Output.setData(fe2Data.get());
}
if (getTupleJoinRowGroupData && !joinedRGMem)
{
// joinedRG.setUseStringTable(true);
joinedRGMem.reset(new RGData(joinedRG));
joinedRG.setData(joinedRGMem.get());
}
}
void BatchPrimitiveProcessor::freeLargeBuffers()
{
/* Get rid of large buffers */
if (ot == ROW_GROUP && outputRG.getMaxDataSizeWithStrings() > maxIdleBufferSize)
outRowGroupData.reset();
if (fe1 && fe1Input.getMaxDataSizeWithStrings() > maxIdleBufferSize)
fe1Data.reset();
if (fe2 && fe2Output.getMaxDataSizeWithStrings() > maxIdleBufferSize)
fe2Data.reset();
if (getTupleJoinRowGroupData && joinedRG.getMaxDataSizeWithStrings() > maxIdleBufferSize)
joinedRGMem.reset();
}
void BatchPrimitiveProcessor::nextLBID()
{
uint32_t i;
for (i = 0; i < filterCount; i++)
filterSteps[i]->nextLBID();
for (i = 0; i < projectCount; i++)
projectSteps[i]->nextLBID();
}
SBPP BatchPrimitiveProcessor::duplicate()
{
SBPP bpp;
uint32_t i;
// cout << "duplicating a bpp\n";
bpp.reset(new BatchPrimitiveProcessor());
bpp->ot = ot;
bpp->versionInfo = versionInfo;
bpp->txnID = txnID;
bpp->sessionID = sessionID;
bpp->stepID = stepID;
bpp->uniqueID = uniqueID;
bpp->needStrValues = needStrValues;
bpp->wideColumnsWidths = wideColumnsWidths;
bpp->gotAbsRids = gotAbsRids;
bpp->gotValues = gotValues;
bpp->LBIDTrace = LBIDTrace;
bpp->hasScan = hasScan;
bpp->hasFilterStep = hasFilterStep;
bpp->filtOnString = filtOnString;
bpp->hasRowGroup = hasRowGroup;
bpp->getTupleJoinRowGroupData = getTupleJoinRowGroupData;
bpp->bop = bop;
bpp->hasPassThru = hasPassThru;
bpp->forHJ = forHJ;
bpp->processorThreads = processorThreads; // is a power-of-2 at this point
bpp->ptMask = processorThreads - 1;
if (ot == ROW_GROUP)
{
bpp->outputRG = outputRG;
if (fe1)
{
bpp->fe1.reset(new FuncExpWrapper(*fe1));
bpp->fe1Input = fe1Input;
}
if (fe2)
{
bpp->fe2.reset(new FuncExpWrapper(*fe2));
bpp->fe2Output = fe2Output;
}
}
bpp->doJoin = doJoin;
if (doJoin)
{
pthread_mutex_lock(&bpp->objLock);
/* There are add'l join vars, but only these are necessary for processing
a join */
bpp->tJoinerSizes = tJoinerSizes;
bpp->joinerCount = joinerCount;
bpp->joinTypes = joinTypes;
bpp->largeSideKeyColumns = largeSideKeyColumns;
bpp->tJoiners = tJoiners;
// bpp->_pools = _pools;
bpp->typelessJoin = typelessJoin;
bpp->tlLargeSideKeyColumns = tlLargeSideKeyColumns;
bpp->tlSmallSideKeyColumns = tlSmallSideKeyColumns;
bpp->tlJoiners = tlJoiners;
bpp->tlSmallSideKeyLengths = tlSmallSideKeyLengths;
bpp->storedKeyAllocators = storedKeyAllocators;
bpp->joinNullValues = joinNullValues;
bpp->doMatchNulls = doMatchNulls;
bpp->hasJoinFEFilters = hasJoinFEFilters;
bpp->hasSmallOuterJoin = hasSmallOuterJoin;
bpp->mJOINHasSkewedKeyColumn = mJOINHasSkewedKeyColumn;
bpp->mSmallSideRGPtr = mSmallSideRGPtr;
bpp->mSmallSideKeyColumnsPtr = mSmallSideKeyColumnsPtr;
if (!getTupleJoinRowGroupData && mJOINHasSkewedKeyColumn)
{
idbassert(!smallSideRGs.empty());
bpp->smallSideRGs.push_back(smallSideRGs[0]);
}
if (hasJoinFEFilters)
{
bpp->joinFERG = joinFERG;
bpp->joinFEFilters.reset(new scoped_ptr<FuncExpWrapper>[joinerCount]);
for (i = 0; i < joinerCount; i++)
if (joinFEFilters[i])
bpp->joinFEFilters[i].reset(new FuncExpWrapper(*joinFEFilters[i]));
}
if (getTupleJoinRowGroupData)
{
bpp->smallSideRGs = smallSideRGs;
bpp->largeSideRG = largeSideRG;
bpp->smallSideRowLengths = smallSideRowLengths;
bpp->smallSideRowData = smallSideRowData;
bpp->smallNullRowData = smallNullRowData;
bpp->smallNullPointers = smallNullPointers;
bpp->joinedRG = joinedRG;
}
#ifdef __FreeBSD__
pthread_mutex_unlock(&bpp->objLock);
#endif
}
bpp->filterCount = filterCount;
bpp->filterSteps.resize(filterCount);
for (i = 0; i < filterCount; ++i)
bpp->filterSteps[i] = filterSteps[i]->duplicate();
bpp->projectCount = projectCount;
bpp->projectSteps.resize(projectCount);
for (i = 0; i < projectCount; ++i)
bpp->projectSteps[i] = projectSteps[i]->duplicate();
if (fAggregator.get() != NULL)
{
bpp->fAggregateRG = fAggregateRG;
bpp->fAggregator.reset(new RowAggregation(fAggregator->getGroupByCols(), fAggregator->getAggFunctions()));
bpp->fAggregator->timeZone(fAggregator->timeZone());
}
bpp->sendRidsAtDelivery = sendRidsAtDelivery;
bpp->prefetchThreshold = prefetchThreshold;
bpp->sock = sock;
bpp->writelock = writelock;
bpp->hasDictStep = hasDictStep;
bpp->sendThread = sendThread;
bpp->newConnection = true;
bpp->initProcessor();
return bpp;
}
#if 0
bool BatchPrimitiveProcessor::operator==(const BatchPrimitiveProcessor& bpp) const
{
uint32_t i;
if (ot != bpp.ot)
return false;
if (versionInfo != bpp.versionInfo)
return false;
if (txnID != bpp.txnID)
return false;
if (sessionID != bpp.sessionID)
return false;
if (stepID != bpp.stepID)
return false;
if (uniqueID != bpp.uniqueID)
return false;
if (gotValues != bpp.gotValues)
return false;
if (gotAbsRids != bpp.gotAbsRids)
return false;
if (needStrValues != bpp.needStrValues)
return false;
if (filterCount != bpp.filterCount)
return false;
if (projectCount != bpp.projectCount)
return false;
if (sendRidsAtDelivery != bpp.sendRidsAtDelivery)
return false;
if (hasScan != bpp.hasScan)
return false;
if (hasFilterStep != bpp.hasFilterStep)
return false;
if (filtOnString != bpp.filtOnString)
return false;
if (doJoin != bpp.doJoin)
return false;
if (doJoin)
/* Join equality test is a bit out of date */
if (joiner != bpp.joiner || joinerSize != bpp.joinerSize)
return false;
for (i = 0; i < filterCount; i++)
if (*filterSteps[i] != *bpp.filterSteps[i])
return false;
return true;
}
#endif
void BatchPrimitiveProcessor::asyncLoadProjectColumns()
{
// relLBID is the LBID related to the primMsg->LBID,
// it is used to keep the read ahead boundary for asyncLoads
// 1. scan driven case: load blocks in # to (# + blocksReadAhead - 1) range,
// where # is a multiple of ColScanReadAheadBlocks in Columnstore.xml
// 2. non-scan driven case: load blocks in the logical block.
// because 1 logical block per primMsg, asyncLoad only once per message.
for (uint64_t i = 0; i < projectCount; ++i)
{
// only care about column commands
ColumnCommand* col = dynamic_cast<ColumnCommand*>(projectSteps[i].get());
if (col != NULL)
{
asyncLoaded[i] = asyncLoaded[i] && (relLBID[i] % blocksReadAhead != 0);
relLBID[i] += col->getWidth();
if (!asyncLoaded[i] && col->willPrefetch())
{
loadBlockAsync(col->getLBID(), versionInfo, txnID, col->getCompType(), &cachedIO, &physIO, LBIDTrace,
sessionID, &counterLock, &busyLoaderCount, sendThread, &vssCache);
asyncLoaded[i] = true;
}
}
}
}
bool BatchPrimitiveProcessor::generateJoinedRowGroup(rowgroup::Row& baseRow, const uint32_t depth)
{
Row& smallRow = smallRows[depth];
const bool lowestLvl = (depth == joinerCount - 1);
while (gjrgRowNumber < ridCount &&
gjrgPlaceHolders[depth] < tSmallSideMatches[depth][gjrgRowNumber].size() && !gjrgFull)
{
const vector<uint32_t>& results = tSmallSideMatches[depth][gjrgRowNumber];
const uint32_t size = results.size();
if (depth == 0)
{
outputRG.getRow(gjrgRowNumber, &largeRow);
applyMapping(gjrgMappings[joinerCount], largeRow, &baseRow);
baseRow.setRid(largeRow.getRelRid());
}
// cout << "rowNum = " << gjrgRowNumber << " at depth " << depth << " size is " << size << endl;
for (uint32_t& i = gjrgPlaceHolders[depth]; i < size && !gjrgFull; i++)
{
if (results[i] != (uint32_t)-1)
{
smallSideRGs[depth].getRow(results[i], &smallRow);
// rowOffset = ((uint64_t) results[i]) * smallRowSize;
// smallRow.setData(&rowDataAtThisLvl.rowData[rowOffset] + emptySize);
}
else
smallRow.setPointer(smallNullPointers[depth]);
// cout << "small row: " << smallRow.toString() << endl;
applyMapping(gjrgMappings[depth], smallRow, &baseRow);
if (!lowestLvl)
generateJoinedRowGroup(baseRow, depth + 1);
else
{
copyRow(baseRow, &joinedRow);
// memcpy(joinedRow.getData(), baseRow.getData(), joinedRow.getSize());
// cerr << "joined row " << joinedRG.getRowCount() << ": " << joinedRow.toString() << endl;
joinedRow.nextRow();
joinedRG.incRowCount();
if (joinedRG.getRowCount() == 8192)
{
i++;
gjrgFull = true;
}
}
if (gjrgFull)
break;
}
if (depth == 0 && gjrgPlaceHolders[0] == tSmallSideMatches[0][gjrgRowNumber].size())
{
gjrgPlaceHolders[0] = 0;
gjrgRowNumber++;
}
}
// if (depth == 0)
// cout << "gjrg returning " << (uint32_t) gjrgFull << endl;
if (!gjrgFull)
gjrgPlaceHolders[depth] = 0;
return gjrgFull;
}
void BatchPrimitiveProcessor::resetGJRG()
{
gjrgFull = false;
joinedRG.resetRowGroup(baseRid);
joinedRG.getRow(0, &joinedRow);
joinedRG.setDBRoot(dbRoot);
}
void BatchPrimitiveProcessor::initGJRG()
{
for (uint32_t z = 0; z < joinerCount; z++)
gjrgPlaceHolders[z] = 0;
gjrgRowNumber = 0;
}
inline void BatchPrimitiveProcessor::getJoinResults(const Row& r, uint32_t jIndex, vector<uint32_t>& v)
{
uint bucket;
if (!typelessJoin[jIndex])
{
if (r.isNullValue(largeSideKeyColumns[jIndex]))
{
/* Bug 3524. This matches everything. */
if (joinTypes[jIndex] & ANTI)
{
TJoiner::iterator it;
for (uint i = 0; i < processorThreads; ++i)
for (it = tJoiners[jIndex][i]->begin(); it != tJoiners[jIndex][i]->end(); ++it)
v.push_back(it->second);
return;
}
else
return;
}
uint64_t largeKey;
uint32_t colIndex = largeSideKeyColumns[jIndex];
if (r.isUnsigned(colIndex))
{
largeKey = r.getUintField(colIndex);
}
else
{
largeKey = r.getIntField(colIndex);
}
bucket = bucketPicker((char*)&largeKey, 8, bpSeed) & ptMask;
pair<TJoiner::iterator, TJoiner::iterator> range = tJoiners[jIndex][bucket]->equal_range(largeKey);
for (; range.first != range.second; ++range.first)
v.push_back(range.first->second);
if (doMatchNulls[jIndex]) // add the nulls to the match list
{
bucket = bucketPicker((char*)&joinNullValues[jIndex], 8, bpSeed) & ptMask;
range = tJoiners[jIndex][bucket]->equal_range(joinNullValues[jIndex]);
for (; range.first != range.second; ++range.first)
v.push_back(range.first->second);
}
}
else
{
/* Bug 3524. Large-side NULL + ANTI join matches everything. */
if (joinTypes[jIndex] & ANTI)
{
bool hasNullValue = false;
for (uint32_t i = 0; i < tlLargeSideKeyColumns[jIndex].size(); i++)
if (r.isNullValue(tlLargeSideKeyColumns[jIndex][i]))
{
hasNullValue = true;
break;
}
if (hasNullValue)
{
TLJoiner::iterator it;
for (uint i = 0; i < processorThreads; ++i)
for (it = tlJoiners[jIndex][i]->begin(); it != tlJoiners[jIndex][i]->end(); ++it)
v.push_back(it->second);
return;
}
}
TypelessData largeKey(&r);
bucket = r.hashTypeless(tlLargeSideKeyColumns[jIndex], mSmallSideKeyColumnsPtr,
mSmallSideRGPtr ? &mSmallSideRGPtr->getColWidths() : nullptr) &
ptMask;
pair<TLJoiner::iterator, TLJoiner::iterator> range = tlJoiners[jIndex][bucket]->equal_range(largeKey);
for (; range.first != range.second; ++range.first)
v.push_back(range.first->second);
}
}
void BatchPrimitiveProcessor::buildVSSCache(uint32_t loopCount)
{
vector<int64_t> lbidList;
vector<BRM::VSSData> vssData;
uint32_t i;
int rc;
for (i = 0; i < filterCount; i++)
filterSteps[i]->getLBIDList(loopCount, &lbidList);
for (i = 0; i < projectCount; i++)
projectSteps[i]->getLBIDList(loopCount, &lbidList);
rc = brm->bulkVSSLookup(lbidList, versionInfo, (int)txnID, &vssData);
if (rc == 0)
for (i = 0; i < vssData.size(); i++)
vssCache.insert(make_pair(lbidList[i], vssData[i]));
// cout << "buildVSSCache inserted " << vssCache.size() << " elements" << endl;
}
} // namespace primitiveprocessor
View Git Blame Copy Permalink