void computeAndAddSelParamIfPossible(
CacheWA& cwa, BindWA& bindWA, ExprValueId& child,
BaseColumn *base, ConstValue *val)
{
ColStatsSharedPtr cStatsPtr = (base->getTableDesc()->tableColStats()).
getColStatsPtrForColumn(base->getValueId());
if (cStatsPtr == NULL )
return;
HistogramSharedPtr hist = cStatsPtr->getHistogram();
if ( hist == NULL )
return;
CostScalar sel;
NABoolean canComputeSelectivity = hist -> computeSelectivityForEquality(
val, cStatsPtr->getRowcount(), cStatsPtr->getTotalUec(),
sel);
if ( canComputeSelectivity == TRUE ) {
const NAType * newType = base->getNAColumn()->getType();
// for char datatypes, assign the caseinsensitive attribute from
// the const node.
if (newType->getTypeQualifier() == NA_CHARACTER_TYPE)
{
newType =
base->getNAColumn()->getType()->newCopy(cwa.wHeap());
((CharType*)newType)->setCaseinsensitive(((CharType*)val->getType())->isCaseinsensitive());
}
cwa.replaceWithNewOrOldSelParam(val, newType, Selectivity(sel),
child, bindWA);
}
}
// ---------------------------------------------------------------------
// Utility Routine: pickOutputs
//
// From the given ColStatDescList, populate columnStats_ with column
// descriptors that are useful based on the characteristic outputs for
// the group.
//
// Always include in the output the current histograms of the input data,
// and, if the histogram is contained in the required output list, then
// this is a useful histogram and will also be output.
//
// ---------------------------------------------------------------------
void EstLogProp::pickOutputs( ColStatDescList & columnStats,
const EstLogPropSharedPtr& inputEstLogProp,
const ValueIdSet specifiedOutputs,
const ValueIdSet predSet)
{
const ColStatDescList & outerColStatsList = inputEstLogProp->getColStats();
ValueIdSet colsRequiringHistograms = specifiedOutputs;
// (i) see if the selection predicates contain any constant value or a
// constant expression
// (ii) check if there are any columns of this table being joined to some other
// columns, which do not appear as characteristics outputs. There should be
// histograms available for these columns, as these might be needed later.
// This problem was seen for temporary tables created as normal_tables by the
// triggers.
colsRequiringHistograms.addSet(predSet.getColumnsForHistogram());
colStats().setMCSkewedValueLists(columnStats.getMCSkewedValueLists()) ;
NABoolean colStatDescAdded = FALSE;
for (CollIndex i=0; i < columnStats.entries(); i++)
{
// we probably don't need 'em all, but this is the easiest way to
// grab all of the multi-column uec information we'll need later
colStats().insertIntoUecList (columnStats.getUecList()) ;
colStats().setScanRowCountWithoutHint(columnStats.getScanRowCountWithoutHint());
NABoolean found = FALSE;
// Note: The following inserts into a ColStatDescList should not
// have to be deep copies. From this point on, ColStatDescs that
// describe the output of the calling operator are read-only.
ColStatDescSharedPtr colStatDesc = columnStats[i];
// the value-id we're looking for
const ValueId columnId = colStatDesc->getVEGColumn() ;
for (CollIndex j=0 ; j < outerColStatsList.entries() ; j++)
{
if (columnId == outerColStatsList[j]->getVEGColumn() OR
(CmpCommon::context()->showQueryStats()))
{
colStats().insert(colStatDesc) ;
found = TRUE;
if(!colStatDescAdded)
colStatDescAdded = TRUE;
break ; // jump to next ColStatDesc
}
}
// OK, the valueid doesn't match directly -- but there are still a
// couple of things to check in order to verify whether or not we're
// interested in keeping the i'th ColStatDesc ...
ValueId throwaway ; // used by the second clause below
if ( NOT found AND
(columnId != NULL_VALUE_ID) AND
(colsRequiringHistograms.contains (columnId) OR
colsRequiringHistograms.referencesTheGivenValue (columnId, throwaway) OR
columnId.isInvolvedInJoinAndConst() OR
CmpCommon::context()->showQueryStats() )
)
{
colStats().insert(colStatDesc);
found = TRUE;
if(!colStatDescAdded)
colStatDescAdded = TRUE;
}
if (CURRSTMT_OPTDEFAULTS->incorporateSkewInCosting())
{
// if the column is referenced for histogram, but is
// not needed beyond this time , then we shall save its
// max freq, which might be used later in costing if this
// column is a part of the partitioning key
ColStatsSharedPtr stat = colStatDesc->getColStats();
if (!(stat->isVirtualColForHist() ) && NOT found &&
!(stat->isOrigFakeHist() ) )
{
const ValueId col = colStatDesc->getColumn();
ColAnalysis * colAnalysis = col.colAnalysis();
if (colAnalysis)
{
NAColumn * column = stat->getStatColumns()[0];
if (column->isReferencedForHistogram())
{
CostScalar maxFreq = columnStats.getMaxFreq(columnId);
colAnalysis->setMaxFreq(maxFreq);
colAnalysis->setFinalUec(stat->getTotalUec());
colAnalysis->setFinalRC(stat->getRowcount());
}
}
}
}
} // for columnStats.entries()
if(!colStatDescAdded && columnStats.entries() > 0)
colStats().insert(columnStats[0]) ;
} // pickOutputs
EstLogPropSharedPtr AppliedStatMan::getStatsForCANodeId(
CANodeId jbbc,
const EstLogPropSharedPtr &inLP,
const ValueIdSet * predIdSet)
{
EstLogPropSharedPtr inputLP = inLP;
if(inputLP == (*GLOBAL_EMPTY_INPUT_LOGPROP))
inputLP = jbbc.getJBBInput();
EstLogPropSharedPtr outputEstLogProp = NULL;
// 1. Try to find Logical Properties from cache if cacheable.
// The estimate Logical Properties can be cacheable if all local
// predicates are to be applied and if inNodeSet is provided,
// or the inLP are cacheable
if ((inputLP->isCacheable()) && (predIdSet == NULL) )
{
CANodeIdSet combinedSet = jbbc;
// get the nodeIdSet of the outer child, if not already given. This
// along with the present jbbc is used as a key in the cache
CANodeIdSet * inputNodeSet;
inputNodeSet = inputLP->getNodeSet();
// if inLP are cacheable these should have a nodeSet attached
CCMPASSERT(inputNodeSet != NULL);
if (inputNodeSet)
{
combinedSet.insert(*inputNodeSet);
// if estLogProp for all local predicates is required,
// then it might already exist in the cache
outputEstLogProp = getCachedStatistics(&combinedSet);
}
}
if (outputEstLogProp == NULL)
{
// 2. properties do not exist in the cache, so synthesize them.
//if specified by the user apply those predicates,
// else apply predicates in the original expr
NodeAnalysis * jbbcNode = jbbc.getNodeAnalysis();
TableAnalysis * tableAnalysis = jbbcNode->getTableAnalysis();
if (tableAnalysis && predIdSet)
{
TableDesc * tableDesc = tableAnalysis->getTableDesc();
const QualifiedName& qualName =
tableDesc->getNATable()->getTableName();
CorrName name(qualName, STMTHEAP);
Scan *scanExpr = new STMTHEAP Scan(name, tableDesc, REL_SCAN, STMTHEAP);
Cardinality rc = tableDesc->getNATable()->getEstRowCount();
const CardinalityHint* cardHint = tableDesc->getCardinalityHint();
if ( cardHint )
rc = (cardHint->getScanCardinality()).getValue();
if ( !cardHint && tableDesc->getNATable()->isHbaseTable() ) {
NATable* nt = (NATable*)(tableDesc->getNATable());
StatsList* statsList = nt->getColStats();
if ( statsList && statsList->entries() > 0 ) {
ColStatsSharedPtr cStatsPtr =
statsList->getSingleColumnColStats(0);
if ( cStatsPtr )
rc = (cStatsPtr->getRowcount()).getValue();
}
}
scanExpr->setBaseCardinality(MIN_ONE (rc));
GroupAttributes * gaExpr = new STMTHEAP GroupAttributes();
scanExpr->setSelectionPredicates(*predIdSet);
ValueIdSet requiredOutputs = jbbc.getNodeAnalysis()->\
getOriginalExpr()->getGroupAttr()->getCharacteristicOutputs();
gaExpr->setCharacteristicOutputs(requiredOutputs);
scanExpr->setGroupAttr(gaExpr);
gaExpr->setLogExprForSynthesis(scanExpr);
EstLogPropSharedPtr nonCacheableInLP(new (HISTHEAP) EstLogProp (*inputLP));
nonCacheableInLP->setCacheableFlag(FALSE);
scanExpr->synthLogProp();
outputEstLogProp = scanExpr->getGroupAttr()->outputLogProp(nonCacheableInLP);
}
else
{
NodeAnalysis * nodeAnalysis = jbbc.getNodeAnalysis();
RelExpr * relExpr = nodeAnalysis->getModifiedExpr();
if (relExpr == NULL)
relExpr = nodeAnalysis->getOriginalExpr();
// synthesize and cache estLogProp for the given inLP.
outputEstLogProp = relExpr->getGroupAttr()->outputLogProp(inputLP);
}
}
return outputEstLogProp;
} // getStatsForCANodeId