virtual void init()
{
HashDistributeMasterBase::init();
// JCSMORE should common up some with indexread
IHThorKeyedDistributeArg *helper = (IHThorKeyedDistributeArg *)queryHelper();
StringBuffer scoped;
OwnedRoxieString indexFileName(helper->getIndexFileName());
queryThorFileManager().addScope(container.queryJob(), indexFileName, scoped);
file.setown(queryThorFileManager().lookup(container.queryJob(), indexFileName));
if (!file)
throw MakeActivityException(this, 0, "KeyedDistribute: Failed to find key: %s", scoped.str());
if (0 == file->numParts())
throw MakeActivityException(this, 0, "KeyedDistribute: Can't distribute based on an empty key: %s", scoped.str());
checkFormatCrc(this, file, helper->getFormatCrc(), true);
Owned<IFileDescriptor> fileDesc = file->getFileDescriptor();
Owned<IPartDescriptor> tlkDesc = fileDesc->getPart(fileDesc->numParts()-1);
if (!tlkDesc->queryProperties().hasProp("@kind") || 0 != stricmp("topLevelKey", tlkDesc->queryProperties().queryProp("@kind")))
throw MakeActivityException(this, 0, "Cannot distribute using a non-distributed key: '%s'", scoped.str());
unsigned location;
OwnedIFile iFile;
StringBuffer filePath;
if (!getBestFilePart(this, *tlkDesc, iFile, location, filePath))
throw MakeThorException(TE_FileNotFound, "Top level key part does not exist, for key: %s", file->queryLogicalName());
OwnedIFileIO iFileIO = iFile->open(IFOread);
assertex(iFileIO);
tlkMb.append(iFileIO->size());
::read(iFileIO, 0, (size32_t)iFileIO->size(), tlkMb);
queryThorFileManager().noteFileRead(container.queryJob(), file);
}
void sendPartialCount(CSlaveActivity &activity, rowcount_t partialCount)
{
CMessageBuffer msg;
msg.append(partialCount);
if (!activity.queryContainer().queryJob().queryJobComm().send(msg, 0, activity.queryMpTag(), 5000))
throw MakeThorException(0, "Failed to give partial result to master");
}
virtual void validateFile(IDistributedFile *file)
{
IHThorDiskReadBaseArg *helper = (IHThorDiskReadBaseArg *)queryHelper();
bool codeGenGrouped = 0 != (TDXgrouped & helper->getFlags());
bool isGrouped = fileDesc->isGrouped();
if (isGrouped != codeGenGrouped)
{
Owned<IException> e = MakeActivityWarning(&container, TE_GroupMismatch, "DFS and code generated group info. differs: DFS(%s), CodeGen(%s), using DFS info", isGrouped?"grouped":"ungrouped", codeGenGrouped?"grouped":"ungrouped");
queryJobChannel().fireException(e);
}
IOutputMetaData *recordSize = helper->queryDiskRecordSize()->querySerializedDiskMeta();
if (recordSize->isFixedSize()) // fixed size
{
if (0 != fileDesc->queryProperties().getPropInt("@recordSize"))
{
size32_t rSz = fileDesc->queryProperties().getPropInt("@recordSize");
if (isGrouped)
rSz--; // eog byte not to be included in this test.
if (rSz != recordSize->getMinRecordSize())
throw MakeThorException(TE_RecordSizeMismatch, "Published record size %d for file %s, does not match coded record size %d", rSz, fileName.get(), recordSize->getMinRecordSize());
}
if (!fileDesc->isCompressed() && (TDXcompress & helper->getFlags()))
{
size32_t rSz = recordSize->getMinRecordSize();
if (isGrouped) rSz++;
if (rSz >= MIN_ROWCOMPRESS_RECSIZE)
{
Owned<IException> e = MakeActivityWarning(&container, TE_CompressionMismatch, "Ignoring compression attribute on file '%s', which is not published as compressed in DFS", fileName.get());
queryJobChannel().fireException(e);
}
}
}
if (0 == (TDRnocrccheck & helper->getFlags()))
checkFormatCrc(this, file, helper->getFormatCrc(), false);
}
void CPartialResultAggregator::sendResult(const void *row)
{
CMessageBuffer mb;
if (row)
{
CMemoryRowSerializer mbs(mb);
activity.queryRowSerializer()->serialize(mbs,(const byte *)row);
}
if (!activity.queryContainer().queryJob().queryJobComm().send(mb, 0, activity.queryMpTag(), 5000))
throw MakeThorException(0, "Failed to give partial result to master");
}
virtual unsigned hash(const void *data)
{
if (1 == count)
return offsetTable[0].index;
offset_t fpos = iFetchHandler.extractFpos(data);
if (isLocalFpos(fpos))
return getLocalFposPart(fpos);
const void *result = bsearch(&fpos, offsetTable, count, sizeof(FPosTableEntry), slaveLookup);
if (!result)
throw MakeThorException(TE_FetchOutOfRange, "FETCH: Offset not found in offset table; fpos=%"I64F"d", fpos);
return ((FPosTableEntry *)result)->index;
}
virtual void process()
{
if (totalCountKnown) return;
if (container.queryLocalOrGrouped())
return;
totalCount = ::getCount(*this, container.queryJob().querySlaves(), stopAfter, mpTag);
if (totalCount > stopAfter)
totalCount = stopAfter;
CMessageBuffer msg;
msg.append(totalCount);
if (!queryJobChannel().queryJobComm().send(msg, 1, mpTag, 5000))
throw MakeThorException(0, "Failed to give result to slave");
}
virtual void process()
{
CMasterActivity::process();
bool results = false;
unsigned nslaves = container.queryJob().querySlaves();
while (nslaves--)
{
CMessageBuffer mb;
if (abortSoon || !receiveMsg(mb, RANK_ALL, replyTag, NULL)) break;
StringBuffer str;
mb.getSender().getUrlStr(str);
size32_t sz;
mb.read(sz);
if (sz)
{
if (results)
throw MakeThorException(TE_UnexpectedMultipleSlaveResults, "Received greater than one result from slaves");
IHThorRemoteResultArg *helper = (IHThorRemoteResultArg *)queryHelper();
Owned<IThorRowInterfaces> resultRowIf = createRowInterfaces(helper->queryOutputMeta());
CThorStreamDeserializerSource mds(sz, mb.readDirect(sz));
RtlDynamicRowBuilder rowBuilder(resultRowIf->queryRowAllocator());
size32_t sz = resultRowIf->queryRowDeserializer()->deserialize(rowBuilder, mds);
OwnedConstThorRow result = rowBuilder.finalizeRowClear(sz);
helper->sendResult(result);
results = true;
}
}
if (!results && !abortSoon)
{
ActPrintLog("WARNING: no results");
IHThorRemoteResultArg *helper = (IHThorRemoteResultArg *)queryHelper();
//helper->sendResult(NULL);
// Jake I think this always cores (so raise exception instead)
throw MakeThorException(TE_UnexpectedMultipleSlaveResults, "Received no results from slaves");
}
}
virtual void process()
{
IRecordSize *recordSize = helper->queryOutputMeta();
Owned<IThorRowInterfaces> rowIf = createThorRowInterfaces(queryRowManager(), helper->queryOutputMeta(), queryId(), queryCodeContext());
OwnedConstThorRow result = getAggregate(*this, container.queryJob().querySlaves(), *rowIf, *helper, mpTag);
if (!result)
return;
CMessageBuffer msg;
CMemoryRowSerializer mbs(msg);
rowIf->queryRowSerializer()->serialize(mbs, (const byte *)result.get());
if (!queryJobChannel().queryJobComm().send(msg, 1, mpTag, 5000))
throw MakeThorException(0, "Failed to give result to slave");
}
IRowStream * doGlobalSelfJoin()
{
#if THOR_TRACE_LEVEL > 5
ActPrintLog("SELFJOIN: Performing global self-join");
#endif
sorter->Gather(::queryRowInterfaces(input), input, compare, NULL, NULL, keyserializer, NULL, false, isUnstable(), abortSoon, NULL);
stopInput(input);
input = NULL;
if(abortSoon)
{
barrier->cancel();
return NULL;
}
if (!barrier->wait(false)) {
Sleep(1000); // let original error through
throw MakeThorException(TE_BarrierAborted,"SELFJOIN: Barrier Aborted");
}
rowcount_t totalrows;
return sorter->startMerge(totalrows);
}
void process()
{
CWorkUnitWriteMasterBase::process();
unsigned nslaves = container.queryJob().querySlaves();
CMessageBuffer mb;
unsigned s=0;
for (; s<nslaves; s++)
{
loop
{
if (!container.queryJob().queryJobComm().send(mb, s+1, mpTag)) return;
if (!receiveMsg(mb, s+1, mpTag)) return;
if (0 == mb.length())
break;
unsigned numGot;
mb.read(numGot);
unsigned l=mb.remaining();
if (workunitWriteLimit && totalSize+resultData.length()+l > workunitWriteLimit)
{
StringBuffer errMsg("Dataset too large to output to workunit (limit is set to ");
errMsg.append(workunitWriteLimit/0x100000).append(") megabytes, in result (");
if (resultName.length())
errMsg.append("name=").append(resultName);
else
errMsg.append("sequence=").append(resultSeq);
errMsg.append(")");
throw MakeThorException(TE_WorkUnitWriteLimitExceeded, "%s", errMsg.str());
}
resultData.append(l, mb.readDirect(l));
mb.clear();
numResults += numGot;
if (-1 != flushThreshold && resultData.length() >= (unsigned)flushThreshold)
flushResults();
}
}
flushResults(true);
}
virtual void process() override
{
ActPrintLog("INDEXWRITE: Start");
init();
IRowStream *stream = inputStream;
ThorDataLinkMetaInfo info;
input->getMetaInfo(info);
outRowAllocator.setown(getRowAllocator(helper->queryDiskRecordSize()));
start();
if (refactor)
{
assertex(isLocal);
if (active)
{
unsigned targetWidth = partDesc->queryOwner().numParts()-(buildTlk?1:0);
assertex(0 == container.queryJob().querySlaves() % targetWidth);
unsigned partsPerNode = container.queryJob().querySlaves() / targetWidth;
unsigned myPart = queryJobChannel().queryMyRank();
IArrayOf<IRowStream> streams;
streams.append(*LINK(stream));
--partsPerNode;
// Should this be merging 1,11,21,31 etc.
unsigned p=0;
unsigned fromPart = targetWidth+1 + (partsPerNode * (myPart-1));
for (; p<partsPerNode; p++)
{
streams.append(*createRowStreamFromNode(*this, fromPart++, queryJobChannel().queryJobComm(), mpTag, abortSoon));
}
ICompare *icompare = helper->queryCompare();
assertex(icompare);
Owned<IRowLinkCounter> linkCounter = new CThorRowLinkCounter;
myInputStream.setown(createRowStreamMerger(streams.ordinality(), streams.getArray(), icompare, false, linkCounter));
stream = myInputStream;
}
else // serve nodes, creating merged parts
rowServer.setown(createRowServer(this, stream, queryJobChannel().queryJobComm(), mpTag));
}
processed = THORDATALINK_STARTED;
// single part key support
// has to serially pull all data fron nodes 2-N
// nodes 2-N, could/should start pushing some data (as it's supposed to be small) to cut down on serial nature.
unsigned node = queryJobChannel().queryMyRank();
if (singlePartKey)
{
if (1 == node)
{
try
{
open(*partDesc, false, helper->queryDiskRecordSize()->isVariableSize());
loop
{
OwnedConstThorRow row = inputStream->ungroupedNextRow();
if (!row)
break;
if (abortSoon) return;
processRow(row);
}
unsigned node = 2;
while (node <= container.queryJob().querySlaves())
{
Linked<IOutputRowDeserializer> deserializer = ::queryRowDeserializer(input);
CMessageBuffer mb;
Owned<ISerialStream> stream = createMemoryBufferSerialStream(mb);
CThorStreamDeserializerSource rowSource;
rowSource.setStream(stream);
bool successSR;
loop
{
{
BooleanOnOff tf(receivingTag2);
successSR = queryJobChannel().queryJobComm().sendRecv(mb, node, mpTag2);
}
if (successSR)
{
if (rowSource.eos())
break;
Linked<IEngineRowAllocator> allocator = ::queryRowAllocator(input);
do
{
RtlDynamicRowBuilder rowBuilder(allocator);
size32_t sz = deserializer->deserialize(rowBuilder, rowSource);
OwnedConstThorRow fRow = rowBuilder.finalizeRowClear(sz);
processRow(fRow);
}
while (!rowSource.eos());
}
}
node++;
}
}
catch (CATCHALL)
{
close(*partDesc, partCrc, true);
throw;
}
close(*partDesc, partCrc, true);
doStopInput();
}
else
{
CMessageBuffer mb;
CMemoryRowSerializer mbs(mb);
Linked<IOutputRowSerializer> serializer = ::queryRowSerializer(input);
loop
{
BooleanOnOff tf(receivingTag2);
if (queryJobChannel().queryJobComm().recv(mb, 1, mpTag2)) // node 1 asking for more..
{
if (abortSoon) break;
mb.clear();
do
{
OwnedConstThorRow row = inputStream->ungroupedNextRow();
if (!row) break;
serializer->serialize(mbs, (const byte *)row.get());
} while (mb.length() < SINGLEPART_KEY_TRANSFER_SIZE); // NB: at least one row
if (!queryJobChannel().queryJobComm().reply(mb))
throw MakeThorException(0, "Failed to send index data to node 1, from node %d", node);
if (0 == mb.length())
break;
}
}
}
}
void init(MemoryBuffer &data, MemoryBuffer &slaveData)
{
isLocal = 0 != (TIWlocal & helper->getFlags());
mpTag = container.queryJob().deserializeMPTag(data);
mpTag2 = container.queryJob().deserializeMPTag(data);
data.read(active);
if (active)
{
data.read(logicalFilename);
partDesc.setown(deserializePartFileDescriptor(data));
}
data.read(singlePartKey);
data.read(refactor);
if (singlePartKey)
buildTlk = false;
else
{
data.read(buildTlk);
if (firstNode())
{
if (buildTlk)
tlkDesc.setown(deserializePartFileDescriptor(data));
else if (!isLocal) // exising tlk then..
{
OwnedRoxieString diName(helper->getDistributeIndexName());
assertex(diName.get());
tlkDesc.setown(deserializePartFileDescriptor(data));
unsigned c;
data.read(c);
while (c--)
{
RemoteFilename rf;
rf.deserialize(data);
if (!existingTlkIFile)
{
Owned<IFile> iFile = createIFile(rf);
if (iFile->exists())
existingTlkIFile.set(iFile);
}
}
if (!existingTlkIFile)
throw MakeThorException(TE_FileNotFound, "Top level key part does not exist, for key: %s", diName.get());
}
}
}
IOutputMetaData * diskSize = helper->queryDiskRecordSize();
assertex(!(diskSize->getMetaFlags() & MDFneedserializedisk));
if (diskSize->isVariableSize())
{
if (TIWmaxlength & helper->getFlags())
maxDiskRecordSize = helper->getMaxKeySize();
else
maxDiskRecordSize = KEYBUILD_MAXLENGTH; //Current default behaviour, could be improved in the future
}
else
maxDiskRecordSize = diskSize->getFixedSize();
reportOverflow = false;
}
void start()
{
ActivityTimer s(totalCycles, timeActivities);
input = inputs.item(0);
try
{
try {
startInput(input);
}
catch (IException *e)
{
fireException(e);
barrier->cancel();
throw;
}
catch (CATCHALL)
{
Owned<IException> e = MakeActivityException(this, 0, "Unknown exception starting sort input");
fireException(e);
barrier->cancel();
throw;
}
dataLinkStart();
Linked<IRowInterfaces> rowif = queryRowInterfaces(input);
Owned<IRowInterfaces> auxrowif = createRowInterfaces(helper->querySortedRecordSize(),queryActivityId(),queryCodeContext());
sorter->Gather(
rowif,
input,
helper->queryCompare(),
helper->queryCompareLeftRight(),
NULL,helper->querySerialize(),
NULL,
false,
isUnstable(),
abortSoon,
auxrowif);
stopInput(input);
input = NULL;
if (abortSoon)
{
ActPrintLogEx(&queryContainer(), thorlog_null, MCwarning, "MSortSlaveActivity::start aborting");
barrier->cancel();
return;
}
}
catch (IException *e)
{
fireException(e);
barrier->cancel();
throw;
}
catch (CATCHALL)
{
Owned<IException> e = MakeActivityException(this, 0, "Unknown exception gathering sort input");
fireException(e);
barrier->cancel();
throw;
}
ActPrintLog("SORT waiting barrier.1");
if (!barrier->wait(false)) {
Sleep(1000); // let original error through
throw MakeThorException(TE_BarrierAborted,"SORT: Barrier Aborted");
}
ActPrintLog("SORT barrier.1 raised");
output.setown(sorter->startMerge(totalrows));
}
void CWriteMasterBase::publish()
{
if (published) return;
published = true;
if (!(diskHelperBase->getFlags() & (TDXtemporary|TDXjobtemp)))
updateActivityResult(container.queryJob().queryWorkUnit(), diskHelperBase->getFlags(), diskHelperBase->getSequence(), fileName, recordsProcessed);
IPropertyTree &props = fileDesc->queryProperties();
props.setPropInt64("@recordCount", recordsProcessed);
if (0 == (diskHelperBase->getFlags() & TDXtemporary) || container.queryJob().queryUseCheckpoints())
{
if (0 != (diskHelperBase->getFlags() & TDWexpires))
setExpiryTime(props, diskHelperBase->getExpiryDays());
if (TDWupdate & diskHelperBase->getFlags())
{
unsigned eclCRC;
unsigned __int64 totalCRC;
diskHelperBase->getUpdateCRCs(eclCRC, totalCRC);
props.setPropInt("@eclCRC", eclCRC);
props.setPropInt64("@totalCRC", totalCRC);
}
}
container.queryTempHandler()->registerFile(fileName, container.queryOwner().queryGraphId(), diskHelperBase->getTempUsageCount(), TDXtemporary & diskHelperBase->getFlags(), getDiskOutputKind(diskHelperBase->getFlags()), &clusters);
if (!dlfn.isExternal())
{
bool temporary = 0 != (diskHelperBase->getFlags()&TDXtemporary);
if (!temporary && (queryJob().querySlaves() < fileDesc->numParts()))
{
// create empty parts for a fileDesc being published that is larger than this clusters
size32_t recordSize = 0;
IOutputMetaData *diskRowMeta = diskHelperBase->queryDiskRecordSize()->querySerializedDiskMeta();
if (diskRowMeta->isFixedSize() && (TAKdiskwrite == container.getKind()))
{
recordSize = diskRowMeta->getMinRecordSize();
if (0 != (diskHelperBase->getFlags() & TDXgrouped))
recordSize += 1;
}
unsigned compMethod = COMPRESS_METHOD_LZW;
// rowdiff used if recordSize > 0, else fallback to compMethod
if (getOptBool(THOROPT_COMP_FORCELZW, false))
{
recordSize = 0; // by default if fixed length (recordSize set), row diff compression is used. This forces compMethod.
compMethod = COMPRESS_METHOD_LZW;
}
else if (getOptBool(THOROPT_COMP_FORCEFLZ, false))
compMethod = COMPRESS_METHOD_FASTLZ;
else if (getOptBool(THOROPT_COMP_FORCELZ4, false))
compMethod = COMPRESS_METHOD_LZ4;
bool blockCompressed;
bool compressed = fileDesc->isCompressed(&blockCompressed);
for (unsigned clusterIdx=0; clusterIdx<fileDesc->numClusters(); clusterIdx++)
{
StringBuffer clusterName;
fileDesc->getClusterGroupName(clusterIdx, clusterName, &queryNamedGroupStore());
PROGLOG("Creating blank parts for file '%s', cluster '%s'", fileName.get(), clusterName.str());
unsigned p=0;
while (p<fileDesc->numParts())
{
if (p == targetOffset)
p += queryJob().querySlaves();
IPartDescriptor *partDesc = fileDesc->queryPart(p);
CDateTime createTime, modifiedTime;
for (unsigned c=0; c<partDesc->numCopies(); c++)
{
RemoteFilename rfn;
partDesc->getFilename(c, rfn);
StringBuffer path;
rfn.getPath(path);
try
{
ensureDirectoryForFile(path.str());
OwnedIFile iFile = createIFile(path.str());
Owned<IFileIO> iFileIO;
if (compressed)
iFileIO.setown(createCompressedFileWriter(iFile, recordSize, false, true, NULL, compMethod));
else
iFileIO.setown(iFile->open(IFOcreate));
dbgassertex(iFileIO.get());
iFileIO.clear();
// ensure copies have matching datestamps, as they would do normally (backupnode expects it)
if (partDesc->numCopies() > 1)
{
if (0 == c)
iFile->getTime(&createTime, &modifiedTime, NULL);
else
iFile->setTime(&createTime, &modifiedTime, NULL);
}
}
catch (IException *e)
{
if (0 == c)
throw;
Owned<IThorException> e2 = MakeThorException(e);
e->Release();
e2->setAction(tea_warning);
queryJob().fireException(e2);
}
}
partDesc->queryProperties().setPropInt64("@size", 0);
p++;
}
clusterIdx++;
}
}
queryThorFileManager().publish(container.queryJob(), fileName, *fileDesc, NULL);
}
}
virtual void init()
{
CMasterActivity::init();
OwnedRoxieString indexFileName(helper->getIndexFileName());
Owned<IDistributedFile> dataFile;
Owned<IDistributedFile> indexFile = queryThorFileManager().lookup(container.queryJob(), indexFileName, false, 0 != (helper->getJoinFlags() & JFindexoptional), true);
unsigned keyReadWidth = (unsigned)container.queryJob().getWorkUnitValueInt("KJKRR", 0);
if (!keyReadWidth || keyReadWidth>container.queryJob().querySlaves())
keyReadWidth = container.queryJob().querySlaves();
initMb.clear();
initMb.append(indexFileName.get());
if (helper->diskAccessRequired())
numTags += 2;
initMb.append(numTags);
unsigned t=0;
for (; t<numTags; t++)
{
tags[t] = container.queryJob().allocateMPTag();
initMb.append(tags[t]);
}
bool keyHasTlk = false;
if (indexFile)
{
unsigned numParts = 0;
localKey = indexFile->queryAttributes().getPropBool("@local");
if (container.queryLocalData() && !localKey)
throw MakeActivityException(this, 0, "Keyed Join cannot be LOCAL unless supplied index is local");
checkFormatCrc(this, indexFile, helper->getIndexFormatCrc(), true);
Owned<IFileDescriptor> indexFileDesc = indexFile->getFileDescriptor();
IDistributedSuperFile *superIndex = indexFile->querySuperFile();
unsigned superIndexWidth = 0;
unsigned numSuperIndexSubs = 0;
if (superIndex)
{
numSuperIndexSubs = superIndex->numSubFiles(true);
bool first=true;
// consistency check
Owned<IDistributedFileIterator> iter = superIndex->getSubFileIterator(true);
ForEach(*iter)
{
IDistributedFile &f = iter->query();
unsigned np = f.numParts()-1;
IDistributedFilePart &part = f.queryPart(np);
const char *kind = part.queryAttributes().queryProp("@kind");
bool hasTlk = NULL != kind && 0 == stricmp("topLevelKey", kind); // if last part not tlk, then deemed local (might be singlePartKey)
if (first)
{
first = false;
keyHasTlk = hasTlk;
superIndexWidth = f.numParts();
if (keyHasTlk)
--superIndexWidth;
}
else
{
if (hasTlk != keyHasTlk)
throw MakeActivityException(this, 0, "Local/Single part keys cannot be mixed with distributed(tlk) keys in keyedjoin");
if (keyHasTlk && superIndexWidth != f.numParts()-1)
throw MakeActivityException(this, 0, "Super sub keys of different width cannot be mixed with distributed(tlk) keys in keyedjoin");
}
}
if (keyHasTlk)
numParts = superIndexWidth * numSuperIndexSubs;
else
numParts = superIndex->numParts();
}
else
{
numParts = indexFile->numParts();
if (numParts)
{
const char *kind = indexFile->queryPart(indexFile->numParts()-1).queryAttributes().queryProp("@kind");
keyHasTlk = NULL != kind && 0 == stricmp("topLevelKey", kind);
if (keyHasTlk)
--numParts;
}
}
if (numParts)
{
initMb.append(numParts);
initMb.append(superIndexWidth); // 0 if not superIndex
initMb.append((superIndex && superIndex->isInterleaved()) ? numSuperIndexSubs : 0);
unsigned p=0;
UnsignedArray parts;
for (; p<numParts; p++)
parts.append(p);
indexFileDesc->serializeParts(initMb, parts);
if (localKey)
keyHasTlk = false; // not used
initMb.append(keyHasTlk);
if (keyHasTlk)
{
if (numSuperIndexSubs)
initMb.append(numSuperIndexSubs);
else
initMb.append((unsigned)1);
Owned<IDistributedFileIterator> iter;
IDistributedFile *f;
if (superIndex)
{
iter.setown(superIndex->getSubFileIterator(true));
f = &iter->query();
}
else
f = indexFile;
loop
{
unsigned location;
OwnedIFile iFile;
StringBuffer filePath;
Owned<IFileDescriptor> fileDesc = f->getFileDescriptor();
Owned<IPartDescriptor> tlkDesc = fileDesc->getPart(fileDesc->numParts()-1);
if (!getBestFilePart(this, *tlkDesc, iFile, location, filePath))
throw MakeThorException(TE_FileNotFound, "Top level key part does not exist, for key: %s", f->queryLogicalName());
OwnedIFileIO iFileIO = iFile->open(IFOread);
assertex(iFileIO);
size32_t tlkSz = (size32_t)iFileIO->size();
initMb.append(tlkSz);
::read(iFileIO, 0, tlkSz, initMb);
if (!iter || !iter->next())
break;
f = &iter->query();
}
}
if (helper->diskAccessRequired())
{
OwnedRoxieString fetchFilename(helper->getFileName());
if (fetchFilename)
{
dataFile.setown(queryThorFileManager().lookup(container.queryJob(), fetchFilename, false, 0 != (helper->getFetchFlags() & FFdatafileoptional), true));
if (dataFile)
{
if (superIndex)
throw MakeActivityException(this, 0, "Superkeys and full keyed joins are not supported");
Owned<IFileDescriptor> dataFileDesc = getConfiguredFileDescriptor(*dataFile);
void *ekey;
size32_t ekeylen;
helper->getFileEncryptKey(ekeylen,ekey);
bool encrypted = dataFileDesc->queryProperties().getPropBool("@encrypted");
if (0 != ekeylen)
{
memset(ekey,0,ekeylen);
free(ekey);
if (!encrypted)
{
Owned<IException> e = MakeActivityWarning(&container, TE_EncryptionMismatch, "Ignoring encryption key provided as file '%s' was not published as encrypted", helper->getFileName());
container.queryJob().fireException(e);
}
}
else if (encrypted)
throw MakeActivityException(this, 0, "File '%s' was published as encrypted but no encryption key provided", fetchFilename.get());
unsigned dataReadWidth = (unsigned)container.queryJob().getWorkUnitValueInt("KJDRR", 0);
if (!dataReadWidth || dataReadWidth>container.queryJob().querySlaves())
dataReadWidth = container.queryJob().querySlaves();
Owned<IGroup> grp = container.queryJob().querySlaveGroup().subset((unsigned)0, dataReadWidth);
dataFileMapping.setown(getFileSlaveMaps(dataFile->queryLogicalName(), *dataFileDesc, container.queryJob().queryUserDescriptor(), *grp, false, false, NULL));
dataFileMapping->serializeFileOffsetMap(offsetMapMb.clear());
}
else
indexFile.clear();
}
}
}
else
void prepareKey(IDistributedFile *index)
{
IDistributedFile *f = index;
IDistributedSuperFile *super = f->querySuperFile();
unsigned nparts = f->numParts(); // includes tlks if any, but unused in array
performPartLookup.ensure(nparts);
bool checkTLKConsistency = (nullptr != super) && !localKey && (0 != (TIRsorted & indexBaseHelper->getFlags()));
if (nofilter)
{
while (nparts--) performPartLookup.append(true);
if (!checkTLKConsistency) return;
}
else
{
while (nparts--) performPartLookup.append(false); // parts to perform lookup set later
}
Owned<IDistributedFileIterator> iter;
if (super)
{
iter.setown(super->getSubFileIterator(true));
verifyex(iter->first());
f = &iter->query();
}
unsigned width = f->numParts();
if (!localKey)
--width;
assertex(width);
unsigned tlkCrc = 0;
bool first = true;
unsigned superSubIndex=0;
bool fileCrc = false, rowCrc = false;
for (;;)
{
Owned<IDistributedFilePart> part = f->getPart(width);
if (checkTLKConsistency)
{
unsigned _tlkCrc;
if (part->getCrc(_tlkCrc))
fileCrc = true;
else if (part->queryAttributes().hasProp("@crc")) // NB: key "@crc" is not a crc on the file, but data within.
{
_tlkCrc = part->queryAttributes().getPropInt("@crc");
rowCrc = true;
}
else if (part->queryAttributes().hasProp("@tlkCrc")) // backward compat.
{
_tlkCrc = part->queryAttributes().getPropInt("@tlkCrc");
rowCrc = true;
}
else
{
if (rowCrc || fileCrc)
{
checkTLKConsistency = false;
Owned<IException> e = MakeActivityWarning(&container, 0, "Cannot validate that tlks in superfile %s match, some crc attributes are missing", super->queryLogicalName());
queryJobChannel().fireException(e);
}
}
if (rowCrc && fileCrc)
{
checkTLKConsistency = false;
Owned<IException> e = MakeActivityWarning(&container, 0, "Cannot validate that tlks in superfile %s match, due to mixed crc types.", super->queryLogicalName());
queryJobChannel().fireException(e);
}
if (checkTLKConsistency)
{
if (first)
{
tlkCrc = _tlkCrc;
first = false;
}
else if (tlkCrc != _tlkCrc)
throw MakeActivityException(this, 0, "Sorted output on super files comprising of non coparitioned sub keys is not supported (TLK's do not match)");
}
}
if (!nofilter)
{
Owned<IKeyIndex> keyIndex;
unsigned copy;
for (copy=0; copy<part->numCopies(); copy++)
{
RemoteFilename rfn;
OwnedIFile ifile = createIFile(part->getFilename(rfn,copy));
if (ifile->exists())
{
StringBuffer remotePath;
rfn.getRemotePath(remotePath);
unsigned crc = 0;
part->getCrc(crc);
keyIndex.setown(createKeyIndex(remotePath.str(), crc, false, false));
break;
}
}
if (!keyIndex)
throw MakeThorException(TE_FileNotFound, "Top level key part does not exist, for key: %s", index->queryLogicalName());
unsigned fixedSize = indexBaseHelper->queryDiskRecordSize()->querySerializedDiskMeta()->getFixedSize(); // used only if fixed
Owned <IKeyManager> tlk = createLocalKeyManager(keyIndex, fixedSize, NULL);
indexBaseHelper->createSegmentMonitors(tlk);
tlk->finishSegmentMonitors();
tlk->reset();
while (tlk->lookup(false))
{
if (tlk->queryFpos())
performPartLookup.replace(true, (aindex_t)(super?super->numSubFiles(true)*(tlk->queryFpos()-1)+superSubIndex:tlk->queryFpos()-1));
}
}
if (!super||!iter->next())
break;
superSubIndex++;
f = &iter->query();
if (width != f->numParts()-1)
throw MakeActivityException(this, 0, "Super key %s, with mixture of sub key width are not supported.", f->queryLogicalName());
}
}
virtual void process() override
{
ActPrintLog("INDEXWRITE: Start");
init();
IRowStream *stream = inputStream;
ThorDataLinkMetaInfo info;
input->getMetaInfo(info);
outRowAllocator.setown(getRowAllocator(helper->queryDiskRecordSize()));
start();
if (refactor)
{
assertex(isLocal);
if (active)
{
unsigned targetWidth = partDesc->queryOwner().numParts()-(buildTlk?1:0);
assertex(0 == container.queryJob().querySlaves() % targetWidth);
unsigned partsPerNode = container.queryJob().querySlaves() / targetWidth;
unsigned myPart = queryJobChannel().queryMyRank();
IArrayOf<IRowStream> streams;
streams.append(*LINK(stream));
--partsPerNode;
// Should this be merging 1,11,21,31 etc.
unsigned p=0;
unsigned fromPart = targetWidth+1 + (partsPerNode * (myPart-1));
for (; p<partsPerNode; p++)
{
streams.append(*createRowStreamFromNode(*this, fromPart++, queryJobChannel().queryJobComm(), mpTag, abortSoon));
}
ICompare *icompare = helper->queryCompare();
assertex(icompare);
Owned<IRowLinkCounter> linkCounter = new CThorRowLinkCounter;
myInputStream.setown(createRowStreamMerger(streams.ordinality(), streams.getArray(), icompare, false, linkCounter));
stream = myInputStream;
}
else // serve nodes, creating merged parts
rowServer.setown(createRowServer(this, stream, queryJobChannel().queryJobComm(), mpTag));
}
processed = THORDATALINK_STARTED;
// single part key support
// has to serially pull all data fron nodes 2-N
// nodes 2-N, could/should start pushing some data (as it's supposed to be small) to cut down on serial nature.
unsigned node = queryJobChannel().queryMyRank();
if (singlePartKey)
{
if (1 == node)
{
try
{
open(*partDesc, false, helper->queryDiskRecordSize()->isVariableSize());
for (;;)
{
OwnedConstThorRow row = inputStream->ungroupedNextRow();
if (!row)
break;
if (abortSoon) return;
processRow(row);
}
unsigned node = 2;
while (node <= container.queryJob().querySlaves())
{
Linked<IOutputRowDeserializer> deserializer = ::queryRowDeserializer(input);
CMessageBuffer mb;
Owned<ISerialStream> stream = createMemoryBufferSerialStream(mb);
CThorStreamDeserializerSource rowSource;
rowSource.setStream(stream);
bool successSR;
for (;;)
{
{
BooleanOnOff tf(receivingTag2);
successSR = queryJobChannel().queryJobComm().sendRecv(mb, node, mpTag2);
}
if (successSR)
{
if (rowSource.eos())
break;
Linked<IEngineRowAllocator> allocator = ::queryRowAllocator(input);
do
{
RtlDynamicRowBuilder rowBuilder(allocator);
size32_t sz = deserializer->deserialize(rowBuilder, rowSource);
OwnedConstThorRow fRow = rowBuilder.finalizeRowClear(sz);
processRow(fRow);
}
while (!rowSource.eos());
}
}
node++;
}
}
catch (CATCHALL)
{
close(*partDesc, partCrc, true);
throw;
}
close(*partDesc, partCrc, true);
stop();
}
else
{
CMessageBuffer mb;
CMemoryRowSerializer mbs(mb);
Linked<IOutputRowSerializer> serializer = ::queryRowSerializer(input);
for (;;)
{
BooleanOnOff tf(receivingTag2);
if (queryJobChannel().queryJobComm().recv(mb, 1, mpTag2)) // node 1 asking for more..
{
if (abortSoon) break;
mb.clear();
do
{
OwnedConstThorRow row = inputStream->ungroupedNextRow();
if (!row) break;
serializer->serialize(mbs, (const byte *)row.get());
} while (mb.length() < SINGLEPART_KEY_TRANSFER_SIZE); // NB: at least one row
if (!queryJobChannel().queryJobComm().reply(mb))
throw MakeThorException(0, "Failed to send index data to node 1, from node %d", node);
if (0 == mb.length())
break;
}
}
}
}
else
{
if (!refactor || active)
{
try
{
StringBuffer partFname;
getPartFilename(*partDesc, 0, partFname);
ActPrintLog("INDEXWRITE: process: handling fname : %s", partFname.str());
open(*partDesc, false, helper->queryDiskRecordSize()->isVariableSize());
ActPrintLog("INDEXWRITE: write");
BooleanOnOff tf(receiving);
if (!refactor || !active)
receiving = false;
do
{
OwnedConstThorRow row = inputStream->ungroupedNextRow();
if (!row)
break;
processRow(row);
} while (!abortSoon);
ActPrintLog("INDEXWRITE: write level 0 complete");
}
catch (CATCHALL)
{
close(*partDesc, partCrc, isLocal && !buildTlk && 1 == node);
throw;
}
close(*partDesc, partCrc, isLocal && !buildTlk && 1 == node);
stop();
ActPrintLog("INDEXWRITE: Wrote %" RCPF "d records", processed & THORDATALINK_COUNT_MASK);
if (buildTlk)
{
ActPrintLog("INDEXWRITE: sending rows");
NodeInfoArray tlkRows;
CMessageBuffer msg;
if (firstNode())
{
if (processed & THORDATALINK_COUNT_MASK)
{
if (enableTlkPart0)
tlkRows.append(* new CNodeInfo(0, firstRow.get(), firstRowSize, totalCount));
tlkRows.append(* new CNodeInfo(1, lastRow.get(), lastRowSize, totalCount));
}
}
else
{
if (processed & THORDATALINK_COUNT_MASK)
{
CNodeInfo row(queryJobChannel().queryMyRank(), lastRow.get(), lastRowSize, totalCount);
row.serialize(msg);
}
queryJobChannel().queryJobComm().send(msg, 1, mpTag);
}
if (firstNode())
{
ActPrintLog("INDEXWRITE: Waiting on tlk to complete");
// JCSMORE if refactor==true, is rowsToReceive here right??
unsigned rowsToReceive = (refactor ? (tlkDesc->queryOwner().numParts()-1) : container.queryJob().querySlaves()) -1; // -1 'cos got my own in array already
ActPrintLog("INDEXWRITE: will wait for info from %d slaves before writing TLK", rowsToReceive);
while (rowsToReceive--)
{
msg.clear();
receiveMsg(msg, RANK_ALL, mpTag); // NH->JCS RANK_ALL_OTHER not supported for recv
if (abortSoon)
return;
if (msg.length())
{
CNodeInfo *ni = new CNodeInfo();
ni->deserialize(msg);
tlkRows.append(*ni);
}
}
tlkRows.sort(CNodeInfo::compare);
StringBuffer path;
getPartFilename(*tlkDesc, 0, path);
ActPrintLog("INDEXWRITE: creating toplevel key file : %s", path.str());
try
{
open(*tlkDesc, true, helper->queryDiskRecordSize()->isVariableSize());
if (tlkRows.length())
{
CNodeInfo &lastNode = tlkRows.item(tlkRows.length()-1);
memset(lastNode.value, 0xff, lastNode.size);
}
ForEachItemIn(idx, tlkRows)
{
CNodeInfo &info = tlkRows.item(idx);
builder->processKeyData((char *)info.value, info.pos, info.size);
}
close(*tlkDesc, tlkCrc, true);
}
catch (CATCHALL)
{
abortSoon = true;
close(*tlkDesc, tlkCrc, true);
removeFiles(*partDesc);
throw;
}
}
}
else if (!isLocal && firstNode())