1
0
mirror of https://github.com/mariadb-corporation/mariadb-columnstore-engine.git synced 2025-07-24 22:42:05 +03:00

Improve batch inserts.

1) Instead of making dbrm calls to writeVBEntry() per block,
     we make these calls per batch. This can have non-trivial
     reductions in the overhead of these calls if the batch size
     is large.

  2) In dmlproc, do not deserialize the whole insertpackage, which
     consists of the complete record set per column, which would be
     wasteful as we only need some metadata fields from insertpackage
     here. This is only done for batch inserts at the moment, this
     should also be applied to single inserts.
This commit is contained in:
Gagan Goel
2020-05-10 19:38:06 -04:00
parent 04fdacb927
commit d4d0ebdf5d
13 changed files with 306 additions and 36 deletions

View File

@ -72,6 +72,28 @@ int DMLTable::read(messageqcpp::ByteStream& bytestream)
return retval;
}
void DMLTable::readMetaData(messageqcpp::ByteStream& bytestream)
{
// read the table name
bytestream >> fName;
// read the schema name
bytestream >> fSchema;
}
void DMLTable::readRowData(messageqcpp::ByteStream& bytestream)
{
messageqcpp::ByteStream::quadbyte rowNum;
bytestream >> rowNum;
for (unsigned int i = 0; i < rowNum; i++)
{
Row* aRow = new Row();
aRow->read(bytestream);
fRows.push_back(aRow);
}
}
int DMLTable::write(messageqcpp::ByteStream& bytestream)
{
int retval = 1;

View File

@ -91,6 +91,20 @@ public:
int read(messageqcpp::ByteStream& bytestream);
/** @brief read a DMLTable metadata from a ByteStream
*
* @param bytestream the ByteStream to read from
*/
void readMetaData(messageqcpp::ByteStream& bytestream);
/** @brief read a DMLTable row data from a ByteStream
*
* @param bytestream the ByteStream to read from
*/
void readRowData(messageqcpp::ByteStream& bytestream);
/** @brief write a DMLTable to a ByteStream
*
* @param bytestream the ByteStream to write to

View File

@ -66,16 +66,16 @@ int InsertDMLPackage::write(messageqcpp::ByteStream& bytestream)
bytestream << (uint8_t)fLogging;
bytestream << (uint8_t)fLogending;
if (fTable != 0)
{
retval = fTable->write(bytestream);
}
bytestream << fTableOid;
bytestream << static_cast<const messageqcpp::ByteStream::byte>(fIsInsertSelect);
bytestream << static_cast<const messageqcpp::ByteStream::byte>(fIsBatchInsert);
bytestream << static_cast<const messageqcpp::ByteStream::byte>(fIsAutocommitOn);
if (fTable != 0)
{
retval = fTable->write(bytestream);
}
return retval;
}
@ -100,15 +100,50 @@ int InsertDMLPackage::read(messageqcpp::ByteStream& bytestream)
bytestream >> logending;
fLogending = (logending != 0);
bytestream >> fTableOid;
bytestream >> reinterpret_cast<messageqcpp::ByteStream::byte&>(fIsInsertSelect);
bytestream >> reinterpret_cast<messageqcpp::ByteStream::byte&>(fIsBatchInsert);
bytestream >> reinterpret_cast<messageqcpp::ByteStream::byte&>(fIsAutocommitOn);
fTable = new DMLTable();
retval = fTable->read(bytestream);
bytestream >> fTableOid;
bytestream >> reinterpret_cast< messageqcpp::ByteStream::byte&>(fIsInsertSelect);
bytestream >> reinterpret_cast< messageqcpp::ByteStream::byte&>(fIsBatchInsert);
bytestream >> reinterpret_cast< messageqcpp::ByteStream::byte&>(fIsAutocommitOn);
return retval;
}
void InsertDMLPackage::readMetaData(messageqcpp::ByteStream& bytestream)
{
messageqcpp::ByteStream::quadbyte session_id;
bytestream >> session_id;
fSessionID = session_id;
bytestream >> fUuid;
std::string dmlStatement;
bytestream >> fDMLStatement;
bytestream >> fSQLStatement;
bytestream >> fSchemaName;
bytestream >> fTimeZone;
uint8_t logging;
bytestream >> logging;
fLogging = (logging != 0);
uint8_t logending;
bytestream >> logending;
fLogending = (logending != 0);
bytestream >> fTableOid;
bytestream >> reinterpret_cast<messageqcpp::ByteStream::byte&>(fIsInsertSelect);
bytestream >> reinterpret_cast<messageqcpp::ByteStream::byte&>(fIsBatchInsert);
bytestream >> reinterpret_cast<messageqcpp::ByteStream::byte&>(fIsAutocommitOn);
fTable = new DMLTable();
fTable->readMetaData(bytestream);
}
// Has to be called after InsertDMLPackage::readMetaData()
void InsertDMLPackage::readRowData(messageqcpp::ByteStream& bytestream)
{
fTable->readRowData(bytestream);
}
int InsertDMLPackage::buildFromBuffer(std::string& buffer, int columns, int rows)
{
#ifdef DML_PACKAGE_DEBUG

View File

@ -73,6 +73,18 @@ public:
*/
EXPORT int read(messageqcpp::ByteStream& bytestream);
/** @brief read InsertDMLPackage metadata from bytestream
*
* @param bytestream the ByteStream to read from
*/
EXPORT void readMetaData(messageqcpp::ByteStream& bytestream);
/** @brief read InsertDMLPackage row data from bytestream
*
* @param bytestream the ByteStream to read from
*/
EXPORT void readRowData(messageqcpp::ByteStream& bytestream);
/** @brief build a InsertDMLPackage from a string buffer
*
* @param buffer

View File

@ -548,7 +548,7 @@ void PackageHandler::run()
dmlpackage::InsertDMLPackage insertPkg;
//boost::shared_ptr<messageqcpp::ByteStream> insertBs (new messageqcpp::ByteStream);
messageqcpp::ByteStream bsSave = *(fByteStream.get());
insertPkg.read(*(fByteStream.get()));
insertPkg.readMetaData(*(fByteStream.get()));
#ifdef MCOL_140
if (fConcurrentSupport)
@ -584,8 +584,8 @@ void PackageHandler::run()
//cout << "This is batch insert " << insertPkg->get_isBatchInsert() << endl;
if (insertPkg.get_isBatchInsert())
{
fByteStream->reset();
//cout << "This is batch insert " << endl;
//boost::shared_ptr<messageqcpp::ByteStream> insertBs (new messageqcpp::ByteStream(fByteStream));
BatchInsertProc* batchProcessor = NULL;
{
boost::mutex::scoped_lock lk(DMLProcessor::batchinsertProcessorMapLock);
@ -900,7 +900,11 @@ void PackageHandler::run()
}
else // Single Insert
{
//insertPkg.readTable(*(fByteStream.get()));
// make sure insertPkg.readMetaData() is called before
// this on fByteStream!
// TODO: Similar to batch inserts, don't
// deserialize the row data here for single inserts.
insertPkg.readRowData(*(fByteStream.get()));
insertPkg.set_TxnID(fTxnid);
fProcessor.reset(new dmlpackageprocessor::InsertPackageProcessor(fDbrm, insertPkg.get_SessionID()));
result = fProcessor->processPackage(insertPkg);

View File

@ -507,6 +507,7 @@ const uint8_t RELEASE_LBID_RANGES = 91;
/* More main BRM functions 100-110 */
const uint8_t BULK_UPDATE_DBROOT = 100;
const uint8_t GET_SYSTEM_CATALOG = 101;
const uint8_t BULK_WRITE_VB_ENTRY = 102;
/* Error codes returned by the DBRM functions. */

View File

@ -2226,6 +2226,42 @@ int DBRM::writeVBEntry(VER_t transID, LBID_t lbid, OID_t vbOID,
return err;
}
int DBRM::bulkWriteVBEntry(VER_t transID,
const std::vector<BRM::LBID_t>& lbids,
OID_t vbOID,
const std::vector<uint32_t>& vbFBOs) DBRM_THROW
{
#ifdef BRM_INFO
if (fDebug)
{
TRACER_WRITELATER("bulkWriteVBEntry");
TRACER_WRITE;
}
#endif
ByteStream command, response;
uint8_t err;
command << BULK_WRITE_VB_ENTRY << (uint32_t) transID;
serializeInlineVector(command, lbids);
command << (uint32_t) vbOID;
serializeInlineVector(command, vbFBOs);
err = send_recv(command, response);
if (err != ERR_OK)
return err;
if (response.length() != 1)
return ERR_NETWORK;
response >> err;
CHECK_EMPTY(response);
return err;
}
struct _entry
{
_entry(LBID_t l) : lbid(l) { };

View File

@ -608,6 +608,20 @@ public:
EXPORT int writeVBEntry(VER_t transID, LBID_t lbid, OID_t vbOID,
uint32_t vbFBO) DBRM_THROW;
/** @brief Bulk registers a version buffer entry.
*
* Similar to writeVBEntry, but registers the version buffer
* entries in bulk for a list of lbids and vbFBOs, for a given
* transID and vbOID.
* @note The version buffer locations must hold the 'copy' lock
* first.
* @return 0 on success, non-0 on error (see brmtypes.h)
*/
EXPORT int bulkWriteVBEntry(VER_t transID,
const std::vector<BRM::LBID_t>& lbids,
OID_t vbOID,
const std::vector<uint32_t>& vbFBOs) DBRM_THROW;
/** @brief Retrieves a list of uncommitted LBIDs.
*
* Retrieves a list of uncommitted LBIDs for the given transaction ID.

View File

@ -375,6 +375,10 @@ void SlaveComm::processCommand(ByteStream& msg)
do_writeVBEntry(msg);
break;
case BULK_WRITE_VB_ENTRY:
do_bulkWriteVBEntry(msg);
break;
case BEGIN_VB_COPY:
do_beginVBCopy(msg);
break;
@ -1758,6 +1762,49 @@ void SlaveComm::do_writeVBEntry(ByteStream& msg)
doSaveDelta = true;
}
void SlaveComm::do_bulkWriteVBEntry(ByteStream& msg)
{
VER_t transID;
std::vector<BRM::LBID_t> lbids;
OID_t vbOID;
std::vector<uint32_t> vbFBOs;
uint32_t tmp;
int err;
ByteStream reply;
#ifdef BRM_VERBOSE
cerr << "WorkerComm: do_bulkWriteVBEntry()" << endl;
#endif
msg >> tmp;
transID = tmp;
deserializeInlineVector(msg, lbids);
msg >> tmp;
vbOID = tmp;
deserializeInlineVector(msg, vbFBOs);
if (printOnly)
{
cout << "bulkWriteVBEntry: transID=" << transID << endl;
for (size_t i = 0; i < lbids.size(); i++)
cout << "bulkWriteVBEntry arg " << i + 1 << ": lbid=" << lbids[i] << " vbOID=" <<
vbOID << " vbFBO=" << vbFBOs[i] << endl;
return;
}
err = slave->bulkWriteVBEntry(transID, lbids, vbOID, vbFBOs);
reply << (uint8_t) err;
#ifdef BRM_VERBOSE
cerr << "WorkerComm: do_bulkWriteVBEntry() err code is " << err << endl;
#endif
if (!standalone)
master.write(reply);
doSaveDelta = true;
}
void SlaveComm::do_beginVBCopy(ByteStream& msg)
{
VER_t transID;

View File

@ -91,6 +91,7 @@ private:
void do_bulkSetHWM(messageqcpp::ByteStream& msg);
void do_bulkSetHWMAndCP(messageqcpp::ByteStream& msg);
void do_writeVBEntry(messageqcpp::ByteStream& msg);
void do_bulkWriteVBEntry(messageqcpp::ByteStream& msg);
void do_beginVBCopy(messageqcpp::ByteStream& msg);
void do_endVBCopy(messageqcpp::ByteStream& msg);
void do_vbRollback1(messageqcpp::ByteStream& msg);

View File

@ -523,6 +523,70 @@ int SlaveDBRMNode::writeVBEntry(VER_t transID, LBID_t lbid, OID_t vbOID,
return 0;
}
int SlaveDBRMNode::bulkWriteVBEntry(VER_t transID,
const std::vector<BRM::LBID_t>& lbids,
OID_t vbOID,
const std::vector<uint32_t>& vbFBOs) throw()
{
VER_t oldVerID;
/*
LBIDRange r;
r.start = lbid;
r.size = 1;
if (!copylocks.isLocked(r))
cout << "Copylock error: lbid " << lbid << " isn't locked\n";
*/
try
{
vbbm.lock(VBBM::WRITE);
locked[0] = true;
vss.lock(VSS::WRITE);
locked[1] = true;
for (size_t i = 0; i < lbids.size(); i++)
{
// figure out the current version of the block
// NOTE! This will currently error out to preserve the assumption that
// larger version numbers imply more recent changes. If we ever change that
// assumption, we'll need to revise the vbRollback() fcns as well.
oldVerID = vss.getCurrentVersion(lbids[i], NULL);
if (oldVerID == transID)
continue;
else if (oldVerID > transID)
{
ostringstream str;
str << "WorkerDBRMNode::bulkWriteVBEntry(): Overlapping transactions detected. "
"Transaction " << transID << " cannot overwrite blocks written by "
"transaction " << oldVerID;
log(str.str());
return ERR_OLDTXN_OVERWRITING_NEWTXN;
}
vbbm.insert(lbids[i], oldVerID, vbOID, vbFBOs[i]);
if (oldVerID > 0)
vss.setVBFlag(lbids[i], oldVerID, true);
else
vss.insert(lbids[i], oldVerID, true, false);
// XXXPAT: There's a problem if we use transID as the new version here.
// Need to use at least oldVerID + 1. OldverID can be > TransID
vss.insert(lbids[i], transID, false, true);
}
}
catch (exception& e)
{
cerr << e.what() << endl;
return -1;
}
return 0;
}
int SlaveDBRMNode::beginVBCopy(VER_t transID, uint16_t vbOID,
const LBIDRange_v& ranges, VBRange_v& freeList, bool flushPMCache) throw()
{

View File

@ -364,6 +364,20 @@ public:
EXPORT int writeVBEntry(VER_t transID, LBID_t lbid, OID_t vbOID,
uint32_t vbFBO) throw();
/** @brief Bulk registers a version buffer entry.
*
* Similar to writeVBEntry, but registers the version buffer
* entries in bulk for a list of lbids and vbFBOs, for a given
* transID and vbOID.
* @note The version buffer locations must hold the 'copy' lock
* first.
* @return 0 on success, -1 on error
*/
EXPORT int bulkWriteVBEntry(VER_t transID,
const std::vector<BRM::LBID_t>& lbids,
OID_t vbOID,
const std::vector<uint32_t>& vbFBOs) throw();
/** @brief Atomically prepare to copy data to the version buffer
*
* Atomically sets the copy flag on the specified LBID ranges

View File

@ -1737,38 +1737,44 @@ int BRMWrapper::writeVB(IDBDataFile* pSourceFile, const VER_t transID, const OID
if (rc != NO_ERROR)
goto cleanup;
for (; processedBlocks < (k + rangeListCount); processedBlocks++)
std::vector<BRM::LBID_t> lbids(k);
std::vector<uint32_t> vbFBOs(k);
size_t idx = 0;
for (; processedBlocks < (k + rangeListCount); processedBlocks++, idx++)
{
rc = blockRsltnMgrPtr->writeVBEntry(transID, rangeList[processedBlocks].start,
freeList[i].vbOID, freeList[i].vbFBO + (processedBlocks - rangeListCount));
lbids[idx] = rangeList[processedBlocks].start;
vbFBOs[idx] = freeList[i].vbFBO + (processedBlocks - rangeListCount);
}
//cout << (uint64_t)rangeList[processedBlocks].start << endl;
if (rc != NO_ERROR)
rc = blockRsltnMgrPtr->bulkWriteVBEntry(transID, lbids, freeList[i].vbOID,
vbFBOs);
if (rc != NO_ERROR)
{
switch (rc)
{
switch (rc)
{
case ERR_DEADLOCK:
rc = ERR_BRM_DEAD_LOCK;
break;
case ERR_DEADLOCK:
rc = ERR_BRM_DEAD_LOCK;
break;
case ERR_VBBM_OVERFLOW:
rc = ERR_BRM_VB_OVERFLOW;
break;
case ERR_VBBM_OVERFLOW:
rc = ERR_BRM_VB_OVERFLOW;
break;
case ERR_NETWORK:
rc = ERR_BRM_NETWORK;
break;
case ERR_NETWORK:
rc = ERR_BRM_NETWORK;
break;
case ERR_READONLY:
rc = ERR_BRM_READONLY;
break;
case ERR_READONLY:
rc = ERR_BRM_READONLY;
break;
default:
rc = ERR_BRM_WR_VB_ENTRY;
}
goto cleanup;
default:
rc = ERR_BRM_WR_VB_ENTRY;
}
goto cleanup;
}
}
}