mirror of
https://github.com/mariadb-corporation/mariadb-columnstore-engine.git
synced 2025-04-18 21:44:02 +03:00
738 lines
26 KiB
C++
738 lines
26 KiB
C++
/* Copyright (C) 2014 InfiniDB, Inc.
|
|
|
|
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: we_colbufmgr.cpp 4726 2013-08-07 03:38:36Z bwilkinson $
|
|
*
|
|
****************************************************************************/
|
|
|
|
/** @file
|
|
* ColumnBufferManager methods used to coordinate writes to db column files.
|
|
*
|
|
* ColumnBufferManager methods collect data in an internal buffer from
|
|
* multiple threads, and then periodically flushes the buffer to the relevant
|
|
* DB column segment file. However, if the output is to be compressed, then
|
|
* instead of flushing the buffer to disk, the data is instead copied to a to-
|
|
* be-compressed buffer, where it is collected and later flushed to disk by
|
|
* ColumnBufferCompressed.
|
|
*/
|
|
|
|
#include "we_colbufmgr.h"
|
|
#include "we_colbuf.h"
|
|
#include "we_colbufcompressed.h"
|
|
#include "we_columninfo.h"
|
|
#include "we_bulkstatus.h"
|
|
#include "we_log.h"
|
|
#include "blocksize.h"
|
|
#include <sstream>
|
|
#include <boost/date_time/posix_time/posix_time_types.hpp>
|
|
|
|
namespace
|
|
{
|
|
// Minimum time to wait for a condition, so as to periodically wake up and
|
|
// check the global job status, to see if the job needs to terminate.
|
|
const int COND_WAIT_SECONDS = 1;
|
|
} // namespace
|
|
|
|
namespace WriteEngine
|
|
{
|
|
//------------------------------------------------------------------------------
|
|
// ColumnBufferManger constructor
|
|
//------------------------------------------------------------------------------
|
|
ColumnBufferManager::ColumnBufferManager(ColumnInfo* pColInfo, int width, Log* logger, int compressionType)
|
|
: fBufWriteOffset(0)
|
|
, fBufFreeOffset(0)
|
|
, fResizePending(false)
|
|
, fColWidth(width)
|
|
, fMaxRowId(std::numeric_limits<WriteEngine::RID>::max())
|
|
, fColInfo(pColInfo)
|
|
, fLog(logger)
|
|
{
|
|
if (compressionType)
|
|
fCBuf = new ColumnBufferCompressed(pColInfo, logger);
|
|
else
|
|
fCBuf = new ColumnBuffer(pColInfo, logger);
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// ColumnBufferManger destructor
|
|
//------------------------------------------------------------------------------
|
|
ColumnBufferManager::~ColumnBufferManager()
|
|
{
|
|
if (fCBuf)
|
|
delete fCBuf;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// Reserve a section out of the output buffer for the application code to begin
|
|
// using to store data destined for the database.
|
|
// startRowId (in) - starting RID for the rows to be stored in this section
|
|
// nRowsIn (in) - desired number of rows to be stored in this section
|
|
// secRowCnt (out) - number of rows that the returned ColumnBufferSection can
|
|
// hold. This may be smaller than nRowsIn (for dictionary
|
|
// column) if the rows cross an extent boundary.
|
|
// The application code will need to make a second call
|
|
// later, to create a section to handle the remaining rows
|
|
// that this section could not hold.
|
|
// cbs (out) - reserved section pointer
|
|
// lastInputRowInExtent(out) - last input Row number (relative to start of job)
|
|
// that can go into the current extent we are loading.
|
|
//
|
|
// Basic algorithm:
|
|
// 1. Wait to ensure that ColumnBufferSection allocations are made in input
|
|
// row order
|
|
// 2. Wait for any pending output buffer expansion to be completed
|
|
// 3. If buffer does not have enough room to contain nRowsIn of data, then
|
|
// a. If there are any ColumnBufferSections in use, then
|
|
// wait for these ColumnBufferSections to be released
|
|
// b. If there is any data in the output buffer, then
|
|
// flush the data (rounded down to nearest 8192-byte block)
|
|
// c. Expand the output buffer, and
|
|
// copy the remaining data to the front of the newly expanded buffer
|
|
//
|
|
// @bug 3456
|
|
// Note that when determining whether the output buffer has enough room or
|
|
// not, we are comparing to remainingSpace-1 and not remainingSpace. We
|
|
// don't want output buffer to get completely full, because we can't
|
|
// distinguish between a buffer that is full, and one that is empty;
|
|
// as both cases are indicated by fBufFreeOffset == fBufWriteOffset.
|
|
// If we allow output buffer to get exactly full, then we end up
|
|
// losing track of data because we will think the buffer is empty.
|
|
//
|
|
// 6. Allocate new ColumnBufferSection
|
|
// 7. Update fBufFreeOffset, fSectionsInUse, fMaxRowId
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::reserveSection(RID startRowId, uint32_t nRowsIn, uint32_t& secRowCnt,
|
|
ColumnBufferSection** cbs, RID& lastInputRowInExtent)
|
|
{
|
|
#ifdef PROFILE
|
|
Stats::startParseEvent(WE_STATS_WAIT_TO_RESERVE_OUT_BUF);
|
|
#endif
|
|
*cbs = 0;
|
|
boost::posix_time::seconds wait_seconds(COND_WAIT_SECONDS);
|
|
|
|
boost::mutex::scoped_lock lock(fColInfo->colMutex());
|
|
|
|
//..Ensure that ColumnBufferSection allocations are made in input row order
|
|
bool bWaitedForInSequence = false;
|
|
|
|
while (1)
|
|
{
|
|
RID startRowTest = (std::numeric_limits<WriteEngine::RID>::max() == fMaxRowId) ? 0 : fMaxRowId + 1;
|
|
|
|
if (startRowTest == startRowId)
|
|
break;
|
|
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
bWaitedForInSequence = true;
|
|
std::ostringstream oss;
|
|
oss << "OID-" << fColInfo->curCol.dataFile.fid << "; Waiting for in-sequence";
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
|
|
fOutOfSequence.timed_wait(lock, wait_seconds);
|
|
|
|
// See if JobStatus has been set to terminate by another thread
|
|
if (BulkStatus::getJobStatus() == EXIT_FAILURE)
|
|
{
|
|
throw SecondaryShutdownException(
|
|
"ColumnBufferManager::"
|
|
"reserveSection(1) responding to job termination");
|
|
}
|
|
}
|
|
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
if (bWaitedForInSequence)
|
|
{
|
|
std::ostringstream oss;
|
|
oss << "OID-" << fColInfo->curCol.dataFile.fid << "; Resume after waiting for in-sequence";
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
}
|
|
|
|
//..Check/wait for any pending output buffer expansion to be completed
|
|
bool bWaitedForResize = false;
|
|
|
|
while (fResizePending)
|
|
{
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
bWaitedForResize = true;
|
|
std::ostringstream oss;
|
|
oss << "OID-" << fColInfo->curCol.dataFile.fid << "; Waiting for pending resize";
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
|
|
fResizeInProgress.timed_wait(lock, wait_seconds);
|
|
|
|
// See if JobStatus has been set to terminate by another thread
|
|
if (BulkStatus::getJobStatus() == EXIT_FAILURE)
|
|
{
|
|
throw SecondaryShutdownException(
|
|
"ColumnBufferManager::"
|
|
"reserveSection(2) responding to job termination");
|
|
}
|
|
}
|
|
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
if (bWaitedForResize)
|
|
{
|
|
std::ostringstream oss;
|
|
oss << "OID-" << fColInfo->curCol.dataFile.fid << "; Resume after waiting for pending resize";
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
}
|
|
|
|
#ifdef PROFILE
|
|
Stats::stopParseEvent(WE_STATS_WAIT_TO_RESERVE_OUT_BUF);
|
|
#endif
|
|
|
|
// Through the use of the mutex lock and the fResizePending flag, nobody
|
|
// should be changing the buffer size out from under us; so okay to save in
|
|
// local variable till we call resizeColumnBuffer() to expand the buffer.
|
|
int bufferSize = fCBuf->getSize();
|
|
int remainingSpace = 0;
|
|
|
|
if (bufferSize > 0)
|
|
{
|
|
// Calculate remaining space
|
|
remainingSpace = bufferSize - (fBufFreeOffset + bufferSize - fBufWriteOffset) % bufferSize;
|
|
}
|
|
|
|
//..Restrict the new section to the extent boundary if applicable.
|
|
// We assume here that the colMutex() lock will assure the integrity
|
|
// of the values used in evaluating or recalculating spaceRequired.
|
|
int nRows = 0;
|
|
RETURN_ON_ERROR(rowsExtentCheck(nRowsIn, nRows));
|
|
|
|
int spaceRequired = nRows * fColWidth;
|
|
|
|
if (nRows > 0)
|
|
{
|
|
//..If not enough room to add nRows to output buffer, wait for pending
|
|
// sections to be released, so that we can flush and resize the buffer.
|
|
//..@bug 3456: compare to remainingSpace-1 and not remainingSpace.
|
|
// See note in function description that precedes this function.
|
|
if (spaceRequired > (remainingSpace - 1))
|
|
{
|
|
//#ifdef PROFILE
|
|
// Stats::startParseEvent(WE_STATS_WAIT_TO_RESIZE_OUT_BUF);
|
|
//#endif
|
|
fResizePending = true;
|
|
bool bWaitedForSectionsInUse = false;
|
|
|
|
// Wait for all other threads to finish writing pending sections
|
|
// to the output buffer, before we resize the buffer
|
|
while (fSectionsInUse.size() > 0)
|
|
{
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
bWaitedForSectionsInUse = true;
|
|
std::ostringstream oss;
|
|
oss << "OID-" << fColInfo->curCol.dataFile.fid
|
|
<< "; Waiting to resize output buffer; "
|
|
"sections in-use: "
|
|
<< fSectionsInUse.size();
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
|
|
fBufInUse.timed_wait(lock, wait_seconds);
|
|
|
|
// See if JobStatus has been set to quit by another thread
|
|
if (BulkStatus::getJobStatus() == EXIT_FAILURE)
|
|
{
|
|
throw SecondaryShutdownException(
|
|
"ColumnBufferManager::"
|
|
"reserveSection(3) responding to job termination");
|
|
}
|
|
}
|
|
|
|
//#ifdef PROFILE
|
|
// Stats::stopParseEvent(WE_STATS_WAIT_TO_RESIZE_OUT_BUF);
|
|
// Stats::startParseEvent(WE_STATS_RESIZE_OUT_BUF);
|
|
//#endif
|
|
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
if (bWaitedForSectionsInUse)
|
|
{
|
|
std::ostringstream oss;
|
|
oss << "OID-" << fColInfo->curCol.dataFile.fid << "; Resume after waiting to resize output buffer";
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
}
|
|
|
|
// @bug 1977 correct problem; writing extra blocks
|
|
// Flush remaining data blocks to disk "if" buffer contains data
|
|
if (bufferSize > 0)
|
|
{
|
|
if (fBufFreeOffset != fBufWriteOffset)
|
|
RETURN_ON_ERROR(writeToFile((fBufFreeOffset + bufferSize - 1) % bufferSize));
|
|
}
|
|
|
|
resizeColumnBuffer(spaceRequired);
|
|
bufferSize = fCBuf->getSize(); // update bufferSize after resize-
|
|
// ColumnBuffer() expanded the buffer
|
|
fResizePending = false;
|
|
fResizeInProgress.notify_all();
|
|
//#ifdef PROFILE
|
|
// Stats::stopParseEvent(WE_STATS_RESIZE_OUT_BUF);
|
|
//#endif
|
|
} // (spaceRequired > remainingSpace-1)
|
|
|
|
*cbs = new ColumnBufferSection(fCBuf, startRowId, startRowId + nRows - 1, fColWidth, fBufFreeOffset);
|
|
fBufFreeOffset = (fBufFreeOffset + nRows * fColWidth) % bufferSize;
|
|
fSectionsInUse.push_back(*cbs);
|
|
|
|
fMaxRowId = startRowId + nRows - 1;
|
|
fOutOfSequence.notify_all();
|
|
} // (nRows > 0)
|
|
|
|
secRowCnt = nRows;
|
|
|
|
// Get/return last input Row number for the extent this buffer goes in.
|
|
// If we determine this set of rows will cross over to the next extent,
|
|
// then we tell ColumnInfo to bump the last Row to the end of the next
|
|
// extent, in preparation for the next Read buffer going into the next
|
|
// extent. We do this even though we have not yet allocated the next
|
|
// extent from the extent map.
|
|
lastInputRowInExtent = fColInfo->lastInputRowInExtent();
|
|
|
|
if ((startRowId + nRowsIn) > lastInputRowInExtent)
|
|
fColInfo->lastInputRowInExtentInc();
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// Release the data in the specified ColumnBufferSection, meaning the data in
|
|
// that section is ready to be written to the database.
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::releaseSection(ColumnBufferSection* cbs)
|
|
{
|
|
#ifdef PROFILE
|
|
Stats::startParseEvent(WE_STATS_WAIT_TO_RELEASE_OUT_BUF);
|
|
#endif
|
|
boost::mutex::scoped_lock lock(fColInfo->colMutex());
|
|
#ifdef PROFILE
|
|
Stats::stopParseEvent(WE_STATS_WAIT_TO_RELEASE_OUT_BUF);
|
|
#endif
|
|
cbs->setStatus(WRITE_COMPLETE);
|
|
|
|
int lastWriteOffset = fBufWriteOffset;
|
|
|
|
std::list<ColumnBufferSection*>::iterator it = fSectionsInUse.begin();
|
|
|
|
if (it != fSectionsInUse.end())
|
|
{
|
|
ColumnBufferSection* cbs_temp = *it;
|
|
|
|
while (WRITE_COMPLETE == cbs_temp->getStatus())
|
|
{
|
|
lastWriteOffset = cbs_temp->getStartOffset() + cbs_temp->getSectionSize() - 1;
|
|
|
|
delete cbs_temp;
|
|
it = fSectionsInUse.erase(it);
|
|
|
|
if (it == fSectionsInUse.end())
|
|
break;
|
|
|
|
cbs_temp = *it;
|
|
}
|
|
}
|
|
|
|
fBufInUse.notify_all();
|
|
|
|
RETURN_ON_ERROR(writeToFile(lastWriteOffset));
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// Expand the output buffer for the column this ColumBufferManager is managing.
|
|
// "spaceRequired" indicates the number of additional bytes that are needed.
|
|
//------------------------------------------------------------------------------
|
|
void ColumnBufferManager::resizeColumnBuffer(int spaceRequired)
|
|
{
|
|
int bufferSize = fCBuf->getSize();
|
|
int bufferSizeOld = bufferSize;
|
|
int dataRemaining = (bufferSize > 0) ? ((fBufFreeOffset - fBufWriteOffset + bufferSize) % bufferSize) : 0;
|
|
|
|
int resizeAction = 0;
|
|
|
|
if (0 == bufferSize)
|
|
{
|
|
bufferSize = (int)(spaceRequired * 1.2); // Additional 20% to account
|
|
// for changes in number of rows
|
|
// because of varying line-widths
|
|
resizeAction = 1;
|
|
}
|
|
else
|
|
{
|
|
if (spaceRequired > bufferSize)
|
|
{
|
|
bufferSize = spaceRequired * 2;
|
|
resizeAction = 2;
|
|
}
|
|
else
|
|
{
|
|
bufferSize *= 2; // Double the buffer size
|
|
resizeAction = 3;
|
|
}
|
|
}
|
|
|
|
// Round off the bufferSize to size of a disk block
|
|
if (bufferSize % BLOCK_SIZE > 0)
|
|
{
|
|
bufferSize = (((int)(bufferSize / BLOCK_SIZE)) + 1) * BLOCK_SIZE;
|
|
}
|
|
|
|
if (resizeAction > 0)
|
|
{
|
|
if (fLog->isDebug(DEBUG_2))
|
|
{
|
|
RID numRowsInBuffer = dataRemaining / fColWidth;
|
|
RID firstRid = fMaxRowId - numRowsInBuffer + 1;
|
|
|
|
std::ostringstream oss;
|
|
oss << "Resizing out buffer (case" << resizeAction << ") for OID-" << fColInfo->curCol.dataFile.fid
|
|
<< "; oldSize-" << bufferSizeOld << "; freeOff-" << fBufFreeOffset << "; writeOff-"
|
|
<< fBufWriteOffset << "; startRID-" << firstRid << "; rows-" << numRowsInBuffer << "; reqBytes-"
|
|
<< spaceRequired << "; newSize-" << bufferSize;
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
}
|
|
|
|
// @bug 1977 correct problem; writing extra blocks
|
|
// If we have no data in buffer, we still call resizeAndCopy()
|
|
// to expand the buffer; we just pass -1 for the buffer offsets.
|
|
if (fBufFreeOffset == fBufWriteOffset)
|
|
{
|
|
fCBuf->resizeAndCopy(bufferSize, -1, -1);
|
|
}
|
|
else
|
|
{
|
|
int endOffset = (fBufFreeOffset + bufferSize - 1) % bufferSize;
|
|
fCBuf->resizeAndCopy(bufferSize, fBufWriteOffset, endOffset);
|
|
}
|
|
|
|
fBufFreeOffset = dataRemaining;
|
|
fBufWriteOffset = 0;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// Writes data from our output buffer starting at fBufWriteOffset, through the
|
|
// "endOffset" that is specified (rounded down to nearest 8192-byte block).
|
|
// Trailing data (less than 1 block) is left in the output buffer.
|
|
// Keep in mind that the output buffer is a circular buffer, so if the
|
|
// applicable bytes "wrap-around" in the buffer, then this function will
|
|
// perform 2 I/O operations to write the 2 noncontiguous chunks of data.
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::writeToFile(int endOffset)
|
|
{
|
|
int bufferSize = fCBuf->getSize();
|
|
|
|
if (endOffset == fBufWriteOffset)
|
|
return NO_ERROR;
|
|
|
|
unsigned int writeSize = (endOffset - fBufWriteOffset + bufferSize) % bufferSize + 1;
|
|
|
|
// Don't bother writing anything if we don't at least have a BLOCK_SIZE
|
|
// set of bytes to write out; which means we need to be sure to flush
|
|
// the buffer at the end, because we could have leftover bytes that we
|
|
// have not yet written out.
|
|
if (writeSize < BLOCK_SIZE)
|
|
return NO_ERROR;
|
|
|
|
writeSize = writeSize - writeSize % BLOCK_SIZE; // round down to mult of blksiz
|
|
endOffset = (fBufWriteOffset + writeSize - 1) % bufferSize;
|
|
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
std::ostringstream oss;
|
|
oss << "Writing OID-" << fColInfo->curCol.dataFile.fid << "; bufWriteOff-" << fBufWriteOffset
|
|
<< "; bufFreeOff-" << fBufFreeOffset << "; endWrite-" << endOffset << "; bytesToWrite-" << writeSize
|
|
<< "; bufSize-" << bufferSize;
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
|
|
// Account for circular buffer by making 2 calls to write the data,
|
|
// if we are wrapping around at the end of the buffer.
|
|
if (endOffset < fBufWriteOffset)
|
|
{
|
|
RETURN_ON_ERROR(writeToFileExtentCheck(fBufWriteOffset, bufferSize - fBufWriteOffset));
|
|
fBufWriteOffset = 0;
|
|
}
|
|
|
|
RETURN_ON_ERROR(writeToFileExtentCheck(fBufWriteOffset, endOffset - fBufWriteOffset + 1));
|
|
fBufWriteOffset = (endOffset + 1) % bufferSize;
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// Writes the specified bytes from the internal buffer to the db column file.
|
|
// The data to be written, starts at "startOffset" in the internal buffer and
|
|
// is "writeSize" bytes long.
|
|
// This function also checks to see if an extent needs to be added to the db
|
|
// column file for this number of bytes. If a second extent is required,
|
|
// then the current db file will be filled out with the 1st part of the buffer,
|
|
// and the remaining buffer data will be written to the next segment file in
|
|
// the DBRoot, partition, segement number sequence.
|
|
// This function also catches and handles the case where an abbreviated
|
|
// extent needs to be expanded to a full extent on disk. When fillUpWEmpties is
|
|
// set then CS finishes with writing and has to fill with magics this block
|
|
// up to its boundary.
|
|
//
|
|
// WARNING: This means this function may change the information in the
|
|
// ColumnInfo struct that owns this ColumnBufferManager, if a
|
|
// second db column file has to be opened to finish writing the
|
|
// internal buffer, or if an abbreviated extent is expanded.
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::writeToFileExtentCheck(uint32_t startOffset, uint32_t writeSize, bool fillUpWEmpties)
|
|
{
|
|
if (fLog->isDebug(DEBUG_3))
|
|
{
|
|
std::ostringstream oss;
|
|
oss << "Col extent check: OID-" << fColInfo->curCol.dataFile.fid << "; DBRoot-"
|
|
<< fColInfo->curCol.dataFile.fDbRoot << "; part-" << fColInfo->curCol.dataFile.fPartition << "; seg-"
|
|
<< fColInfo->curCol.dataFile.fSegment << "; Wanting to write " << writeSize
|
|
<< " bytes, with avail space " << fColInfo->availFileSize;
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
|
|
// Don't need a mutex lock here because if writeToFile() is calling
|
|
// us, we already have a lock; and if flush() is calling us, then
|
|
// all parsing is complete, so we should have no thread contention.
|
|
|
|
// If extent out of space, see if this is an abbrev extent we can expand
|
|
long long availableFileSize = fColInfo->availFileSize;
|
|
|
|
if ((availableFileSize < writeSize) && (fColInfo->isAbbrevExtent()))
|
|
{
|
|
int rc = fColInfo->expandAbbrevExtent(true);
|
|
|
|
if (rc != NO_ERROR)
|
|
{
|
|
WErrorCodes ec;
|
|
std::ostringstream oss;
|
|
oss << "writeToFileExtentCheck: expand extent failed: " << ec.errorString(rc);
|
|
fLog->logMsg(oss.str(), rc, MSGLVL_ERROR);
|
|
return rc;
|
|
}
|
|
|
|
availableFileSize = fColInfo->availFileSize;
|
|
}
|
|
|
|
if (availableFileSize >= writeSize)
|
|
{
|
|
int rc = fCBuf->writeToFile(startOffset, writeSize, fillUpWEmpties);
|
|
|
|
if (rc != NO_ERROR)
|
|
{
|
|
WErrorCodes ec;
|
|
std::ostringstream oss;
|
|
oss << "writeToFileExtentCheck: write1 extent failed: " << ec.errorString(rc);
|
|
fLog->logMsg(oss.str(), rc, MSGLVL_ERROR);
|
|
return rc;
|
|
}
|
|
|
|
// MCOL-498 Fill this block up to its boundary.
|
|
if (fillUpWEmpties)
|
|
{
|
|
writeSize = BLOCK_SIZE;
|
|
}
|
|
|
|
fColInfo->updateBytesWrittenCounts(writeSize);
|
|
}
|
|
else
|
|
{
|
|
// We use ColumnInfo to help us add an extent to the "next"
|
|
// segment file, if needed.
|
|
// Current extent does not have enough room for buffer, so we
|
|
// have to break up the buffer into 2 extents; creating a new
|
|
// extent and switching the db column file "on-the-fly".
|
|
int writeSize1 = availableFileSize;
|
|
|
|
if (writeSize1 > 0)
|
|
{
|
|
int rc = fCBuf->writeToFile(startOffset, writeSize1);
|
|
|
|
if (rc != NO_ERROR)
|
|
{
|
|
WErrorCodes ec;
|
|
std::ostringstream oss;
|
|
oss << "writeToFileExtentCheck: write2 extent failed: " << ec.errorString(rc);
|
|
fLog->logMsg(oss.str(), rc, MSGLVL_ERROR);
|
|
return rc;
|
|
}
|
|
|
|
fColInfo->updateBytesWrittenCounts(writeSize1);
|
|
}
|
|
|
|
int rc = fColInfo->extendColumn(true);
|
|
|
|
if (rc != NO_ERROR)
|
|
{
|
|
WErrorCodes ec;
|
|
std::ostringstream oss;
|
|
oss << "writeToFileExtentCheck: extend column failed: " << ec.errorString(rc);
|
|
fLog->logMsg(oss.str(), rc, MSGLVL_ERROR);
|
|
return rc;
|
|
}
|
|
|
|
int writeSize2 = writeSize - writeSize1;
|
|
rc = fCBuf->writeToFile(startOffset + writeSize1, writeSize2, fillUpWEmpties);
|
|
|
|
if (rc != NO_ERROR)
|
|
{
|
|
WErrorCodes ec;
|
|
std::ostringstream oss;
|
|
oss << "writeToFileExtentCheck: write3 extent failed: " << ec.errorString(rc);
|
|
fLog->logMsg(oss.str(), rc, MSGLVL_ERROR);
|
|
return rc;
|
|
}
|
|
|
|
// MCOL-498 Fill this block up to its boundary.
|
|
if (fillUpWEmpties)
|
|
{
|
|
writeSize2 = BLOCK_SIZE;
|
|
}
|
|
|
|
fColInfo->updateBytesWrittenCounts(writeSize2);
|
|
}
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// Flush the contents of internal fCBuf (column buffer) to disk. If CS flushes
|
|
// less then BLOCK_SIZE bytes then it propagates this event down the stack.
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::flush()
|
|
{
|
|
if (fBufFreeOffset == fBufWriteOffset)
|
|
{
|
|
if (fLog->isDebug(DEBUG_2))
|
|
{
|
|
std::ostringstream oss;
|
|
oss << "Skipping write flush for: OID-" << fColInfo->curCol.dataFile.fid << "; DBRoot-"
|
|
<< fColInfo->curCol.dataFile.fDbRoot << "; part-" << fColInfo->curCol.dataFile.fPartition
|
|
<< "; seg-" << fColInfo->curCol.dataFile.fSegment
|
|
<< "; both fBufFreeOffset and fBufWriteOffset = " << fBufFreeOffset;
|
|
fLog->logMsg(oss.str(), MSGLVL_INFO2);
|
|
}
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
int bufferSize = fCBuf->getSize();
|
|
|
|
// MCOL-498 There are less the BLOCK_SIZE bytes in the buffer left
|
|
// so propagate this info down the stack to fill the buffer up
|
|
// with empty magics.
|
|
// Account for circular buffer by making 2 calls to write the data,
|
|
// if we are wrapping around at the end of the buffer.
|
|
if (fBufFreeOffset < fBufWriteOffset)
|
|
{
|
|
bool fillUpWEmpties =
|
|
(static_cast<unsigned int>(bufferSize - fBufWriteOffset) >= BLOCK_SIZE) ? false : true;
|
|
RETURN_ON_ERROR(writeToFileExtentCheck(fBufWriteOffset, bufferSize - fBufWriteOffset, fillUpWEmpties));
|
|
fBufWriteOffset = 0;
|
|
}
|
|
// MCOL-498 fill the buffer up with empty magics.
|
|
RETURN_ON_ERROR(writeToFileExtentCheck(fBufWriteOffset, fBufFreeOffset - fBufWriteOffset, true));
|
|
fBufWriteOffset = fBufFreeOffset;
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// In the middle of parsing a Read buffer, if we detect that the buffer will
|
|
// need to be split among 2 different extents, then this function should be
|
|
// called to flush the first part of the buffer into the first extent, before
|
|
// writing the second part of the buffer into the second extent. This function
|
|
// will wait for the pending sections to be parsed so that they can all be
|
|
// flushed to the segment file containing the first extent. This function is
|
|
// currently only needed for Dictionary column processing.
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::intermediateFlush()
|
|
{
|
|
boost::posix_time::seconds wait_seconds(COND_WAIT_SECONDS);
|
|
boost::mutex::scoped_lock lock(fColInfo->colMutex());
|
|
|
|
// Wait for all other threads which are currently parsing rows,
|
|
// to finish parsing the data in those sections.
|
|
#ifdef PROFILE
|
|
Stats::startParseEvent(WE_STATS_WAIT_FOR_INTERMEDIATE_FLUSH);
|
|
#endif
|
|
|
|
while (fSectionsInUse.size() > 0)
|
|
{
|
|
fBufInUse.timed_wait(lock, wait_seconds);
|
|
|
|
// See if JobStatus has been set to terminate by another thread
|
|
if (BulkStatus::getJobStatus() == EXIT_FAILURE)
|
|
{
|
|
throw SecondaryShutdownException(
|
|
"ColumnBufferManager::"
|
|
"intermediateFlush() responding to job termination");
|
|
}
|
|
}
|
|
|
|
#ifdef PROFILE
|
|
Stats::stopParseEvent(WE_STATS_WAIT_FOR_INTERMEDIATE_FLUSH);
|
|
#endif
|
|
|
|
RETURN_ON_ERROR(flush());
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// This function is a no-op for non-dictionary columns, but it is provided as
|
|
// a hook for ColumnBufferManagerDctnry to provide a function that adds extents
|
|
// as sections from the output buffer are being copied to the column segment
|
|
// file(s).
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::rowsExtentCheck(int nRows, int& nRows2)
|
|
{
|
|
nRows2 = nRows;
|
|
|
|
return NO_ERROR;
|
|
}
|
|
|
|
//------------------------------------------------------------------------------
|
|
// This wrapper function wraps a mutex lock around the call to extendColumn().
|
|
// For typical numeric column processing, extendColumn() is called from within
|
|
// ColumnBufferManager::writeToFileExtentCheck() which already has a mutex lock.
|
|
// But dictionary token processing detects/extends a token column outside of
|
|
// ColumnBufferManager; hence the need for this function to be called to employ
|
|
// a mutex lock around the call to extendColumn().
|
|
//------------------------------------------------------------------------------
|
|
int ColumnBufferManager::extendTokenColumn()
|
|
{
|
|
boost::mutex::scoped_lock lock(fColInfo->colMutex());
|
|
|
|
return fColInfo->extendColumn(false);
|
|
}
|
|
|
|
} // namespace WriteEngine
|