|
|
@@ -388,13 +388,6 @@ CMasterActivity::~CMasterActivity()
|
|
|
delete [] data;
|
|
|
}
|
|
|
|
|
|
-void CMasterActivity::addReadFile(IDistributedFile *file, bool temp)
|
|
|
-{
|
|
|
- readFiles.append(*LINK(file));
|
|
|
- if (!temp) // NB: Temps not listed in workunit
|
|
|
- queryThorFileManager().noteFileRead(container.queryJob(), file);
|
|
|
-}
|
|
|
-
|
|
|
IDistributedFile *CMasterActivity::queryReadFile(unsigned f)
|
|
|
{
|
|
|
if (f>=readFiles.ordinality())
|
|
|
@@ -402,20 +395,31 @@ IDistributedFile *CMasterActivity::queryReadFile(unsigned f)
|
|
|
return &readFiles.item(f);
|
|
|
}
|
|
|
|
|
|
-void CMasterActivity::preStart(size32_t parentExtractSz, const byte *parentExtract)
|
|
|
+IDistributedFile *CMasterActivity::findReadFile(const char *lfnName)
|
|
|
+{
|
|
|
+ auto it = readFilesMap.find(lfnName);
|
|
|
+ if (it != readFilesMap.end())
|
|
|
+ return LINK(it->second);
|
|
|
+ return nullptr;
|
|
|
+}
|
|
|
+
|
|
|
+IDistributedFile *CMasterActivity::lookupReadFile(const char *lfnName, bool jobTemp, bool temp, bool opt)
|
|
|
{
|
|
|
- CActivityBase::preStart(parentExtractSz, parentExtract);
|
|
|
- IArrayOf<IDistributedFile> tmpFiles;
|
|
|
- tmpFiles.swapWith(readFiles);
|
|
|
- ForEachItemIn(f, tmpFiles)
|
|
|
+ StringBuffer normalizedFileName;
|
|
|
+ queryThorFileManager().addScope(container.queryJob(), lfnName, normalizedFileName, jobTemp|temp);
|
|
|
+ Owned<IDistributedFile> file = findReadFile(normalizedFileName);
|
|
|
+ if (!file)
|
|
|
{
|
|
|
- IDistributedFile &file = tmpFiles.item(f);
|
|
|
- IDistributedSuperFile *super = file.querySuperFile();
|
|
|
- if (super)
|
|
|
- getSuperFileSubs(super, readFiles, true);
|
|
|
- else
|
|
|
- readFiles.append(*LINK(&file));
|
|
|
+ file.setown(queryThorFileManager().lookup(container.queryJob(), lfnName, jobTemp|temp, opt, true, container.activityIsCodeSigned()));
|
|
|
+ if (file)
|
|
|
+ {
|
|
|
+ readFiles.append(*LINK(file));
|
|
|
+ readFilesMap[normalizedFileName.str()] = file;
|
|
|
+ if (!temp) // NB: Temps not listed in workunit
|
|
|
+ queryThorFileManager().noteFileRead(container.queryJob(), file);
|
|
|
+ }
|
|
|
}
|
|
|
+ return file.getClear();
|
|
|
}
|
|
|
|
|
|
MemoryBuffer &CMasterActivity::queryInitializationData(unsigned slave) const
|
|
|
@@ -455,7 +459,16 @@ void CMasterActivity::threadmain()
|
|
|
|
|
|
void CMasterActivity::init()
|
|
|
{
|
|
|
- readFiles.kill();
|
|
|
+ // Files are added to readFiles during initialization,
|
|
|
+ // If this is a CQ query act., then it will be repeatedly re-initialized if it depends on master context,
|
|
|
+ // e.g. due to a variable filename.
|
|
|
+ // Therefore, do not clear readFiles on reinitialization, to avoid repeatedly [expensively] looking them up.
|
|
|
+ bool inCQ = container.queryOwner().queryOwner() && !container.queryOwner().isGlobal();
|
|
|
+ if (!inCQ)
|
|
|
+ {
|
|
|
+ readFiles.kill();
|
|
|
+ readFilesMap.clear();
|
|
|
+ }
|
|
|
}
|
|
|
|
|
|
void CMasterActivity::startProcess(bool async)
|
|
|
@@ -480,6 +493,7 @@ void CMasterActivity::kill()
|
|
|
{
|
|
|
CActivityBase::kill();
|
|
|
readFiles.kill();
|
|
|
+ readFilesMap.clear();
|
|
|
}
|
|
|
|
|
|
bool CMasterActivity::fireException(IException *_e)
|
|
|
@@ -542,35 +556,84 @@ void CMasterActivity::getEdgeStats(IStatisticGatherer & stats, unsigned idx)
|
|
|
void CMasterActivity::done()
|
|
|
{
|
|
|
CActivityBase::done();
|
|
|
- ForEachItemIn(s, readFiles)
|
|
|
+ if (readFiles.ordinality())
|
|
|
{
|
|
|
- IDistributedFile &file = readFiles.item(s);
|
|
|
- file.setAccessed();
|
|
|
+ IArrayOf<IDistributedFile> tmpFiles;
|
|
|
+ ForEachItemIn(f, readFiles)
|
|
|
+ {
|
|
|
+ IDistributedFile &file = readFiles.item(f);
|
|
|
+ IDistributedSuperFile *super = file.querySuperFile();
|
|
|
+ if (super)
|
|
|
+ {
|
|
|
+ getSuperFileSubs(super, tmpFiles, true);
|
|
|
+ tmpFiles.append(*LINK(&file));
|
|
|
+ }
|
|
|
+ else
|
|
|
+ tmpFiles.append(*LINK(&file));
|
|
|
+ }
|
|
|
+ ForEachItemIn(s, tmpFiles)
|
|
|
+ {
|
|
|
+ IDistributedFile &file = tmpFiles.item(s);
|
|
|
+ file.setAccessed();
|
|
|
+ }
|
|
|
}
|
|
|
}
|
|
|
|
|
|
void CMasterActivity::updateFileReadCostStats(std::vector<OwnedPtr<CThorStatsCollection>> & subFileStats)
|
|
|
{
|
|
|
- if (!subFileStats.empty())
|
|
|
- {
|
|
|
- unsigned numSubFiles = subFileStats.size();
|
|
|
- for (unsigned i=0; i<numSubFiles; i++)
|
|
|
- {
|
|
|
- IDistributedFile *file = queryReadFile(i);
|
|
|
- if (file)
|
|
|
+ /* JCSMORE->SHAMSER: (separate JIRA needed)
|
|
|
+ * there can be >1 read file if this act. is in a child query/loop, it could be processing a different file per iteration,
|
|
|
+ * meaning there could be an arbitrary number of readfiles, in that case activity::init would be called multiple times.
|
|
|
+ *
|
|
|
+ * I hit an assert during testing reading a super in a CQ:
|
|
|
+ libjlib.so!raiseAssertException(const char * assertion, const char * file, unsigned int line) (\home\jsmith\git\HPCC-Platform\system\jlib\jexcept.cpp:660)
|
|
|
+ libjlib.so!MemoryBuffer::read(MemoryBuffer * const this, unsigned char & value) (\home\jsmith\git\HPCC-Platform\system\jlib\jbuff.cpp:693)
|
|
|
+ libjlib.so!MemoryBuffer::readPacked(MemoryBuffer * const this) (\home\jsmith\git\HPCC-Platform\system\jlib\jbuff.cpp:813)
|
|
|
+ libjlib.so!MemoryBuffer::readPacked(MemoryBuffer * const this, unsigned int & value) (\home\jsmith\git\HPCC-Platform\system\jlib\jbuff.cpp:824)
|
|
|
+ libjlib.so!CRuntimeStatisticCollection::deserialize(CRuntimeStatisticCollection * const this, MemoryBuffer & in) (\home\jsmith\git\HPCC-Platform\system\jlib\jstats.cpp:2524)
|
|
|
+ libactivitymasters_lcr.so!CThorStatsCollection::deserialize(CThorStatsCollection * const this, unsigned int node, MemoryBuffer & mb) (\home\jsmith\git\HPCC-Platform\thorlcr\graph\thgraphmaster.ipp:53)
|
|
|
+ libactivitymasters_lcr.so!CDiskReadMasterBase::deserializeStats(CDiskReadMasterBase * const this, unsigned int node, MemoryBuffer & mb) (\home\jsmith\git\HPCC-Platform\thorlcr\activities\thdiskbase.cpp:139)
|
|
|
+ libgraphmaster_lcr.so!CMasterGraph::deserializeStats(CMasterGraph * const this, unsigned int node, MemoryBuffer & mb) (\home\jsmith\git\HPCC-Platform\thorlcr\graph\thgraphmaster.cpp:2781)
|
|
|
+ *
|
|
|
+ * (would be a crash in a Release build)
|
|
|
+ * it's because the diskread init is adding new CThorStatsCollection per init (per execution of the CQ),
|
|
|
+ * which means when deserializing there are too many. The code assumes there is only 1 file being read.
|
|
|
+ *
|
|
|
+ * I've temporarily changed the code where subFileStats's are added, to prevent more being added per iteration,
|
|
|
+ * but it needs re-thinking to handle the workers potentially dealing with different logical files
|
|
|
+ * (+ index read to handle super files, and case where act. is reading >1 file, i.e. KJ)
|
|
|
+ * I've changed this code rely on the 1st readFiles for now (not the # of subFileStats, which may be more)
|
|
|
+ * NB: also changed when the expansion of readFiles (from supers to subfiles) happens, a new super was being added
|
|
|
+ * each CQ iteration and re-expanded, meaing readFiles kept growing.
|
|
|
+ *
|
|
|
+ * Also, superkey1.ecl hits a dbgasserex whilst deserializaing stats (before and after these PR changes),
|
|
|
+ * but is caught/ignored. I haven't investigated further.
|
|
|
+ */
|
|
|
+
|
|
|
+ IDistributedFile *file = queryReadFile(0);
|
|
|
+ if (file)
|
|
|
+ {
|
|
|
+ IDistributedSuperFile *super = file->querySuperFile();
|
|
|
+ if (super)
|
|
|
+ {
|
|
|
+ unsigned numSubFiles = super->numSubFiles(true); //subFileStats.size();
|
|
|
+ if (subFileStats.size())
|
|
|
{
|
|
|
- stat_type numDiskReads = subFileStats[i]->getStatisticSum(StNumDiskReads);
|
|
|
- StringBuffer clusterName;
|
|
|
- file->getClusterName(0, clusterName);
|
|
|
- diskAccessCost += money2cost_type(calcFileAccessCost(clusterName, 0, numDiskReads));
|
|
|
- file->addAttrValue("@numDiskReads", numDiskReads);
|
|
|
+ assertex(numSubFiles <= subFileStats.size());
|
|
|
+ for (unsigned i=0; i<subFileStats.size(); i++)
|
|
|
+ {
|
|
|
+ IDistributedFile &subFile = super->querySubFile(i, true);
|
|
|
+ const char *subName = subFile.queryLogicalName();
|
|
|
+ PROGLOG("subName = %s", subName);
|
|
|
+ stat_type numDiskReads = subFileStats[i]->getStatisticSum(StNumDiskReads);
|
|
|
+ StringBuffer clusterName;
|
|
|
+ subFile.getClusterName(0, clusterName);
|
|
|
+ diskAccessCost += money2cost_type(calcFileAccessCost(clusterName, 0, numDiskReads));
|
|
|
+ subFile.addAttrValue("@numDiskReads", numDiskReads);
|
|
|
+ }
|
|
|
}
|
|
|
}
|
|
|
- }
|
|
|
- else
|
|
|
- {
|
|
|
- IDistributedFile *file = queryReadFile(0);
|
|
|
- if (file)
|
|
|
+ else
|
|
|
{
|
|
|
stat_type numDiskReads = statsCollection.getStatisticSum(StNumDiskReads);
|
|
|
StringBuffer clusterName;
|
Jake Smith