ValueIdSet AppliedStatMan::getPotentialOutputs(
const CANodeIdSet & jbbcsNodeSet)
{
ValueIdSet potentialOutputs;
for (CANodeId jbbc = jbbcsNodeSet.init();
jbbcsNodeSet.next(jbbc);
jbbcsNodeSet.advance(jbbc))
{
if (NodeAnalysis * jbbcNodeAnalysis = jbbc.getNodeAnalysis())
{
ValueIdSet outputs;
const Join * jbbcParentJoin = jbbcNodeAnalysis->getJBBC()->
getOriginalParentJoin();
if((!jbbcParentJoin) ||
(jbbcParentJoin && jbbcParentJoin->isInnerNonSemiJoin()))
outputs = jbbcNodeAnalysis->getOriginalExpr()->\
getGroupAttr()->getCharacteristicOutputs();
else if (jbbcParentJoin->isLeftJoin())
outputs = jbbcParentJoin->nullInstantiatedOutput();
potentialOutputs.insert(outputs);
}
}
return potentialOutputs;
} // AppliedStatMan::getPotentialOutputs
// LCOV_EXCL_START - cnu
CostScalar
TableDesc::getBaseRowCntIfUniqueJoinCol(const ValueIdSet &joinedCols)
{
// get the joining columns for this table
ValueIdList userColumns;
// get All user columns for this table;
getUserColumnList(userColumns);
ValueIdSet userColumnSet(userColumns);
ValueIdSet joinedColsCopy(joinedCols);
ValueIdSet thisTableJoinCols = joinedColsCopy.intersect(userColumnSet);
if (thisTableJoinCols.isEmpty() )
return csMinusOne;
CostScalar baseRowCount = csMinusOne;
if (thisTableJoinCols.doColumnsConstituteUniqueIndex(this) )
baseRowCount = tableColStats()[0]->getColStats()->getRowcount();
return baseRowCount;
} // TableDesc::getBaseRowCntIfUniqueJoinCol
void
generateMarkedEntries(Generator *generator, ValueIdSet &marks)
{
for(ValueId vid = marks.init(); marks.next(vid); marks.advance(vid)) {
MapInfo *mapInfo =
generator->getMapInfoAsIs(vid);
if(mapInfo)
mapInfo->codeGenerated();
}
}
// PhysSequence::computeHistoryAttributes
//
// Helper function to compute the attribute for the history buffer based
// on the items projected from the child and the computed history items.
// Also, adds the attribute information the the map table.
//
void
PhysSequence::computeHistoryAttributes(Generator *generator,
MapTable *localMapTable,
Attributes **attrs,
const ValueIdSet &historyIds) const
{
// Get a local handle on some of the generator objects.
//
CollHeap *wHeap = generator->wHeap();
// Populate the attribute vector with the flattened list of sequence
// functions and/or sequence function arguments that must be in the
// history row. Add convert nodes for the items that are not sequence
// functions to force them to be moved into the history row.
//
if(NOT historyIds.isEmpty())
{
Int32 i = 0;
ValueId valId;
for (valId = historyIds.init();
historyIds.next(valId);
historyIds.advance(valId))
{
// If this is not a sequence function, then insert a convert
// node.
//
if(!valId.getItemExpr()->isASequenceFunction())
{
// Get a handle on the original expression and erase
// the value ID.
//
ItemExpr *origExpr = valId.getItemExpr();
origExpr->setValueId(NULL_VALUE_ID);
origExpr->markAsUnBound();
// Construct the cast expression with the original expression
// as the child -- must have undone the child value ID to
// avoid recursion later.
//
ItemExpr *castExpr = new(wHeap)
Cast(origExpr, &(valId.getType()));
// Replace the expression for the original value ID and the
// synthesize the types and value ID for the new expression.
//
valId.replaceItemExpr(castExpr);
castExpr->synthTypeAndValueId(TRUE);
}
attrs[i++] = (generator->addMapInfoToThis(localMapTable, valId, 0))->getAttr();
}
}
} // PhysSequence::computeHistoryAttributes
NABoolean TableDesc::hasIdentityColumnInClusteringKey() const
{
ValueIdSet pKeyColumns = clusteringIndex_->getIndexKey();
NAColumn * column = NULL;
for(ValueId id = pKeyColumns.init(); pKeyColumns.next(id);
pKeyColumns.advance(id))
{
column = id.getNAColumn();
if (column && column->isIdentityColumn())
return TRUE;
}
return FALSE;
}
ValueIdSet TableDesc::getLocalPreds()
{
ValueIdSet localPreds;
localPreds.clear();
// We can get this information from TableAnalysis
const TableAnalysis * tableAnalysis = getTableAnalysis();
// if no tableAnalysis exists, return FALSE
if(tableAnalysis)
localPreds = tableAnalysis->getLocalPreds();
return localPreds;
}
void NormWA::locateVEGRegionAndMarkToBeMergedRecursively(const ValueId & vid)
{
VEGRegion* toBeMergedRegion = locateVEGRegionAndMarkToBeMerged(vid);
if (toBeMergedRegion)
{
ValueIdSet nullInstValues;
toBeMergedRegion->gatherInstantiateNullMembers(nullInstValues);
for (ValueId exprId = nullInstValues.init();
nullInstValues.next(exprId);
nullInstValues.advance(exprId))
{
locateVEGRegionAndMarkToBeMergedRecursively(exprId);
}
}
}
void MvQueryRewriteHandler::dumpAnalysisToFile(QueryAnalysis* qa, RelExpr* expr)
{
// Dump the QueryAnalysis data to a file.
NAString analysisFileName = fileNamePrefix_ + ".analysis";
NAString str;
expr->unparse(str, OPTIMIZER_PHASE, MVINFO_FORMAT);
str += "\n";
str += qa->getText();
// Add in some stuff to look at join predicates for the JBBCs.
str += "Join Predicates\n";
str += "===============";
char buffer[20];
ARRAY(JBB*) jbbs = qa->getJBBs();
for (CollIndex jbbInx = 0; jbbInx < jbbs.entries(); jbbInx++)
{
JBB* jbb = jbbs[jbbInx];
str_itoa(jbbInx, buffer);
((str += "\nJBB #") += NAString(buffer)) += ":\n";
CANodeIdSet jbbcs = jbb->getJBBCs();
for (CANodeId jbbcId=jbbcs.init(); jbbcs.next(jbbcId); jbbcs.advance(jbbcId) )
{
str_itoa(jbbcId, buffer);
((str += "\nJBBC with CANodeId ") += NAString(buffer)) += ":\n";
ValueIdSet joinPreds = jbbcId.getNodeAnalysis()->getJBBC()->getJoinPreds();
str += valueIdSetGetText(joinPreds);
if (joinPreds.entries() > 0)
{
str.append("\n(value ids of predicates are ");
NABoolean first = true;
for (ValueId jpVid=joinPreds.init(); joinPreds.next(jpVid); joinPreds.advance(jpVid))
{
if (first)
first = FALSE;
else
str.append(", ");
str_itoa(jpVid, buffer);
str.append(buffer);
}
str.append(")\n");
}
}
str += '\n';
}
dumpToFile(analysisFileName.data(), str.data());
} // dumpAnalysisToFile()
void
IndexDesc::getNonKeyColumnSet(ValueIdSet& nonKeyColumnSet) const
{
const ValueIdList
&indexColumns = getIndexColumns(),
&keyColumns = getIndexKey();
// clean up input:
nonKeyColumnSet.clear();
// Add all index columns
CollIndex i = 0;
for (i=0;
i < indexColumns.entries();
i++)
{
nonKeyColumnSet.insert(indexColumns[i]);
}
// And remove all key columns:
for (i=0;
i < keyColumns.entries();
i++)
{
nonKeyColumnSet.remove(keyColumns[i]);
// if this is a secondary index, the base column
// which is part of the index,
// may also be present, remove it:
const ItemExpr *colPtr = keyColumns[i].getItemExpr();
if (colPtr->getOperatorType()
==
ITM_INDEXCOLUMN)
{
const ValueId & colDef = ((IndexColumn *)(colPtr))->getDefinition();
nonKeyColumnSet.remove(colDef);
}
}
} // IndexDesc::getNonKeyColumnSet(ValueIdSet& nonKeyColumnSet) const
void MultiJoin::getPotentialOutputValues(ValueIdSet & outputValues) const
{
outputValues.clear();
CANodeIdSet jbbcs = jbbSubset_.getJBBCs();
Int32 arity = getArity();
for (Lng32 i = 0; i < arity; i++)
{
JBBC * jbbci =
child(i)->getGroupAnalysis()->getNodeAnalysis()->getJBBC();
if(jbbci->parentIsLeftJoin())
outputValues.insertList(jbbci->nullInstantiatedOutput());
else
// Default implementation is good enough for innerNonSemi multi join
outputValues += child(i).getGroupAttr()->getCharacteristicOutputs();
}
} // MultiJoin::getPotentialOutputValues()
void PhysSequence::seperateReadAndReturnItems(
//ValueIdSet & readPhaseSet,
//ValueIdSet & returnPhaseSet,
CollHeap *wHeap)
{
ValueIdSet outputFromChild = child(0)->getGroupAttr()->getCharacteristicOutputs();
ValueIdSet seqFuncs = sequenceFunctions();
for(ValueId valId = seqFuncs.init();
seqFuncs.next(valId);
seqFuncs.advance(valId))
{
computeReadNReturnItems(valId,
valId,
//sequenceFunctions(),
//returnSeqFunctions(),
outputFromChild,
wHeap);
}
}
void HbaseSearchSpec::addColumnNames(const ValueIdSet& vs)
{
// TEMP TEMP. Not all needed column names are being set up.
// for now, return without populating result.
// that will cause all columns to be retrieved.
//return;
for (ValueId vid = vs.init(); vs.next(vid); vs.advance(vid)) {
ItemExpr* ie = vid.getItemExpr();
NAString colName;
if ( ie->getOperatorType() == ITM_BASECOLUMN ) {
colName = ((BaseColumn*)ie)->getColName();
} else
if ( ie->getOperatorType() == ITM_INDEXCOLUMN ) {
colName = ((IndexColumn*)ie)->getNAColumn()->getIndexColName();
}
if (NOT colNames_.contains(colName))
colNames_.insert(colName);
}
}
// -----------------------------------------------------------------------
// MultJoin::recomputeOuterReferences()
// -----------------------------------------------------------------------
void MultiJoin::recomputeOuterReferences()
{
// ---------------------------------------------------------------------
// Delete all those input values that are no longer referenced on
// this operator.
// ---------------------------------------------------------------------
if (NOT getGroupAttr()->getCharacteristicInputs().isEmpty())
{
ValueIdSet outerRefs = getGroupAttr()->getCharacteristicInputs();
// Weed out those expressions not needed by my selectionPred and joinPred
// xxx instead of taking this from getLocalJoinPreds, should I take it
// from MultiJoin selectionPred??? refer to getLocalJoinPreds definition
// and consider preds that referencing inputs!!!
ValueIdSet exprSet = jbbSubset_.getLocalJoinPreds(); // from JbbSubsetAnalysis
// Need to include LocalDependentPreds later when supported. Ok now for inner MultiJoins
exprSet.weedOutUnreferenced(outerRefs);
// Add back those expressiones needed by my children
Int32 arity = getArity();
// outputs produced by JBBCs in this MultiJoin
ValueIdSet jbbcOutputs;
for (Int32 i = 0; i < arity; i++)
{
outerRefs += child(i)->getGroupAttr()->getCharacteristicInputs();
jbbcOutputs += child(i)->getGroupAttr()->getCharacteristicOutputs();
// these inputs are provided by jbbcs in this MultiJoin
}
// account for TSJs i.e. values flowing from
// one jbbc to another within this MultiJoin
outerRefs -= jbbcOutputs;
getGroupAttr()->setCharacteristicInputs(outerRefs);
}
return;
} // MultiJoin::recomputeOuterReferences()
// Is there any column which has a local predicates and no stats
NABoolean TableDesc::isAnyHistWithPredsFakeOrSmallSample(const ValueIdSet &localPreds)
{
// if there are no local predicates return FALSE;
if (localPreds.isEmpty())
return FALSE;
const ColStatDescList & colStatsList = getTableColStats();
// for each predicate, check to see if stats exist
for (ValueId id = localPreds.init();
localPreds.next(id);
localPreds.advance(id))
{
ColStatsSharedPtr colStats = colStatsList.getColStatsPtrForPredicate(id);
if (colStats == NULL)
return FALSE;
if (colStats->isOrigFakeHist() || colStats->isSmallSampleHistogram())
return TRUE;
}
return FALSE;
}
// this method sets the primary key columns. It goes through all the columns
// of the table, and collects the columns which are marked as primary keys
void TableDesc::setPrimaryKeyColumns()
{
ValueIdSet primaryColumns;
for ( CollIndex j = 0 ; j < colList_.entries() ; j++ )
{
ValueId valId = colList_[j];
NAColumn *column = valId.getNAColumn();
if ( column->isPrimaryKey() )
{
primaryColumns.insert(valId) ;
// mark column as referenced for histogram, as we may need its histogram
// during plan generation
if ((column->isUserColumn() || column->isSaltColumn() ) &&
(column->getNATable()->getSpecialType() == ExtendedQualName::NORMAL_TABLE) )
column->setReferencedForMultiIntHist();
}
}
primaryKeyColumns_ = primaryColumns;
}
// TableDesc::isKeyIndex()
// Parameter is an secondary index on the table. Table checks to see
// if the keys of the secondary index is built using the primary key
// of the table. If it is return true otherwise false.
NABoolean TableDesc::isKeyIndex(const IndexDesc * idesc) const
{
ValueIdSet pKeyColumns = clusteringIndex_->getIndexKey();
ValueIdSet indexColumns = idesc->getIndexKey();
ValueIdSet basePKeys=pKeyColumns.convertToBaseIds();
for(ValueId id = indexColumns.init(); indexColumns.next(id);
indexColumns.advance(id))
{
ValueId baseId = ((BaseColumn *)(((IndexColumn *)id.getItemExpr())->
getDefinition().getItemExpr()))->getValueId();
if(NOT basePKeys.contains(baseId))
{
return FALSE;
}
}
return TRUE;
}
void PhysSequence::computeReadNReturnItems( ValueId topSeqVid,
ValueId vid,
const ValueIdSet &outputFromChild,
CollHeap *wHeap)
{
ItemExpr * itmExpr = vid.getItemExpr();
if (outputFromChild.contains(vid))
{
return;
}
//test if itm_minus and then if negative offset ....
if ( itmExpr->getOperatorType() == ITM_OFFSET &&
((ItmSeqOffset *)itmExpr)->getOffsetConstantValue() < 0)
{
readSeqFunctions() -= topSeqVid;
returnSeqFunctions() += topSeqVid;
readSeqFunctions() += itmExpr->child(0)->castToItemExpr()->getValueId();
return;
}
if (itmExpr->getOperatorType() == ITM_MINUS)
{
ItemExpr * chld0 = itmExpr->child(0)->castToItemExpr();
if ( chld0->getOperatorType() == ITM_OFFSET &&
((ItmSeqOffset *)chld0)->getOffsetConstantValue() <0)
{
readSeqFunctions() -= topSeqVid;
returnSeqFunctions() += topSeqVid;
readSeqFunctions() += chld0->child(0)->castToItemExpr()->getValueId();
ItemExpr * chld1 = itmExpr->child(1)->castToItemExpr();
if (chld1->getOperatorType() == ITM_OFFSET &&
((ItmSeqOffset *)chld1)->getOffsetConstantValue() < 0)
{
readSeqFunctions() += chld1->child(0)->castToItemExpr()->getValueId();
}
else
{
readSeqFunctions() += chld1->getValueId();
}
return;
}
}
if (itmExpr->getOperatorType() == ITM_OLAP_MIN ||
itmExpr->getOperatorType() == ITM_OLAP_MAX)
{
ItmSeqOlapFunction * olap = (ItmSeqOlapFunction *)itmExpr;
if (olap->getframeEnd()>0)
{
readSeqFunctions() -= topSeqVid;
returnSeqFunctions() += topSeqVid;
ItemExpr *newChild = new(wHeap) Convert (itmExpr->child(0)->castToItemExpr());
newChild->synthTypeAndValueId(TRUE);
itmExpr->child(0) = newChild;
readSeqFunctions() += newChild->getValueId();
return;
}
}
if (itmExpr->getOperatorType() == ITM_SCALAR_MIN ||
itmExpr->getOperatorType() == ITM_SCALAR_MAX)
{
ItemExpr * chld0 = itmExpr->child(0)->castToItemExpr();
ItemExpr * chld1 = itmExpr->child(1)->castToItemExpr();
if ((chld0->getOperatorType() == ITM_OLAP_MIN && chld1->getOperatorType() == ITM_OLAP_MIN )||
(chld0->getOperatorType() == ITM_OLAP_MAX && chld1->getOperatorType() == ITM_OLAP_MAX ))
{
ItmSeqOlapFunction * olap0 = (ItmSeqOlapFunction *)chld0;
ItmSeqOlapFunction * olap1 = (ItmSeqOlapFunction *)chld1;
if ( olap1->getframeEnd()>0)
{
CMPASSERT(olap0->getframeEnd()==0);
readSeqFunctions() -= topSeqVid;
returnSeqFunctions() += topSeqVid;
readSeqFunctions() += olap0->getValueId();
ItemExpr *newChild = new(wHeap) Convert (olap1->child(0)->castToItemExpr());
newChild->synthTypeAndValueId(TRUE);
olap1->child(0) = newChild;
readSeqFunctions() += newChild->getValueId();
}
else
{
CMPASSERT(olap1->getframeEnd()==0);
readSeqFunctions() -= topSeqVid;
returnSeqFunctions() += topSeqVid;
readSeqFunctions() += olap1->getValueId();
ItemExpr *newChild = new(wHeap) Convert (olap0->child(0)->castToItemExpr());
newChild->synthTypeAndValueId(TRUE);
olap0->child(0) = newChild;
readSeqFunctions() += newChild->getValueId();
}
return;
}
}
for (Int32 i= 0 ; i < itmExpr->getArity(); i++)
{
ItemExpr * chld= itmExpr->child(i);
computeReadNReturnItems(topSeqVid,
chld->getValueId(),
outputFromChild,
wHeap);
}
}//void PhysSequence::computeReadNReturnItems(ItemExpr * other)
/********************************************************************
* Input: Selection predicates for the scan node, boolean indicating if
* it is a indexOnlyIndex, reference parameter that will indicate if
* IndexJoin is viable or not, GroupAttributes for the group and characteristic
* inputs
* Output: MdamFlag indicating if the index key access is good enough for
* MDAM access (if a index does not have good MDAM access we have to
* scan the whole index because single subset also will not have any
* keys to apply)
* IndexJoin flag indicating if index join cost would exceed base table
* access or not.
********************************************************************/
MdamFlags IndexDesc::pruneMdam(const ValueIdSet& preds,
NABoolean indexOnlyIndex,
IndexJoinSelectivityEnum&
selectivityEnum /* out*/ ,
const GroupAttributes * groupAttr,
const ValueIdSet * inputValues) const
{
CollIndex numEmptyColumns=0;
CostScalar numSkips = csOne;
ValueIdSet emptyColumns;
ValueId vid;
if(indexOnlyIndex)
selectivityEnum = INDEX_ONLY_INDEX;
else
selectivityEnum = INDEX_JOIN_VIABLE;
if(preds.isEmpty()) return MDAM_OFF;
//calculate how many key columns don't have any predicates
for(CollIndex i=0;i<indexKey_.entries();i++)
{
if(preds.referencesTheGivenValue(indexKey_[i],vid))
break;
else
numEmptyColumns++;
}
//if we don't have any empty columns or we don't have to evaluate if index
//join is promising or not then just return
if(numEmptyColumns>=1 OR NOT indexOnlyIndex)
{
IndexDescHistograms ixHistogram(*this,
(indexOnlyIndex?numEmptyColumns:indexKey_.entries()));
NABoolean multiColUecAvail = ixHistogram.isMultiColUecInfoAvail();
ColumnOrderList keyPredsByCol(indexKey_);
for(CollIndex j=0;j<numEmptyColumns;j++)
{
emptyColumns.insert(indexKey_[j]);
if(j==0 OR multiColUecAvail == FALSE)
{
//no MCUec so just multiply the empty columns UEC count to
//calculate MDAM skips
numSkips *=(ixHistogram.getColStatsForColumn(indexKey_[j])).
getTotalUec().getCeiling();
}
else // otherwise try to use MCUec
{
NABoolean uecFound = FALSE;
CostScalar correctUec = csOne;
CostScalar combinedUECCount = csOne;
// first let's see if there is multiColUec count for the skipped columns
// so far. If there is that will be number of skips. If there isn't then
// get the best estimate of UEC count for the current column using MCUec
// if possible otherwise just using single column histograms.
combinedUECCount = ixHistogram.getUecCountForColumns(emptyColumns);
if(combinedUECCount >0)
{
numSkips = combinedUECCount;
}
else
{
uecFound = ixHistogram.estimateUecUsingMultiColUec(keyPredsByCol,j,correctUec);
if(uecFound==TRUE)
{
numSkips *= correctUec;
}
else
{
numSkips *=(ixHistogram.getColStatsForColumn(indexKey_[j])).
getTotalUec().getCeiling();
}
}
}
}
CostScalar rowCount = ixHistogram.getRowCount();
CostScalar numIndexBlocks = rowCount /getEstimatedRecordsPerBlock();
CostScalar numProbes = csOne;
CostScalar numBaseTableBlocks = csOne;
CostScalar inputProbes = csOne;
// Pass any selectivity hint provided by the user
const SelectivityHint * selHint = tableDesc_->getSelectivityHint();
const CardinalityHint * cardHint = tableDesc_->getCardinalityHint();
// If it is an index join then compute the number probes into the base
// table. If the alternate index is not selective enough, we will have
// lots of them making the index quite expensive.
if(NOT indexOnlyIndex)
{
if((groupAttr->getInputLogPropList()).entries() >0)
{
//if there are incoming probes to the index. i.e. if the index join
//is under another nested join or TSJ then compute result for all
//probes. We are using the initial inputEstLogProp to compute the
//resulting cardinality. It is possible that for the same group and
//different inputEstLogProp would provide less row count per probe.
//So in FileScanRule::nextSubstitute() we make sure that the context
//inputEstLogProp is in the error range of this inputEstLogProp.
// Ex. select * from lineitem, customer, nation
// where l_custkey < c_custkey and c_custkey = n_nationkey;
//Now if we were evaluating lineitem indexes where the outer was customer
//we would want to exclude alternate index on custkey whereas if nation got
//pushed below customer then range of values would be fewer and max value
//being less would make alternate index on custkey quite attractive.
ixHistogram.
applyPredicatesWhenMultipleProbes(preds,
*((groupAttr->getInputLogPropList())[0]),
*inputValues,
TRUE,
selHint,
cardHint,
NULL,
REL_SCAN);
inputProbes = MIN_ONE((groupAttr->getInputLogPropList())[0]->getResultCardinality());
}
else
{
RelExpr * dummyExpr = new (STMTHEAP) RelExpr(ITM_FIRST_ITEM_OP,
NULL,
NULL,
STMTHEAP);
ixHistogram.applyPredicates(preds, *dummyExpr, selHint, cardHint, REL_SCAN);
}
numProbes = ixHistogram.getRowCount();
numBaseTableBlocks = rowCount / tableDesc_->getClusteringIndex()->
getEstimatedRecordsPerBlock();
double readAhead = CURRSTMT_OPTDEFAULTS->readAheadMaxBlocks();
// although we compute cardinality from the index for all probes we
// do the comparison for per probe. The assumption is that per probe
// the upper bound of cost is scanning the whole base table.
if(numProbes/inputProbes + MINOF((numIndexBlocks / readAhead),numSkips)
> (numBaseTableBlocks/readAhead))
{
selectivityEnum = EXCEEDS_BT_SCAN;
}
}
//Does the number of skips exceed the cost of scanning the index.
if((indexOnlyIndex AND numSkips <=
(numIndexBlocks * CURRSTMT_OPTDEFAULTS->mdamSelectionDefault())) OR
(NOT indexOnlyIndex AND numSkips + numProbes/inputProbes <=
(numBaseTableBlocks * CURRSTMT_OPTDEFAULTS->mdamSelectionDefault())))
return MDAM_ON;
}
else
return MDAM_ON;
return MDAM_OFF;
}
// getHistoryAttributes
//
// Helper function that traverses the set of root sequence functions
// supplied by the compiler and constructs the set of all of the
// attributes that must be materialized in the history row.
//
void PhysSequence::getHistoryAttributes(const ValueIdSet &sequenceFunctions,
const ValueIdSet &outputFromChild,
ValueIdSet &historyAttributes,
NABoolean addConvNodes,
CollHeap *wHeap,
ValueIdMap *origAttributes) const
{
if(addConvNodes && !origAttributes) {
origAttributes = new (wHeap) ValueIdMap();
}
ValueIdSet children;
for(ValueId valId = sequenceFunctions.init();
sequenceFunctions.next(valId);
sequenceFunctions.advance(valId)) {
if(valId.getItemExpr()->isASequenceFunction()) {
ItemExpr *itmExpr = valId.getItemExpr();
switch(itmExpr->getOperatorType())
{
// The child needs to be in the history row.
//
case ITM_OFFSET:
case ITM_ROWS_SINCE:
case ITM_THIS:
case ITM_NOT_THIS:
// If the child needs to be in the history buffer, then
// add a Convert node to force the value to be moved to the
// history buffer.
if (addConvNodes)
{
itmExpr->child(0) =
addConvNode(itmExpr->child(0), origAttributes, wHeap);
}
historyAttributes += itmExpr->child(0)->getValueId();
break;
// The sequence function needs to be in the history row.
//
case ITM_RUNNING_SUM:
case ITM_RUNNING_COUNT:
case ITM_RUNNING_MIN:
case ITM_RUNNING_MAX:
case ITM_LAST_NOT_NULL:
historyAttributes += itmExpr->getValueId();
break;
/*
// after PhysSequence precode gen OLAP sum and count are already transform,ed into running
// this is used during optimization phase--
case ITM_OLAP_SUM:
case ITM_OLAP_COUNT:
case ITM_OLAP_RANK:
case ITM_OLAP_DRANK:
if (addConvNodes)
{
itmExpr->child(0) =
addConvNode(itmExpr->child(0), origAttributes, wHeap);
}
historyAttributes += itmExpr->child(0)->getValueId();
//historyAttributes += itmExpr->getValueId();
break;
*/
// The child and sequence function need to be in the history row.
//
case ITM_OLAP_MIN:
case ITM_OLAP_MAX:
case ITM_MOVING_MIN:
case ITM_MOVING_MAX:
// If the child needs to be in the history buffer, then
// add a Convert node to force the value to be moved to the
// history buffer.
if (addConvNodes)
{
itmExpr->child(0) =
addConvNode(itmExpr->child(0), origAttributes, wHeap);
}
historyAttributes += itmExpr->child(0)->getValueId();
historyAttributes += itmExpr->getValueId();
break;
case ITM_RUNNING_CHANGE:
if (itmExpr->child(0)->getOperatorType() == ITM_ITEM_LIST)
{
// child is a multi-valued expression
//
ExprValueId treePtr = itmExpr->child(0);
ItemExprTreeAsList changeValues(&treePtr,
ITM_ITEM_LIST,
RIGHT_LINEAR_TREE);
CollIndex nc = changeValues.entries();
ItemExpr *newChild = NULL;
if(addConvNodes) {
newChild = addConvNode(changeValues[nc-1], origAttributes, wHeap);
historyAttributes += newChild->getValueId();
} else {
historyAttributes += changeValues[nc-1]->getValueId();
}
// add each item in the list
//
for (CollIndex i = nc; i > 0; i--)
{
if(addConvNodes) {
ItemExpr *conv
= addConvNode(changeValues[i-1], origAttributes, wHeap);
newChild = new(wHeap) ItemList(conv, newChild);
newChild->synthTypeAndValueId(TRUE);
historyAttributes += conv->getValueId();
} else {
historyAttributes += changeValues[i-1]->getValueId();
}
}
if(addConvNodes) {
itmExpr->child(0) = newChild;
}
}
else
{
// If the child needs to be in the history buffer, then
// add a Convert node to force the value to be moved to the
// history buffer.
if (addConvNodes)
{
itmExpr->child(0) =
addConvNode(itmExpr->child(0), origAttributes, wHeap);
}
historyAttributes += itmExpr->child(0)->getValueId();
}
historyAttributes += itmExpr->getValueId();
break;
default:
CMPASSERT(0);
}
}
// Gather all the children, and if not empty, recurse down to the
// next level of the tree.
//
for(Lng32 i = 0; i < valId.getItemExpr()->getArity(); i++)
{
if (!outputFromChild.contains(valId.getItemExpr()->child(i)->getValueId()))
//!valId.getItemExpr()->child(i)->nodeIsPreCodeGenned())
{
children += valId.getItemExpr()->child(i)->getValueId();
}
}
}
if (NOT children.isEmpty())
{
getHistoryAttributes( children,
outputFromChild,
historyAttributes,
addConvNodes,
wHeap,
origAttributes);
}
} // PhysSequence::getHistoryAttributes
// ------------------------------------------------------------------------------
// create my colStats based on my child's output, by converting the columns to
// that of mine
// ------------------------------------------------------------------------------
void EstLogProp::mapOutputsForUpdate(const GenericUpdate & updateExpr,
const ValueIdMap & updateSelectValueIdMap)
{
TableDesc * updateTable = updateExpr.getTableDesc();
for ( CollIndex i = 0; i < colStats().entries(); i++ )
{
ColStatDescSharedPtr colStatPtr = (colStats())[i];
const ValueId columnId = colStatPtr->getVEGColumn();
ValueId updateColVEGOutputId;
updateSelectValueIdMap.mapValueIdUp(updateColVEGOutputId, columnId);
ValueId updateBaseColumnId;
if (updateColVEGOutputId != columnId)
{
updateBaseColumnId = updateColVEGOutputId;
ValueIdSet baseColumns;
updateColVEGOutputId.getItemExpr()->findAll( ITM_BASECOLUMN, baseColumns, TRUE, TRUE );
// from all the columns extracted, get the one for Insert table
TableDesc * thisTable = NULL;
for (ValueId column = baseColumns.init(); baseColumns.next(column);
baseColumns.advance(column) )
{
ItemExpr * columnExpr = column.getItemExpr();
thisTable = ((BaseColumn *)columnExpr)->getTableDesc();
if (thisTable == updateTable)
{
// set my column as the base column
updateBaseColumnId = column;
break;
}
}
ColStatsSharedPtr inColStats = colStatPtr->getColStats();
ColStatsSharedPtr colStatsForUpdate(new (STMTHEAP) ColStats (*inColStats,STMTHEAP));
colStatsForUpdate->setStatColumn(updateBaseColumnId.getNAColumn());
// use this ColStat to generate new ColStat corresponding to the char output
// of the Update expression
ColStatDescSharedPtr colStatDescForUpdate(new (STMTHEAP) ColStatDesc(colStatsForUpdate,
updateBaseColumnId, // ValueId of the column that will be used
// as a column name, VEG and mergeStats
STMTHEAP), STMTHEAP);
colStatDescForUpdate->VEGColumn() = updateColVEGOutputId;
colStatDescForUpdate->mergeState().clear() ;
colStatDescForUpdate->mergeState().insert(updateBaseColumnId);
// Remove the old colStat and insert this colStat into the result colStatDescList
colStats().removeAt( i );
colStats().insertDeepCopyAt(i, colStatDescForUpdate, // colStats to be copied
1, // scale
FALSE);
}
}
}
// ---------------------------------------------------------------------
// 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
Join* MultiJoin::splitSubset(const JBBSubset & leftSet,
const JBBSubset & rightSet,
NABoolean reUseMJ) const
{
// At this point assert that none of the subsets has a group by member
CMPASSERT ( (jbbSubset_.getGB() == NULL_CA_ID) &&
(leftSet.getGB() == NULL_CA_ID) &&
(rightSet.getGB() == NULL_CA_ID) );
#ifndef NDEBUG
// assert that left + right == subSet_
// and left intersect right = phi
CANodeIdSet unionSet(leftSet.getJBBCs());
CANodeIdSet intersectSet(leftSet.getJBBCs());
unionSet += rightSet.getJBBCs();
intersectSet.intersectSet(rightSet.getJBBCs());
CMPASSERT ( (unionSet == jbbSubset_.getJBBCs()) &&
(intersectSet.entries() == 0 ));
#endif
// Note: Joins including left, semi, anti semi are only created when
// a single jbbc connected via one of them is split as a single right
// child. InnerNonSemi joins can be created for any split i.e. any
// number of jbbcs on the left and the right of the join, but special
// joins (i.e. left, semi and anti semi joins) are only created when
// there is a single right child i.e. the rightSet contains only one
// jbbc that is connected via a special join. This is enforced as follows
//
// * The leftSet should be legal: This means that for every jbbc in the
// leftSet any predecessor jbbcs should be present in the leftSet.
// * The rightSet is either a single jbbc or if the rightSet has more
// than one jbbc then it should be legal, note that a jbbc connected
// via a special join is not a legal set by itself but we allow
// creation of special joins assuming the predecessors are present
// in the leftSet.
//
// An implicit assumption here is that 'this' MultiJoin is legal, which
// is fair since apart from the top level multijoin, rest of the multijoins
// are produced by splitting the top level multijoin. This method should
// not produce illegal multijoins, since we check both leftSet and rightSet
// for legality. Only time we don't check for legality is when the rightChild
// is a single jbbc, and a single jbbc does not result in a multijoin.
if(!leftSet.legal())
return NULL;
if((rightSet.getJBBCs().entries() > 1) && (!rightSet.legal()))
return NULL;
// everything here goes to statement heap
CollHeap* outHeap = CmpCommon::statementHeap();
RelExpr* child0 = generateSubsetExpr(leftSet, reUseMJ);
RelExpr* child1 = generateSubsetExpr(rightSet, reUseMJ);
// Flag to remember to pass on the derivedFromRoutineJoin flag if needed.
NABoolean derivedFromRoutineJoin(FALSE);
// now form a JoinExpr with these left and right children.
Join * result = NULL;
// if the rightSet is a single jbbc, then it could be connected via
// a special join. In such a case we have to create the appropriate
// join operator
if(rightSet.getJBBCs().entries() == 1){
JBBC * rightChild = rightSet.getJBBCs().getFirst().getNodeAnalysis()
->getJBBC();
Join * rightChildParentJoin = rightChild->getOriginalParentJoin();
// If rightChildParentJoin is NULL, then the child is the left
// child of the left most join and is considered to be connected
// via a InnerNonSemi join.
if(rightChildParentJoin)
{
if(rightChildParentJoin->derivedFromRoutineJoin())
derivedFromRoutineJoin = TRUE;
if(rightChildParentJoin->isSemiJoin())
result = new (outHeap) Join(child0, child1, REL_SEMIJOIN, NULL);
if(rightChildParentJoin->isAntiSemiJoin())
result = new (outHeap) Join(child0, child1, REL_ANTI_SEMIJOIN, NULL);
if(rightChildParentJoin->isLeftJoin())
{
// left joins can have filter preds, i.e. predicates that
// are applied as filters after applying the join predicate.
// We need to set them here.
result = new (outHeap) Join(child0, child1, REL_LEFT_JOIN, NULL);
result->setSelectionPredicates(rightChild->getLeftJoinFilterPreds());
}
if(rightChildParentJoin->isRoutineJoin())
{
derivedFromRoutineJoin = TRUE;
result = new (outHeap) Join(child0, child1, REL_ROUTINE_JOIN, NULL);
ValueIdSet routineJoinFilterPreds = rightChild->getRoutineJoinFilterPreds();
ValueIdSet predsToAddToRoutineJoin;
// add covered filter preds
for (ValueId filterPred= routineJoinFilterPreds.init();
routineJoinFilterPreds.next(filterPred);
routineJoinFilterPreds.advance(filterPred) )
{
if(jbbSubset_.coversExpr(filterPred))
predsToAddToRoutineJoin += filterPred;
}
result->setSelectionPredicates(predsToAddToRoutineJoin);
}
if(result)
{
// set the join predicate for special joins, note predicates
// for regular InnerNonSemi joins are set as selection predicates
// in the join relexpr.
result->setJoinPred(rightChild->getPredsWithPredecessors());
result->nullInstantiatedOutput().insert(rightChild->
nullInstantiatedOutput());
}
}
}
// The join to be created is a regular InnerNonSemi join
if (!result)
result = new (outHeap) Join(child0, child1, REL_JOIN, NULL);
// Make sure we carry the derivedFromRoutineJoin flag with us
if (derivedFromRoutineJoin)
result->setDerivedFromRoutineJoin();
// Share my groupAttr with result
result->setGroupAttr(getGroupAttr());
// get inner join predicates
ValueIdSet selPreds = rightSet.joinPredsWithOther(leftSet);
// get left join filter preds if any
selPreds += result->getSelectionPredicates();
result->setSelectionPredicates(selPreds);
result->findEquiJoinPredicates();
// May be I could save a little if i pushdown only to the child(ren)
// that are not already JBBCs, i.e. multijoins
result->pushdownCoveredExpr
(result->getGroupAttr()->getCharacteristicOutputs(),
result->getGroupAttr()->getCharacteristicInputs(),
result->selectionPred());
// We used CutOp as children, to avoid pushing predicates to JBBCs.
// Now put the actual expression back in case the child is a JBBCs
if(leftSet.getJBBCs().entries() == 1)
result->setChild(0, getJBBCRelExpr(leftSet.getJBBCs().getFirst()));
// We used CutOp as children, to avoid pushing predicates to JBBCs.
// Now put the actual expression back in case the child is a JBBCs
if(rightSet.getJBBCs().entries() == 1)
result->setChild(1, getJBBCRelExpr(rightSet.getJBBCs().getFirst()));
// Temp fixup. We need to take the selectionPred out of MultiJoin
// for now to prevent that pushed expr from being there. selectionPred
// is not being used now in MultiJoin xxx.
if (leftSet.getJBBCs().entries() > 1)
result->child(0)->selectionPred().clear();
if (rightSet.getJBBCs().entries() > 1)
result->child(1)->selectionPred().clear();
return result;
}
// computeHistoryBuffer
//
// Helper function that traverses the set of root sequence functions
// supplied by the compiler and dynamically determines the size
// of the history buffer.
//
void PhysSequence::computeHistoryRows(const ValueIdSet &sequenceFunctions,//historyIds
Lng32 &computedHistoryRows,
Lng32 &unableToCalculate,
NABoolean &unboundedFollowing,
Lng32 &minFollowingRows,
const ValueIdSet &outputFromChild)
{
ValueIdSet children;
ValueIdSet historyAttributes;
Lng32 value = 0;
for(ValueId valId = sequenceFunctions.init();
sequenceFunctions.next(valId);
sequenceFunctions.advance(valId))
{
if(valId.getItemExpr()->isASequenceFunction())
{
ItemExpr *itmExpr = valId.getItemExpr();
switch(itmExpr->getOperatorType())
{
// THIS and NOT THIS are not dynamically computed
//
case ITM_THIS:
case ITM_NOT_THIS:
break;
// The RUNNING functions and LastNotNull all need to go back just one row.
//
case ITM_RUNNING_SUM:
case ITM_RUNNING_COUNT:
case ITM_RUNNING_MIN:
case ITM_RUNNING_MAX:
case ITM_RUNNING_CHANGE:
case ITM_LAST_NOT_NULL:
computedHistoryRows = MAXOF(computedHistoryRows, 2);
break;
///set to unable to compute for now-- will change later to compte values from frameStart_ and frameEnd_
case ITM_OLAP_SUM:
case ITM_OLAP_COUNT:
case ITM_OLAP_MIN:
case ITM_OLAP_MAX:
case ITM_OLAP_RANK:
case ITM_OLAP_DRANK:
{
if ( !outputFromChild.contains(itmExpr->getValueId()))
{
ItmSeqOlapFunction * olap = (ItmSeqOlapFunction*)itmExpr;
if (olap->isFrameStartUnboundedPreceding()) //(olap->getframeStart() == - INT_MAX)
{
computedHistoryRows = MAXOF(computedHistoryRows, 2);
}
else
{
computedHistoryRows = MAXOF(computedHistoryRows, ABS(olap->getframeStart()) + 2);
}
if (!olap->isFrameEndUnboundedFollowing()) //(olap->getframeEnd() != INT_MAX)
{
computedHistoryRows = MAXOF(computedHistoryRows, ABS(olap->getframeEnd()) + 1);
}
if (olap->isFrameEndUnboundedFollowing()) //(olap->getframeEnd() == INT_MAX)
{
unboundedFollowing = TRUE;
if (olap->getframeStart() > 0)
{
minFollowingRows = ((minFollowingRows > olap->getframeStart()) ?
minFollowingRows : olap->getframeStart());
}
} else if (olap->getframeEnd() > 0)
{
minFollowingRows = ((minFollowingRows > olap->getframeEnd()) ?
minFollowingRows : olap->getframeEnd());
}
}
}
break;
// If 'rows since', we cannot determine how much history is needed.
case ITM_ROWS_SINCE:
unableToCalculate = 1;
break;
// The MOVING and OFFSET functions need to go back as far as the value
// of their second child.
//
// The second argument can be:
// Constant: for these, we can use the constant value to set the upper bound
// for the history buffer.
// ItmScalarMinMax(child0, child1) (with operType = ITM_SCALAR_MIN)
// - if child0 or child1 is a constant, then we can use either one
// to set the upper bound.
case ITM_MOVING_MIN:
case ITM_MOVING_MAX:
case ITM_OFFSET:
for(Lng32 i = 1; i < itmExpr->getArity(); i++)
{
if (itmExpr->child(i)->getOperatorType() != ITM_NOTCOVERED)
{
ItemExpr * exprPtr = itmExpr->child(i);
NABoolean negate;
ConstValue *cv = exprPtr->castToConstValue(negate);
if (cv AND cv->canGetExactNumericValue())
{
Lng32 scale;
Int64 value64 = cv->getExactNumericValue(scale);
if(scale == 0 && value64 >= 0 && value64 < INT_MAX)
{
value64 = (negate ? -value64 : value64);
value = MAXOF((Lng32)value64, value);
}
}
else
{
if (exprPtr->getOperatorType() == ITM_SCALAR_MIN)
{
for(Lng32 j = 0; j < exprPtr->getArity(); j++)
{
if (exprPtr->child(j)->getOperatorType()
!= ITM_NOTCOVERED)
{
ItemExpr * exprPtr1 = exprPtr->child(j);
NABoolean negate1;
ConstValue *cv1 = exprPtr1->castToConstValue(negate1);
if (cv1 AND cv1->canGetExactNumericValue())
{
Lng32 scale1;
Int64 value64_1 = cv1->getExactNumericValue(scale1);
if(scale1 == 0 && value64_1 >= 0 && value64_1 < INT_MAX)
{
value64_1 = (negate1 ? -value64_1 : value64_1);
value = MAXOF((Lng32)value64_1, value);
}
}
}
}
}
} // end of inner else
}// end of if
}// end of for
// Check if the value is greater than zero.
// If it is, then save the value, but first
// increment the returned ConstValue by one.
// Otherwise, the offset or moving value was unable
// to be calculated.
if (value > 0)
{
value++;
computedHistoryRows = MAXOF(computedHistoryRows, value);
value = 0;
}
else
unableToCalculate = 1;
break;
default:
CMPASSERT(0);
}
}
// Gather all the children, and if not empty, recurse down to the
// next level of the tree.
//
for(Lng32 i = 0; i < valId.getItemExpr()->getArity(); i++) {
if (//valId.getItemExpr()->child(i)->getOperatorType() != ITM_NOTCOVERED //old stuff
!outputFromChild.contains(valId.getItemExpr()->child(i)->getValueId()))
{
children += valId.getItemExpr()->child(i)->getValueId();
}
}
}
if (NOT children.isEmpty())
{
computeHistoryRows(children,
computedHistoryRows,
unableToCalculate,
unboundedFollowing,
minFollowingRows,
outputFromChild);
}
} // PhysSequence::computeHistoryRows
// 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
short
PhysSequence::codeGen(Generator *generator)
{
// Get a local handle on some of the generator objects.
//
CollHeap *wHeap = generator->wHeap();
Space *space = generator->getSpace();
ExpGenerator *expGen = generator->getExpGenerator();
MapTable *mapTable = generator->getMapTable();
// Allocate a new map table for this node. This must be done
// before generating the code for my child so that this local
// map table will be sandwiched between the map tables already
// generated and the map tables generated by my offspring.
//
// Only the items available as output from this node will
// be put in the local map table. Before exiting this function, all of
// my offsprings map tables will be removed. Thus, none of the outputs
// from nodes below this node will be visible to nodes above it except
// those placed in the local map table and those that already exist in
// my ancestors map tables. This is the standard mechanism used in the
// generator for managing the access to item expressions.
//
MapTable *localMapTable = generator->appendAtEnd();
// Since this operation doesn't modify the row on the way down the tree,
// go ahead and generate the child subtree. Capture the given composite row
// descriptor and the child's returned TDB and composite row descriptor.
//
ex_cri_desc * givenCriDesc = generator->getCriDesc(Generator::DOWN);
child(0)->codeGen(generator);
ComTdb *childTdb = (ComTdb*)generator->getGenObj();
ex_cri_desc * childCriDesc = generator->getCriDesc(Generator::UP);
ExplainTuple *childExplainTuple = generator->getExplainTuple();
// Make all of my child's outputs map to ATP 1. The child row is only
// accessed in the project expression and it will be the second ATP
// (ATP 1) passed to this expression.
//
localMapTable->setAllAtp(1);
// My returned composite row has an additional tupp.
//
Int32 numberTuples = givenCriDesc->noTuples() + 1;
ex_cri_desc * returnCriDesc
#pragma nowarn(1506) // warning elimination
= new (space) ex_cri_desc(numberTuples, space);
#pragma warn(1506) // warning elimination
// For now, the history buffer row looks just the return row. Later,
// it may be useful to add an additional tupp for sequence function
// itermediates that are not needed above this node -- thus, this
// ATP is kept separate from the returned ATP.
//
const Int32 historyAtp = 0;
const Int32 historyAtpIndex = numberTuples-1;
#pragma nowarn(1506) // warning elimination
ex_cri_desc *historyCriDesc = new (space) ex_cri_desc(numberTuples, space);
#pragma warn(1506) // warning elimination
ExpTupleDesc *historyDesc = 0;
//seperate the read and retur expressions
seperateReadAndReturnItems(wHeap);
// The history buffer consists of items projected directly from the
// child, the root sequence functions, the value arguments of the
// offset functions, and running sequence functions. These elements must
// be materialized in the history buffer in order to be able to compute
// the outputs of this node -- the items projected directly from the child
// (projectValues) and the root sequence functions (sequenceFunctions).
//
// Compute the set of sequence function items that must be materialized
// int the history buffer. -- sequenceItems
//
// Compute the set of items in the history buffer: the union of the
// projected values and the value arguments. -- historyIds
//
// Compute the set of items in the history buffer that are computed:
// the difference between all the elements in the history buffer
// and the projected items. -- computedHistoryIds
//
// KB---will need to return atp with 3 tups only 0,1 and 2
// 2 -->values from history buffer after ther are moved to it
addCheckPartitionChangeExpr(generator, TRUE);
ValueIdSet historyIds;
historyIds += movePartIdsExpr();
historyIds += sequencedColumns();
ValueIdSet outputFromChild = child(0)->getGroupAttr()->getCharacteristicOutputs();
getHistoryAttributes(readSeqFunctions(),outputFromChild, historyIds, TRUE, wHeap);
// Add in the top level sequence functions.
historyIds += readSeqFunctions();
getHistoryAttributes(returnSeqFunctions(),outputFromChild, historyIds, TRUE, wHeap);
// Add in the top level functions.
historyIds += returnSeqFunctions();
// Layout the work tuple format which consists of the projected
// columns and the computed sequence functions. First, compute
// the number of attributes in the tuple.
//
ULng32 numberAttributes
= ((NOT historyIds.isEmpty()) ? historyIds.entries() : 0);
// Allocate an attribute pointer vector from the working heap.
//
Attributes **attrs = new(wHeap) Attributes*[numberAttributes];
// Fill in the attributes vector for the history buffer including
// adding the entries to the map table. Also, compute the value ID
// set for the elements to project from the child row.
//
//??????????re-visit this function??
computeHistoryAttributes(generator,
localMapTable,
attrs,
historyIds);
// Create the tuple descriptor for the history buffer row and
// assign the offsets to the attributes. For now, this layout is
// identical to the returned row. Set the tuple descriptors for
// the return and history rows.
//
ULng32 historyRecLen;
expGen->processAttributes(numberAttributes,
attrs,
ExpTupleDesc::SQLARK_EXPLODED_FORMAT,
historyRecLen,
historyAtp,
historyAtpIndex,
&historyDesc,
ExpTupleDesc::SHORT_FORMAT);
NADELETEBASIC(attrs, wHeap);
#pragma nowarn(1506) // warning elimination
returnCriDesc->setTupleDescriptor(historyAtpIndex, historyDesc);
#pragma warn(1506) // warning elimination
#pragma nowarn(1506) // warning elimination
historyCriDesc->setTupleDescriptor(historyAtpIndex, historyDesc);
#pragma warn(1506) // warning elimination
// If there are any sequence function items, generate the sequence
// function expressions.
//
ex_expr * readSeqExpr = NULL;
if(NOT readSeqFunctions().isEmpty())
{
ValueIdSet seqVals = readSeqFunctions();
seqVals += sequencedColumns();
seqVals += movePartIdsExpr();
expGen->generateSequenceExpression(seqVals,
readSeqExpr);
}
ex_expr *checkPartChangeExpr = NULL;
if (!checkPartitionChangeExpr().isEmpty()) {
ItemExpr * newCheckPartitionChangeTree=
checkPartitionChangeExpr().rebuildExprTree(ITM_AND,TRUE,TRUE);
expGen->generateExpr(newCheckPartitionChangeTree->getValueId(),
ex_expr::exp_SCAN_PRED,
&checkPartChangeExpr);
}
//unsigned long rowLength;
ex_expr * returnExpr = NULL;
if(NOT returnSeqFunctions().isEmpty())
{
expGen->generateSequenceExpression(returnSeqFunctions(),
returnExpr);
}
// Generate expression to evaluate predicate on the output
//
ex_expr *postPred = 0;
if (! selectionPred().isEmpty()) {
ItemExpr * newPredTree =
selectionPred().rebuildExprTree(ITM_AND,TRUE,TRUE);
expGen->generateExpr(newPredTree->getValueId(), ex_expr::exp_SCAN_PRED,
&postPred);
}
// Reset ATP's to zero for parent.
//
localMapTable->setAllAtp(0);
// Generate expression to evaluate the cancel expression
//
ex_expr *cancelExpression = 0;
if (! cancelExpr().isEmpty()) {
ItemExpr * newCancelExprTree =
cancelExpr().rebuildExprTree(ITM_AND,TRUE,TRUE);
expGen->generateExpr(newCancelExprTree->getValueId(), ex_expr::exp_SCAN_PRED,
&cancelExpression);
}
//
// For overflow
//
// ( The following are meaningless if ! unlimitedHistoryRows() )
NABoolean noOverflow =
CmpCommon::getDefault(EXE_BMO_DISABLE_OVERFLOW) == DF_ON ;
NABoolean logDiagnostics =
CmpCommon::getDefault(EXE_DIAGNOSTIC_EVENTS) == DF_ON ;
NABoolean possibleMultipleCalls = generator->getRightSideOfFlow() ;
short scratchTresholdPct =
(short) CmpCommon::getDefaultLong(SCRATCH_FREESPACE_THRESHOLD_PERCENT);
// determione the memory usage (amount of memory as percentage from total
// physical memory used to initialize data structures)
unsigned short memUsagePercent =
(unsigned short) getDefault(BMO_MEMORY_USAGE_PERCENT);
short memPressurePct = (short)getDefault(GEN_MEM_PRESSURE_THRESHOLD);
historyRecLen = ROUND8(historyRecLen);
Lng32 maxNumberOfOLAPBuffers;
Lng32 maxRowsInOLAPBuffer;
Lng32 minNumberOfOLAPBuffers;
Lng32 numberOfWinOLAPBuffers;
Lng32 olapBufferSize;
computeHistoryParams(historyRecLen,
maxRowsInOLAPBuffer,
minNumberOfOLAPBuffers,
numberOfWinOLAPBuffers,
maxNumberOfOLAPBuffers,
olapBufferSize);
ComTdbSequence *sequenceTdb
= new(space) ComTdbSequence(readSeqExpr,
returnExpr,
postPred,
cancelExpression,
getMinFollowingRows(),
#pragma nowarn(1506) // warning elimination
historyRecLen,
historyAtpIndex,
childTdb,
givenCriDesc,
returnCriDesc,
(queue_index)getDefault(GEN_SEQFUNC_SIZE_DOWN),
(queue_index)getDefault(GEN_SEQFUNC_SIZE_UP),
getDefault(GEN_SEQFUNC_NUM_BUFFERS),
getDefault(GEN_SEQFUNC_BUFFER_SIZE),
olapBufferSize,
maxNumberOfOLAPBuffers,
numHistoryRows(),
getUnboundedFollowing(),
logDiagnostics,
possibleMultipleCalls,
scratchTresholdPct,
memUsagePercent,
memPressurePct,
maxRowsInOLAPBuffer,
minNumberOfOLAPBuffers,
numberOfWinOLAPBuffers,
noOverflow,
checkPartChangeExpr);
#pragma warn(1506) // warning elimination
generator->initTdbFields(sequenceTdb);
// update the estimated value of HistoryRowLength with actual value
//setEstHistoryRowLength(historyIds.getRowLength());
double sequenceMemEst = getEstimatedRunTimeMemoryUsage(sequenceTdb);
generator->addToTotalEstimatedMemory(sequenceMemEst);
if(!generator->explainDisabled()) {
Lng32 seqMemEstInKBPerCPU = (Lng32)(sequenceMemEst / 1024) ;
seqMemEstInKBPerCPU = seqMemEstInKBPerCPU/
(MAXOF(generator->compilerStatsInfo().dop(),1));
generator->setOperEstimatedMemory(seqMemEstInKBPerCPU);
generator->
setExplainTuple(addExplainInfo(sequenceTdb,
childExplainTuple,
0,
generator));
generator->setOperEstimatedMemory(0);
}
sequenceTdb->setScratchIOVectorSize((Int16)getDefault(SCRATCH_IO_VECTOR_SIZE_HASH));
sequenceTdb->setOverflowMode(generator->getOverflowMode());
sequenceTdb->setBmoMinMemBeforePressureCheck((Int16)getDefault(EXE_BMO_MIN_SIZE_BEFORE_PRESSURE_CHECK_IN_MB));
if(generator->getOverflowMode() == ComTdb::OFM_SSD )
sequenceTdb->setBMOMaxMemThresholdMB((UInt16)(ActiveSchemaDB()->
getDefaults()).
getAsLong(SSD_BMO_MAX_MEM_THRESHOLD_IN_MB));
else
sequenceTdb->setBMOMaxMemThresholdMB((UInt16)(ActiveSchemaDB()->
getDefaults()).
getAsLong(EXE_MEMORY_AVAILABLE_IN_MB));
// The CQD EXE_MEM_LIMIT_PER_BMO_IN_MB has precedence over the mem quota sys
NADefaults &defs = ActiveSchemaDB()->getDefaults();
UInt16 mmu = (UInt16)(defs.getAsDouble(EXE_MEM_LIMIT_PER_BMO_IN_MB));
UInt16 numBMOsInFrag = (UInt16)generator->getFragmentDir()->getNumBMOs();
if (mmu != 0)
sequenceTdb->setMemoryQuotaMB(mmu);
else {
// Apply quota system if either one the following two is true:
// 1. the memory limit feature is turned off and more than one BMOs
// 2. the memory limit feature is turned on
NABoolean mlimitPerCPU = defs.getAsDouble(EXE_MEMORY_LIMIT_PER_CPU) > 0;
if ( mlimitPerCPU || numBMOsInFrag > 1 ) {
double memQuota =
computeMemoryQuota(generator->getEspLevel() == 0,
mlimitPerCPU,
generator->getBMOsMemoryLimitPerCPU().value(),
generator->getTotalNumBMOsPerCPU(),
generator->getTotalBMOsMemoryPerCPU().value(),
numBMOsInFrag,
generator->getFragmentDir()->getBMOsMemoryUsage()
);
sequenceTdb->setMemoryQuotaMB( UInt16(memQuota) );
}
}
generator->setCriDesc(givenCriDesc, Generator::DOWN);
generator->setCriDesc(returnCriDesc, Generator::UP);
generator->setGenObj(this, sequenceTdb);
return 0;
}
// AppliedStatMan::setupASMCacheForJBB method will be called from
// Query::Analyze after connectivity analysis has been done and
// empty logical properties have been set.
void AppliedStatMan::setupASMCacheForJBB(JBB & jbb)
{
EstLogPropSharedPtr myEstLogProp;
// get all JBBCs of JBB
const CANodeIdSet jbbcNodeIdSet = jbb.getMainJBBSubset().getJBBCs();
CANodeId jbbcId;
// for all jbbcs
for (jbbcId = jbbcNodeIdSet.init();
jbbcNodeIdSet.next(jbbcId);
jbbcNodeIdSet.advance(jbbcId))
{
if (NodeAnalysis * jbbcNode = jbbcId.getNodeAnalysis())
{
// Evaluate local predicates only if it is a table.
RelExpr * jbbcExpr = jbbcNode->getOriginalExpr();
if ((jbbcNode->getTableAnalysis() != NULL) &&
(jbbcExpr->getOperatorType() == REL_SCAN))
{
// get the original expression of the jbbc
Scan * scanExpr = (Scan *) jbbcExpr;
ValueIdSet localPreds = scanExpr->getSelectionPredicates();
// if local predicates have already been computed, then skip
if ((localPreds.entries() > 0) || !(lookup(jbbcId)))
{
// check to see this GA has already been associated with
// a logExpr for synthesis. If not, then synthesize
// log. expression, and then apply local predicates to it
if (NOT scanExpr->getGroupAttr()->existsLogExprForSynthesis())
scanExpr->synthLogProp();
myEstLogProp = getStatsForCANodeId(jbbcId);
}
}
}
}
// Now do a second traversal of the JBB looking for join reducers
for (jbbcId = jbbcNodeIdSet.init();
jbbcNodeIdSet.next(jbbcId);
jbbcNodeIdSet.advance(jbbcId))
{
// now look for all two way joins for this child
if (jbbcId.getNodeAnalysis())
{
// get all JBBCs connected to this JBBC, and do a two-way
// join with all of them
CANodeIdSet connectedNodes = jbbcId.getNodeAnalysis()->\
getJBBC()->getJoinedJBBCs();
for (CANodeId connectedTable = connectedNodes.init();
connectedNodes.next(connectedTable);
connectedNodes.advance(connectedTable))
{
if (connectedTable.getNodeAnalysis())
{
// ASM does not concern itself with the order of the tables,
// hence it is possible that the join has already been computed
CANodeIdSet tableSet = jbbcId;
tableSet.insert(connectedTable);
if ((myEstLogProp = getCachedStatistics(&tableSet)) == NULL)
{
CANodeIdSet setForjbbcId(jbbcId);
CANodeIdSet setForConnectedTable(connectedTable);
myEstLogProp = joinJBBChildren(setForjbbcId, setForConnectedTable);
}
}
}
}
}
} // AppliedStatMan::setupASMCacheForJBB
// compress the histograms based on query predicates on this table
void TableDesc::compressHistogramsForCurrentQuery()
{
// if there are some column statistics
if ((colStats_.entries() != 0) &&
(table_) &&
(table_->getExtendedQualName().getSpecialType() == ExtendedQualName::NORMAL_TABLE))
{ // if 1
// check if query analysis info is available
if(QueryAnalysis::Instance()->isAnalysisON())
{ // if 2
// get a handle to the query analysis
QueryAnalysis* queryAnalysis = QueryAnalysis::Instance();
// get a handle to the table analysis
const TableAnalysis * tableAnalysis = getTableAnalysis();
if(!tableAnalysis)
return;
// iterate over statistics for each column
for(CollIndex i = 0; i < colStats_.entries(); i++)
{ // for 1
// Get a handle to the column's statistics descriptor
ColStatDescSharedPtr columnStatDesc = colStats_[i];
// get a handle to the ColStats
ColStatsSharedPtr colStats = columnStatDesc->getColStats();
// if this is a single column, as opposed to a multicolumn
if(colStats->getStatColumns().entries() == 1)
{ // if 3
// get column's value id
const ValueId columnId = columnStatDesc->getColumn();
// get column analysis
ColAnalysis* colAnalysis = queryAnalysis->getColAnalysis(columnId);
if(!colAnalysis) continue;
ValueIdSet predicatesOnColumn =
colAnalysis->getReferencingPreds();
// we can compress this column's histogram if there
// is a equality predicate against a constant
ItemExpr *constant = NULL;
NABoolean colHasEqualityAgainstConst =
colAnalysis->getConstValue(constant);
// if a equality predicate with a constant was found
// i.e. predicate of the form col = 5
if (colHasEqualityAgainstConst)
{ // if 4
if (constant)
// compress the histogram
columnStatDesc->compressColStatsForQueryPreds(constant,constant);
} // if 4
else { // else 4
// since there is no equality predicates we might still
// be able to compress the column's histogram based on
// range predicates against a constant. Following are
// examples of such predicates
// * col > 1 <-- predicate defines a lower bound
// * col < 3 <-- predicate defines a upper bound
// * col >1 and col < 30 <-- window predicate, define both bounds
ItemExpr * lowerBound = NULL;
ItemExpr * upperBound = NULL;
// Extract predicates from range spec and add it to the
// original predicate set otherwise isARangePredicate() will
// return FALSE, so histgram compression won't happen.
ValueIdSet rangeSpecPred(predicatesOnColumn);
for (ValueId predId= rangeSpecPred.init();
rangeSpecPred.next(predId);
rangeSpecPred.advance(predId))
{
ItemExpr * pred = predId.getItemExpr();
if ( pred->getOperatorType() == ITM_RANGE_SPEC_FUNC )
{
ValueIdSet vs;
((RangeSpecRef *)pred)->getValueIdSetForReconsItemExpr(vs);
// remove rangespec vid from the original set
predicatesOnColumn.remove(predId);
// add preds extracted from rangespec to the original set
predicatesOnColumn.insert(vs);
}
}
// in the following loop we iterate over all the predicates
// on this column. If there is a range predicate e.g. a > 2
// or a < 3, then we use that to define upper and lower bounds.
// Given predicate a > 2, we get a lower bound of 2.
// Given predicate a < 3, we get a upper bound of 3.
// The bound are then passed down to the histogram
// compression methods.
// iterate over predicates to see if any of them is a range
// predicate e.g. a > 2
for (ValueId predId= predicatesOnColumn.init();
predicatesOnColumn.next(predId);
predicatesOnColumn.advance(predId))
{ // for 2
// check if this predicate is a range predicate
ItemExpr * predicateOnColumn = predId.getItemExpr();
if (predicateOnColumn->isARangePredicate())
{ // if 5
// if a predicate is a range predicate we need to find out more
// information regarding the predicate to see if it can be used
// to compress the columns histogram. We look for the following:
// * The predicate is against a constant e.g. a > 3 and not against
// another column e.g. a > b
// Also give a predicate we need to find out what side is the column
// and what side is the constant. Normally people write a range predicate
// as a > 3, but the same could be written as 3 < a.
// Also either on of the operands of the range predicate might be
// a VEG, if so then we need to dig into the VEG to see where is
// the constant and where is the column.
// check the right and left children of this predicate to
// see if one of them is a constant
ItemExpr * leftChildItemExpr = (ItemExpr *) predicateOnColumn->getChild(0);
ItemExpr * rightChildItemExpr = (ItemExpr *) predicateOnColumn->getChild(1);
// by default assume the literal is at right i.e. predicate of
// the form a > 2
NABoolean columnAtRight = FALSE;
// check if right child of predicate is a VEG
if ( rightChildItemExpr->getOperatorType() == ITM_VEG_REFERENCE)
{ // if 6
// if child is a VEG
VEGReference * rightChildVEG = (VEGReference *) rightChildItemExpr;
// check if the VEG contains the current column
// if it does contain the current column then
// the predicate has the column on right and potentially
// a constant on the left.
if(rightChildVEG->getVEG()->getAllValues().contains(columnId))
{ // if 7
// column is at right i.e. predicate is of the form
// 2 < a
columnAtRight = TRUE;
} // if 7
} // if 6
else { // else 6
// child is not a VEG
if ( columnId == rightChildItemExpr->getValueId() )
{ // if 8
// literals are at left i.e. predicate is of the form
// (1,2) < (a, b)
columnAtRight = TRUE;
} // if 8
} // else 6
ItemExpr * potentialConstantExpr = NULL;
// check if the range predicate is against a constant
if (columnAtRight)
{ // if 9
// the left child is potentially a constant
potentialConstantExpr = leftChildItemExpr;
} // if 9
else { // else 9
// the right child is potentially a constant
potentialConstantExpr = rightChildItemExpr;
} // else 9
// initialize constant to NULL before
// looking for next constant
constant = NULL;
// check if potentialConstantExpr contains a constant.
// we need to see if this range predicate is a predicate
// against a constant e.g col > 1 and not a predicate
// against another column e.g. col > anothercol
// if the expression is a VEG
if ( potentialConstantExpr->getOperatorType() == ITM_VEG_REFERENCE)
{ // if 10
// expression is a VEG, dig into the VEG to
// get see if it contains a constant
VEGReference * potentialConstantExprVEG =
(VEGReference *) potentialConstantExpr;
potentialConstantExprVEG->getVEG()->\
getAllValues().referencesAConstValue(&constant);
} // if 10
else { // else 10
// express is not a VEG, it is a constant
if ( potentialConstantExpr->getOperatorType() == ITM_CONSTANT )
constant = potentialConstantExpr;
} // else 10
// if predicate involves a constant, does the constant imply
// a upper bound or lower bound
if (constant)
{ // if 11
// if range predicate has column at right e.g. 3 > a
if (columnAtRight)
{ // if 12
if ( predicateOnColumn->getOperatorType() == ITM_GREATER ||
predicateOnColumn->getOperatorType() == ITM_GREATER_EQ)
{ // if 13
if (!upperBound)
upperBound = constant;
} // if 13
else
{ // else 13
if (!lowerBound)
lowerBound = constant;
} // else 13
} // if 12
else { // else 12
// range predicate has column at left e.g. a < 3
if ( predicateOnColumn->getOperatorType() == ITM_LESS ||
predicateOnColumn->getOperatorType() == ITM_LESS_EQ)
{ // if 14
if (!upperBound)
upperBound = constant;
} // if 14
else
{ // else 14
if (!lowerBound)
lowerBound = constant;
} // else 14
} // else 12
} // if 11
} // if 5
} // for 2
// if we found a upper bound or a lower bound
if (lowerBound || upperBound)
{
// compress the histogram based on range predicates
columnStatDesc->compressColStatsForQueryPreds(lowerBound, upperBound);
}
} // else 4
} // if 3
} // for 1
} // if 2
} // if 1
// All histograms compressed. Set the histCompressed flag to TRUE
histsCompressed(TRUE);
}
// 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
