// This method forms the join expression for join on JBBC specified by jbbcId
// inputEstLogProp should not be cacheable
Join * AppliedStatMan::formJoinExprForJoinOnJBBC(
CANodeIdSet jbbSubset,
CANodeId jbbcId,
const ValueIdSet * jbbcLocalPreds,
const ValueIdSet * joinPreds,
const EstLogPropSharedPtr& inputEstLogProp,
const NABoolean cacheable)
{
NABoolean origInputIsCacheable = inputEstLogProp->isCacheable();
if(origInputIsCacheable)
{
inputEstLogProp->setCacheableFlag(FALSE);
CCMPASSERT("Expecting Non Cacheable Input");
}
RelExpr * jbbcExpr = getExprForCANodeId(jbbcId, inputEstLogProp, jbbcLocalPreds);
jbbcExpr->getGroupAttr()->outputLogProp(inputEstLogProp);
RelExpr * jbbSubsetExpr = jbbSubset.jbbcsToJBBSubset()->getPreferredJoin();
if(!jbbSubsetExpr)
if(jbbSubset.entries()==1)
if(!inputEstLogProp->isCacheable())
{
inputEstLogProp->setCacheableFlag(TRUE);
jbbSubsetExpr = getExprForCANodeId(jbbSubset.getFirst(), inputEstLogProp);
inputEstLogProp->setCacheableFlag(FALSE);
}
else
jbbSubsetExpr = getExprForCANodeId(jbbSubset.getFirst(), inputEstLogProp);
else
{
CCMPASSERT("No Subset expression, need at least one entry in set");
}
RelExpr * leftChildExpr = jbbSubsetExpr;
RelExpr * rightChildExpr = jbbcExpr;
GroupAttributes * galeft = jbbSubsetExpr->getGroupAttr();
GroupAttributes * garight = jbbcExpr->getGroupAttr();
// xxx
JBBC * jbbc = jbbcId.getNodeAnalysis()->getJBBC();
Join * jbbcParentJoin = jbbc->getOriginalParentJoin();
ValueIdSet leftOuterJoinFilterPreds;
Join * joinExpr = NULL;
if(jbbcParentJoin)
{
if(jbbcParentJoin->isSemiJoin())
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_SEMIJOIN, NULL);
if(jbbcParentJoin->isAntiSemiJoin())
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_ANTI_SEMIJOIN, NULL);
if(jbbcParentJoin->isLeftJoin())
{
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_LEFT_JOIN, NULL);
leftOuterJoinFilterPreds += jbbc->getLeftJoinFilterPreds();
}
if(joinExpr)
{
joinExpr->setJoinPred(jbbc->getPredsWithPredecessors());
joinExpr->nullInstantiatedOutput().insert(jbbc->nullInstantiatedOutput());
}
}
if(!joinExpr)
{
// now form a JoinExpr with these left and right children.
joinExpr = new STMTHEAP Join(leftChildExpr, rightChildExpr, REL_JOIN, NULL);
}
ValueIdSet selPredsAndLOJFilter = leftOuterJoinFilterPreds;
selPredsAndLOJFilter += (*joinPreds);
joinExpr->setSelectionPredicates(selPredsAndLOJFilter);
// set groupAttr of this Join expression
GroupAttributes * gaJoin = new STMTHEAP GroupAttributes();
// set required outputs of Join as sum of characteristic
// outputs of the left and the right children
ValueIdSet requiredOutputs;
requiredOutputs.addSet(getPotentialOutputs(jbbSubset));
requiredOutputs.addSet(getPotentialOutputs(jbbcId));
gaJoin->setCharacteristicOutputs(requiredOutputs);
// set JBBSubset for this group, if all estLogProps are cacheable.
// Else JBBSubset is NULL
CANodeIdSet combinedSet = jbbSubset;
combinedSet += jbbcId;
if (cacheable)
gaJoin->getGroupAnalysis()->setLocalJBBView(combinedSet.jbbcsToJBBSubset());
gaJoin->setMinChildEstRowCount(MINOF(garight->getMinChildEstRowCount(), galeft->getMinChildEstRowCount() ) );
// if there are some probes coming into the join
// then join type = tsj.
if ((inputEstLogProp->getResultCardinality() > 1) ||
(inputEstLogProp->getColStats().entries() > 1))
{
if (cacheable)
{
CANodeIdSet inputNodeSet = *(inputEstLogProp->getNodeSet());
gaJoin->setCharacteristicInputs(getPotentialOutputs(inputNodeSet));
}
}
joinExpr->setGroupAttr(gaJoin);
gaJoin->setLogExprForSynthesis(joinExpr);
joinExpr->synthLogProp();
inputEstLogProp->setCacheableFlag(origInputIsCacheable);
return joinExpr;
} // AppliedStatMan::formJoinExprForJoinOnJBBC
// ---------------------------------------------------------------------
// 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
// This method forms the join expression with the estLogProps.
Join * AppliedStatMan::formJoinExprWithEstLogProps(
const EstLogPropSharedPtr& leftEstLogProp,
const EstLogPropSharedPtr& rightEstLogProp,
const EstLogPropSharedPtr& inputEstLogProp,
const ValueIdSet * setOfPredicates,
const NABoolean cacheable,
JBBSubset * combinedJBBSubset)
{
// Form a join expression with these estLogProps.
// form the left child. Since the estLogProps of the left and the
// right children exist, these can be treated as Scan expressions
Scan * leftChildExpr = new STMTHEAP Scan();
GroupAttributes * galeft = new STMTHEAP GroupAttributes();
// set GroupAttr of the leftChild
galeft->inputLogPropList().insert(inputEstLogProp);
galeft->outputLogPropList().insert(leftEstLogProp);
CANodeIdSet * leftNodeSet = leftEstLogProp->getNodeSet();
CANodeId nodeId;
if (leftNodeSet)
{
if (leftNodeSet->entries() == 1)
{
nodeId = leftNodeSet->getFirst();
if(nodeId.getNodeAnalysis()->getTableAnalysis())
leftChildExpr->setTableAttributes(nodeId);
}
CostScalar minEstCard = leftNodeSet->getMinChildEstRowCount();
galeft->setMinChildEstRowCount(minEstCard);
}
leftChildExpr->setGroupAttr(galeft);
galeft->setLogExprForSynthesis(leftChildExpr);
// form the right child and set its groupAttr
Scan * rightChildExpr = new STMTHEAP Scan();
GroupAttributes * garight = new STMTHEAP GroupAttributes();
garight->inputLogPropList().insert(inputEstLogProp);
garight->outputLogPropList().insert(rightEstLogProp);
CANodeIdSet * rightNodeSet = rightEstLogProp->getNodeSet();
// xxx
JBBC * singleRightChild = NULL;
Join * singleRightChildParentJoin = NULL;
ValueIdSet leftOuterJoinFilterPreds;
if (rightNodeSet)
{
if (rightNodeSet->entries() == 1)
{
nodeId = rightNodeSet->getFirst();
if(nodeId.getNodeAnalysis()->getTableAnalysis())
rightChildExpr->setTableAttributes(nodeId);
if(nodeId.getNodeAnalysis()->getJBBC())
{
singleRightChild = nodeId.getNodeAnalysis()->getJBBC();
if(singleRightChild)
singleRightChildParentJoin = singleRightChild->getOriginalParentJoin();
}
}
CostScalar minEstCard = rightNodeSet->getMinChildEstRowCount();
garight->setMinChildEstRowCount(minEstCard);
}
rightChildExpr->setGroupAttr(garight);
garight->setLogExprForSynthesis(rightChildExpr);
Join * joinExpr = NULL;
if(singleRightChild &&
singleRightChildParentJoin)
{
if(singleRightChildParentJoin->isSemiJoin())
joinExpr = new STMTHEAP Join(leftChildExpr,
rightChildExpr,
REL_SEMIJOIN,
NULL);
if(singleRightChildParentJoin->isAntiSemiJoin())
joinExpr = new STMTHEAP Join(leftChildExpr,
rightChildExpr,
REL_ANTI_SEMIJOIN,
NULL);
if(singleRightChildParentJoin->isLeftJoin())
{
joinExpr = new STMTHEAP Join(leftChildExpr,
rightChildExpr,
REL_LEFT_JOIN,
NULL);
leftOuterJoinFilterPreds += singleRightChild->getLeftJoinFilterPreds();
}
if(joinExpr)
{
joinExpr->setJoinPred(singleRightChild->getPredsWithPredecessors());
joinExpr->nullInstantiatedOutput().insert(singleRightChild->
nullInstantiatedOutput());
}
}
if(!joinExpr)
{
// now form a JoinExpr with these left and right children.
joinExpr = new STMTHEAP Join(leftChildExpr, // left child
rightChildExpr, // right child
REL_JOIN, // join type
NULL); // join predicates
}
ValueIdSet selPredsAndLOJFilter = leftOuterJoinFilterPreds;
selPredsAndLOJFilter += (*setOfPredicates);
joinExpr->setSelectionPredicates(selPredsAndLOJFilter);
// set groupAttr of this Join expression
GroupAttributes * gaJoin = new STMTHEAP GroupAttributes();
// set required outputs of Join as sum of characteristic
// outputs of the left and the right children
ValueIdSet requiredOutputs;
if (leftNodeSet)
requiredOutputs.addSet(getPotentialOutputs(*(leftNodeSet)));
if (rightNodeSet)
requiredOutputs.addSet(getPotentialOutputs(*(rightNodeSet)));
gaJoin->setCharacteristicOutputs(requiredOutputs);
// set JBBSubset for this group, if all estLogProps are cacheable.
// Else JBBSubset is NULL
if (cacheable)
gaJoin->getGroupAnalysis()->setLocalJBBView(combinedJBBSubset);
gaJoin->setMinChildEstRowCount(MINOF(garight->getMinChildEstRowCount(), galeft->getMinChildEstRowCount() ) );
joinExpr->setGroupAttr(gaJoin);
// if there are some probes coming into the join
// then join type = tsj.
if ((inputEstLogProp->getResultCardinality() > 1) ||
(inputEstLogProp->getColStats().entries() > 1))
{
if (cacheable)
{
CANodeIdSet inputNodeSet = *(inputEstLogProp->getNodeSet());
gaJoin->setCharacteristicInputs(getPotentialOutputs(inputNodeSet));
}
}
joinExpr->setGroupAttr(gaJoin);
gaJoin->setLogExprForSynthesis(joinExpr);
return joinExpr;
} // AppliedStatMan::formJoinExprWithEstLogProps