use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.util.ImmutableBitSet in project calcite by apache.
the class LoptOptimizeJoinRule method addFilters.
/**
* Determines which join filters can be added to the current join tree. Note
* that the join filter still reflects the original join ordering. It will
* only be adjusted to reflect the new join ordering if the "adjust"
* parameter is set to true.
*
* @param multiJoin join factors being optimized
* @param leftTree left subtree of the join tree
* @param leftIdx if ≥ 0, only consider filters that reference leftIdx in
* leftTree; otherwise, consider all filters that reference any factor in
* leftTree
* @param rightTree right subtree of the join tree
* @param filtersToAdd remaining join filters that need to be added; those
* that are added are removed from the list
* @param adjust if true, adjust filter to reflect new join ordering
*
* @return AND'd expression of the join filters that can be added to the
* current join tree
*/
private RexNode addFilters(LoptMultiJoin multiJoin, LoptJoinTree leftTree, int leftIdx, LoptJoinTree rightTree, List<RexNode> filtersToAdd, boolean adjust) {
// loop through the remaining filters to be added and pick out the
// ones that reference only the factors in the new join tree
final RexBuilder rexBuilder = multiJoin.getMultiJoinRel().getCluster().getRexBuilder();
final ImmutableBitSet.Builder childFactorBuilder = ImmutableBitSet.builder();
childFactorBuilder.addAll(rightTree.getTreeOrder());
if (leftIdx >= 0) {
childFactorBuilder.set(leftIdx);
} else {
childFactorBuilder.addAll(leftTree.getTreeOrder());
}
for (int child : rightTree.getTreeOrder()) {
childFactorBuilder.set(child);
}
final ImmutableBitSet childFactor = childFactorBuilder.build();
RexNode condition = null;
final ListIterator<RexNode> filterIter = filtersToAdd.listIterator();
while (filterIter.hasNext()) {
RexNode joinFilter = filterIter.next();
ImmutableBitSet filterBitmap = multiJoin.getFactorsRefByJoinFilter(joinFilter);
// AND the filter to the current join condition
if (childFactor.contains(filterBitmap)) {
if (condition == null) {
condition = joinFilter;
} else {
condition = rexBuilder.makeCall(SqlStdOperatorTable.AND, condition, joinFilter);
}
filterIter.remove();
}
}
if (adjust && (condition != null)) {
int[] adjustments = new int[multiJoin.getNumTotalFields()];
if (needsAdjustment(multiJoin, adjustments, leftTree, rightTree, false)) {
condition = condition.accept(new RelOptUtil.RexInputConverter(rexBuilder, multiJoin.getMultiJoinFields(), leftTree.getJoinTree().getRowType().getFieldList(), rightTree.getJoinTree().getRowType().getFieldList(), adjustments));
}
}
if (condition == null) {
condition = rexBuilder.makeLiteral(true);
}
return condition;
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.util.ImmutableBitSet in project calcite by apache.
the class LoptOptimizeJoinRule method findRemovableSelfJoins.
/**
* Locates pairs of joins that are self-joins where the join can be removed
* because the join condition between the two factors is an equality join on
* unique keys.
*
* @param multiJoin join factors being optimized
*/
private void findRemovableSelfJoins(RelMetadataQuery mq, LoptMultiJoin multiJoin) {
// Candidates for self-joins must be simple factors
Map<Integer, RelOptTable> simpleFactors = getSimpleFactors(mq, multiJoin);
// See if a simple factor is repeated and therefore potentially is
// part of a self-join. Restrict each factor to at most one
// self-join.
final List<RelOptTable> repeatedTables = new ArrayList<>();
final TreeSet<Integer> sortedFactors = new TreeSet<>();
sortedFactors.addAll(simpleFactors.keySet());
final Map<Integer, Integer> selfJoinPairs = new HashMap<>();
Integer[] factors = sortedFactors.toArray(new Integer[sortedFactors.size()]);
for (int i = 0; i < factors.length; i++) {
if (repeatedTables.contains(simpleFactors.get(factors[i]))) {
continue;
}
for (int j = i + 1; j < factors.length; j++) {
int leftFactor = factors[i];
int rightFactor = factors[j];
if (simpleFactors.get(leftFactor).getQualifiedName().equals(simpleFactors.get(rightFactor).getQualifiedName())) {
selfJoinPairs.put(leftFactor, rightFactor);
repeatedTables.add(simpleFactors.get(leftFactor));
break;
}
}
}
// allow the join to be removed.
for (Integer factor1 : selfJoinPairs.keySet()) {
final int factor2 = selfJoinPairs.get(factor1);
final List<RexNode> selfJoinFilters = new ArrayList<>();
for (RexNode filter : multiJoin.getJoinFilters()) {
ImmutableBitSet joinFactors = multiJoin.getFactorsRefByJoinFilter(filter);
if ((joinFactors.cardinality() == 2) && joinFactors.get(factor1) && joinFactors.get(factor2)) {
selfJoinFilters.add(filter);
}
}
if ((selfJoinFilters.size() > 0) && isSelfJoinFilterUnique(mq, multiJoin, factor1, factor2, selfJoinFilters)) {
multiJoin.addRemovableSelfJoinPair(factor1, factor2);
}
}
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.util.ImmutableBitSet in project calcite by apache.
the class IntersectToDistinctRule method onMatch.
// ~ Methods ----------------------------------------------------------------
public void onMatch(RelOptRuleCall call) {
final Intersect intersect = call.rel(0);
if (intersect.all) {
// nothing we can do
return;
}
final RelOptCluster cluster = intersect.getCluster();
final RexBuilder rexBuilder = cluster.getRexBuilder();
final RelBuilder relBuilder = call.builder();
// 1st level GB: create a GB (col0, col1, count() as c) for each branch
for (RelNode input : intersect.getInputs()) {
relBuilder.push(input);
relBuilder.aggregate(relBuilder.groupKey(relBuilder.fields()), relBuilder.countStar(null));
}
// create a union above all the branches
final int branchCount = intersect.getInputs().size();
relBuilder.union(true, branchCount);
final RelNode union = relBuilder.peek();
// 2nd level GB: create a GB (col0, col1, count(c)) for each branch
// the index of c is union.getRowType().getFieldList().size() - 1
final int fieldCount = union.getRowType().getFieldCount();
final ImmutableBitSet groupSet = ImmutableBitSet.range(fieldCount - 1);
relBuilder.aggregate(relBuilder.groupKey(groupSet, null), relBuilder.countStar(null));
// add a filter count(c) = #branches
relBuilder.filter(relBuilder.equals(relBuilder.field(fieldCount - 1), rexBuilder.makeBigintLiteral(new BigDecimal(branchCount))));
// Project all but the last field
relBuilder.project(Util.skipLast(relBuilder.fields()));
// the schema for intersect distinct is like this
// R3 on all attributes + count(c) as cnt
// finally add a project to project out the last column
call.transformTo(relBuilder.build());
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.util.ImmutableBitSet in project calcite by apache.
the class JoinAssociateRule method onMatch.
// ~ Methods ----------------------------------------------------------------
public void onMatch(final RelOptRuleCall call) {
final Join topJoin = call.rel(0);
final Join bottomJoin = call.rel(1);
final RelNode relA = bottomJoin.getLeft();
final RelNode relB = bottomJoin.getRight();
final RelSubset relC = call.rel(2);
final RelOptCluster cluster = topJoin.getCluster();
final RexBuilder rexBuilder = cluster.getRexBuilder();
if (relC.getConvention() != relA.getConvention()) {
// EnumerableConvention, we're only interested in enumerable subsets.
return;
}
// topJoin
// / \
// bottomJoin C
// / \
// A B
final int aCount = relA.getRowType().getFieldCount();
final int bCount = relB.getRowType().getFieldCount();
final int cCount = relC.getRowType().getFieldCount();
final ImmutableBitSet aBitSet = ImmutableBitSet.range(0, aCount);
final ImmutableBitSet bBitSet = ImmutableBitSet.range(aCount, aCount + bCount);
if (!topJoin.getSystemFieldList().isEmpty()) {
// FIXME Enable this rule for joins with system fields
return;
}
// (Is this too strict?)
if (topJoin.getJoinType() != JoinRelType.INNER || bottomJoin.getJoinType() != JoinRelType.INNER) {
return;
}
// Goal is to transform to
//
// newTopJoin
// / \
// A newBottomJoin
// / \
// B C
// Split the condition of topJoin and bottomJoin into a conjunctions. A
// condition can be pushed down if it does not use columns from A.
final List<RexNode> top = Lists.newArrayList();
final List<RexNode> bottom = Lists.newArrayList();
JoinPushThroughJoinRule.split(topJoin.getCondition(), aBitSet, top, bottom);
JoinPushThroughJoinRule.split(bottomJoin.getCondition(), aBitSet, top, bottom);
// Mapping for moving conditions from topJoin or bottomJoin to
// newBottomJoin.
// target: | B | C |
// source: | A | B | C |
final Mappings.TargetMapping bottomMapping = Mappings.createShiftMapping(aCount + bCount + cCount, 0, aCount, bCount, bCount, aCount + bCount, cCount);
final List<RexNode> newBottomList = Lists.newArrayList();
new RexPermuteInputsShuttle(bottomMapping, relB, relC).visitList(bottom, newBottomList);
RexNode newBottomCondition = RexUtil.composeConjunction(rexBuilder, newBottomList, false);
final Join newBottomJoin = bottomJoin.copy(bottomJoin.getTraitSet(), newBottomCondition, relB, relC, JoinRelType.INNER, false);
// Condition for newTopJoin consists of pieces from bottomJoin and topJoin.
// Field ordinals do not need to be changed.
RexNode newTopCondition = RexUtil.composeConjunction(rexBuilder, top, false);
final Join newTopJoin = topJoin.copy(topJoin.getTraitSet(), newTopCondition, relA, newBottomJoin, JoinRelType.INNER, false);
call.transformTo(newTopJoin);
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.util.ImmutableBitSet in project calcite by apache.
the class LoptSemiJoinOptimizer method removeJoin.
/**
* Determines whether a join of the dimension table in a semijoin can be
* removed. It can be if the dimension keys are unique and the only fields
* referenced from the dimension table are its semijoin keys. The semijoin
* keys can be mapped to the corresponding keys from the fact table (because
* of the equality condition associated with the semijoin keys). Therefore,
* that's why the dimension table can be removed even though those fields
* are referenced elsewhere in the query tree.
*
* @param multiJoin join factors being optimized
* @param semiJoin semijoin under consideration
* @param factIdx id of the fact table in the semijoin
* @param dimIdx id of the dimension table in the semijoin
*/
private void removeJoin(LoptMultiJoin multiJoin, SemiJoin semiJoin, int factIdx, int dimIdx) {
// no need to proceed any further
if (multiJoin.getJoinRemovalFactor(dimIdx) != null) {
return;
}
// Check if the semijoin keys corresponding to the dimension table
// are unique. The semijoin will filter out the nulls.
final ImmutableBitSet dimKeys = ImmutableBitSet.of(semiJoin.getRightKeys());
final RelNode dimRel = multiJoin.getJoinFactor(dimIdx);
if (!RelMdUtil.areColumnsDefinitelyUniqueWhenNullsFiltered(mq, dimRel, dimKeys)) {
return;
}
// check that the only fields referenced from the dimension table
// in either its projection or join conditions are the dimension
// keys
ImmutableBitSet dimProjRefs = multiJoin.getProjFields(dimIdx);
if (dimProjRefs == null) {
int nDimFields = multiJoin.getNumFieldsInJoinFactor(dimIdx);
dimProjRefs = ImmutableBitSet.range(0, nDimFields);
}
if (!dimKeys.contains(dimProjRefs)) {
return;
}
int[] dimJoinRefCounts = multiJoin.getJoinFieldRefCounts(dimIdx);
for (int i = 0; i < dimJoinRefCounts.length; i++) {
if (dimJoinRefCounts[i] > 0) {
if (!dimKeys.get(i)) {
return;
}
}
}
// criteria met; keep track of the fact table and the semijoin that
// allow the join of this dimension table to be removed
multiJoin.setJoinRemovalFactor(dimIdx, factIdx);
multiJoin.setJoinRemovalSemiJoin(dimIdx, semiJoin);
// dimension table doesn't need to use those keys
if (dimProjRefs.cardinality() != 0) {
return;
}
for (int i = 0; i < dimJoinRefCounts.length; i++) {
if (dimJoinRefCounts[i] > 1) {
return;
} else if (dimJoinRefCounts[i] == 1) {
if (!dimKeys.get(i)) {
return;
}
}
}
int[] factJoinRefCounts = multiJoin.getJoinFieldRefCounts(factIdx);
for (Integer key : semiJoin.getLeftKeys()) {
factJoinRefCounts[key]--;
}
}
Aggregations