Examples with HiveAggregate org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveAggregate used on opensource projects

Example 6 with HiveAggregate

use of org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveAggregate in project hive by apache.

the class HiveExceptRewriteRule method createFirstGB.

private RelNode createFirstGB(RelNode input, boolean left, RelOptCluster cluster, RexBuilder rexBuilder) throws CalciteSemanticException {
    final List<RexNode> gbChildProjLst = Lists.newArrayList();
    final List<Integer> groupSetPositions = Lists.newArrayList();
    for (int cInd = 0; cInd < input.getRowType().getFieldList().size(); cInd++) {
        gbChildProjLst.add(rexBuilder.makeInputRef(input, cInd));
        groupSetPositions.add(cInd);
    }
    if (left) {
        gbChildProjLst.add(rexBuilder.makeBigintLiteral(new BigDecimal(2)));
    } else {
        gbChildProjLst.add(rexBuilder.makeBigintLiteral(new BigDecimal(1)));
    }
    // also add the last VCol
    groupSetPositions.add(input.getRowType().getFieldList().size());
    // create the project before GB
    RelNode gbInputRel = HiveProject.create(input, gbChildProjLst, null);
    // groupSetPosition includes all the positions
    final ImmutableBitSet groupSet = ImmutableBitSet.of(groupSetPositions);
    List<AggregateCall> aggregateCalls = Lists.newArrayList();
    RelDataType aggFnRetType = TypeConverter.convert(TypeInfoFactory.longTypeInfo, cluster.getTypeFactory());
    AggregateCall aggregateCall = HiveCalciteUtil.createSingleArgAggCall("count", cluster, TypeInfoFactory.longTypeInfo, input.getRowType().getFieldList().size(), aggFnRetType);
    aggregateCalls.add(aggregateCall);
    return new HiveAggregate(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), gbInputRel, false, groupSet, null, aggregateCalls);
}

Example 7 with HiveAggregate

use of org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveAggregate in project hive by apache.

the class HiveExceptRewriteRule method onMatch.

// ~ Methods ----------------------------------------------------------------
public void onMatch(RelOptRuleCall call) {
    final HiveExcept hiveExcept = call.rel(0);
    final RelOptCluster cluster = hiveExcept.getCluster();
    final RexBuilder rexBuilder = cluster.getRexBuilder();
    Builder<RelNode> bldr = new ImmutableList.Builder<RelNode>();
    // branch
    try {
        bldr.add(createFirstGB(hiveExcept.getInputs().get(0), true, cluster, rexBuilder));
        bldr.add(createFirstGB(hiveExcept.getInputs().get(1), false, cluster, rexBuilder));
    } catch (CalciteSemanticException e) {
        LOG.debug(e.toString());
        throw new RuntimeException(e);
    }
    // create a union above all the branches
    // the schema of union looks like this
    // all keys + VCol + c
    HiveRelNode union = new HiveUnion(cluster, TraitsUtil.getDefaultTraitSet(cluster), bldr.build());
    // 2nd level GB: create a GB (all keys + sum(c) as a + sum(VCol*c) as b) for
    // each branch
    final List<RexNode> gbChildProjLst = Lists.newArrayList();
    final List<Integer> groupSetPositions = Lists.newArrayList();
    int unionColumnSize = union.getRowType().getFieldList().size();
    for (int cInd = 0; cInd < unionColumnSize; cInd++) {
        gbChildProjLst.add(rexBuilder.makeInputRef(union, cInd));
        // the last 2 columns are VCol and c
        if (cInd < unionColumnSize - 2) {
            groupSetPositions.add(cInd);
        }
    }
    try {
        gbChildProjLst.add(multiply(rexBuilder.makeInputRef(union, unionColumnSize - 2), rexBuilder.makeInputRef(union, unionColumnSize - 1), cluster, rexBuilder));
    } catch (CalciteSemanticException e) {
        LOG.debug(e.toString());
        throw new RuntimeException(e);
    }
    RelNode gbInputRel = null;
    try {
        // Here we create a project for the following reasons:
        // (1) GBy only accepts arg as a position of the input, however, we need to sum on VCol*c
        // (2) This can better reuse the function createSingleArgAggCall.
        gbInputRel = HiveProject.create(union, gbChildProjLst, null);
    } catch (CalciteSemanticException e) {
        LOG.debug(e.toString());
        throw new RuntimeException(e);
    }
    // gbInputRel's schema is like this
    // all keys + VCol + c + VCol*c
    List<AggregateCall> aggregateCalls = Lists.newArrayList();
    RelDataType aggFnRetType = TypeConverter.convert(TypeInfoFactory.longTypeInfo, cluster.getTypeFactory());
    // sum(c)
    AggregateCall aggregateCall = HiveCalciteUtil.createSingleArgAggCall("sum", cluster, TypeInfoFactory.longTypeInfo, unionColumnSize - 1, aggFnRetType);
    aggregateCalls.add(aggregateCall);
    // sum(VCol*c)
    aggregateCall = HiveCalciteUtil.createSingleArgAggCall("sum", cluster, TypeInfoFactory.longTypeInfo, unionColumnSize, aggFnRetType);
    aggregateCalls.add(aggregateCall);
    final ImmutableBitSet groupSet = ImmutableBitSet.of(groupSetPositions);
    HiveRelNode aggregateRel = new HiveAggregate(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), gbInputRel, false, groupSet, null, aggregateCalls);
    if (!hiveExcept.all) {
        RelNode filterRel = null;
        try {
            filterRel = new HiveFilter(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), aggregateRel, makeFilterExprForExceptDistinct(aggregateRel, unionColumnSize, cluster, rexBuilder));
        } catch (CalciteSemanticException e) {
            LOG.debug(e.toString());
            throw new RuntimeException(e);
        }
        // finally add a project to project out the last 2 columns
        Set<Integer> projectOutColumnPositions = new HashSet<>();
        projectOutColumnPositions.add(filterRel.getRowType().getFieldList().size() - 2);
        projectOutColumnPositions.add(filterRel.getRowType().getFieldList().size() - 1);
        try {
            call.transformTo(HiveCalciteUtil.createProjectWithoutColumn(filterRel, projectOutColumnPositions));
        } catch (CalciteSemanticException e) {
            LOG.debug(e.toString());
            throw new RuntimeException(e);
        }
    } else {
        List<RexNode> originalInputRefs = Lists.transform(aggregateRel.getRowType().getFieldList(), new Function<RelDataTypeField, RexNode>() {

            @Override
            public RexNode apply(RelDataTypeField input) {
                return new RexInputRef(input.getIndex(), input.getType());
            }
        });
        List<RexNode> copyInputRefs = new ArrayList<>();
        try {
            copyInputRefs.add(makeExprForExceptAll(aggregateRel, unionColumnSize, cluster, rexBuilder));
        } catch (CalciteSemanticException e) {
            LOG.debug(e.toString());
            throw new RuntimeException(e);
        }
        for (int i = 0; i < originalInputRefs.size() - 2; i++) {
            copyInputRefs.add(originalInputRefs.get(i));
        }
        RelNode srcRel = null;
        try {
            srcRel = HiveProject.create(aggregateRel, copyInputRefs, null);
            HiveTableFunctionScan udtf = HiveCalciteUtil.createUDTFForSetOp(cluster, srcRel);
            // finally add a project to project out the 1st columns
            Set<Integer> projectOutColumnPositions = new HashSet<>();
            projectOutColumnPositions.add(0);
            call.transformTo(HiveCalciteUtil.createProjectWithoutColumn(udtf, projectOutColumnPositions));
        } catch (SemanticException e) {
            LOG.debug(e.toString());
            throw new RuntimeException(e);
        }
    }
}

Example 8 with HiveAggregate

use of org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveAggregate in project hive by apache.

the class HiveExpandDistinctAggregatesRule method createCount.

/**
   * @param aggr: the original aggregate
   * @param argList: the original argList in aggregate
   * @param cleanArgList: the new argList without duplicates
   * @param map: the mapping from the original argList to the new argList
   * @param sourceOfForCountDistinct: the sorted positions of groupset
   * @return
   * @throws CalciteSemanticException
   */
private RelNode createCount(Aggregate aggr, List<List<Integer>> argList, List<List<Integer>> cleanArgList, Map<Integer, Integer> map, List<Integer> sourceOfForCountDistinct) throws CalciteSemanticException {
    List<RexNode> originalInputRefs = Lists.transform(aggr.getRowType().getFieldList(), new Function<RelDataTypeField, RexNode>() {

        @Override
        public RexNode apply(RelDataTypeField input) {
            return new RexInputRef(input.getIndex(), input.getType());
        }
    });
    final List<RexNode> gbChildProjLst = Lists.newArrayList();
    // for non-singular args, count can include null, i.e. (,) is counted as 1
    for (List<Integer> list : cleanArgList) {
        RexNode condition = rexBuilder.makeCall(SqlStdOperatorTable.EQUALS, originalInputRefs.get(originalInputRefs.size() - 1), rexBuilder.makeExactLiteral(new BigDecimal(getGroupingIdValue(list, sourceOfForCountDistinct, aggr.getGroupCount()))));
        if (list.size() == 1) {
            int pos = list.get(0);
            RexNode notNull = rexBuilder.makeCall(SqlStdOperatorTable.IS_NOT_NULL, originalInputRefs.get(pos));
            condition = rexBuilder.makeCall(SqlStdOperatorTable.AND, condition, notNull);
        }
        RexNode when = rexBuilder.makeCall(SqlStdOperatorTable.CASE, condition, rexBuilder.makeExactLiteral(BigDecimal.ONE), rexBuilder.constantNull());
        gbChildProjLst.add(when);
    }
    // create the project before GB
    RelNode gbInputRel = HiveProject.create(aggr, gbChildProjLst, null);
    // create the aggregate
    List<AggregateCall> aggregateCalls = Lists.newArrayList();
    RelDataType aggFnRetType = TypeConverter.convert(TypeInfoFactory.longTypeInfo, cluster.getTypeFactory());
    for (int i = 0; i < cleanArgList.size(); i++) {
        AggregateCall aggregateCall = HiveCalciteUtil.createSingleArgAggCall("count", cluster, TypeInfoFactory.longTypeInfo, i, aggFnRetType);
        aggregateCalls.add(aggregateCall);
    }
    Aggregate aggregate = new HiveAggregate(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), gbInputRel, false, ImmutableBitSet.of(), null, aggregateCalls);
    // count(distinct x, y), count(distinct y, x), we find the correct mapping.
    if (map.isEmpty()) {
        return aggregate;
    } else {
        List<RexNode> originalAggrRefs = Lists.transform(aggregate.getRowType().getFieldList(), new Function<RelDataTypeField, RexNode>() {

            @Override
            public RexNode apply(RelDataTypeField input) {
                return new RexInputRef(input.getIndex(), input.getType());
            }
        });
        final List<RexNode> projLst = Lists.newArrayList();
        int index = 0;
        for (int i = 0; i < argList.size(); i++) {
            if (map.containsKey(i)) {
                projLst.add(originalAggrRefs.get(map.get(i)));
            } else {
                projLst.add(originalAggrRefs.get(index++));
            }
        }
        return HiveProject.create(aggregate, projLst, null);
    }
}

Example 9 with HiveAggregate

use of org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveAggregate in project hive by apache.

the class HiveExpandDistinctAggregatesRule method createGroupingSets.

/**
   * @param aggregate: the original aggregate
   * @param argList: the original argList in aggregate
   * @param cleanArgList: the new argList without duplicates
   * @param map: the mapping from the original argList to the new argList
   * @param sourceOfForCountDistinct: the sorted positions of groupset
   * @return
   */
private Aggregate createGroupingSets(Aggregate aggregate, List<List<Integer>> argList, List<List<Integer>> cleanArgList, Map<Integer, Integer> map, List<Integer> sourceOfForCountDistinct) {
    final ImmutableBitSet groupSet = ImmutableBitSet.of(sourceOfForCountDistinct);
    final List<ImmutableBitSet> origGroupSets = new ArrayList<>();
    for (int i = 0; i < argList.size(); i++) {
        List<Integer> list = argList.get(i);
        ImmutableBitSet bitSet = ImmutableBitSet.of(list);
        int prev = origGroupSets.indexOf(bitSet);
        if (prev == -1) {
            origGroupSets.add(bitSet);
            cleanArgList.add(list);
        } else {
            map.put(i, prev);
        }
    }
    // Calcite expects the grouping sets sorted and without duplicates
    Collections.sort(origGroupSets, ImmutableBitSet.COMPARATOR);
    List<AggregateCall> aggregateCalls = new ArrayList<AggregateCall>();
    // Create GroupingID column
    AggregateCall aggCall = AggregateCall.create(HiveGroupingID.INSTANCE, false, new ImmutableList.Builder<Integer>().build(), -1, this.cluster.getTypeFactory().createSqlType(SqlTypeName.INTEGER), HiveGroupingID.INSTANCE.getName());
    aggregateCalls.add(aggCall);
    return new HiveAggregate(cluster, cluster.traitSetOf(HiveRelNode.CONVENTION), aggregate.getInput(), true, groupSet, origGroupSets, aggregateCalls);
}

Example 10 with HiveAggregate

use of org.apache.hadoop.hive.ql.optimizer.calcite.reloperators.HiveAggregate in project hive by apache.

the class HiveRelColumnsAlignment method align.

public RelNode align(Aggregate rel, List<RelFieldCollation> collations) {
    // 1) We extract the group by positions that are part of the collations and
    // sort them so they respect it
    LinkedHashSet<Integer> aggregateColumnsOrder = new LinkedHashSet<>();
    ImmutableList.Builder<RelFieldCollation> propagateCollations = ImmutableList.builder();
    if (!rel.indicator && !collations.isEmpty()) {
        for (RelFieldCollation c : collations) {
            if (c.getFieldIndex() < rel.getGroupCount()) {
                // Group column found
                if (aggregateColumnsOrder.add(c.getFieldIndex())) {
                    propagateCollations.add(c.copy(rel.getGroupSet().nth(c.getFieldIndex())));
                }
            }
        }
    }
    for (int i = 0; i < rel.getGroupCount(); i++) {
        if (!aggregateColumnsOrder.contains(i)) {
            // Not included in the input collations, but can be propagated as this Aggregate
            // will enforce it
            propagateCollations.add(new RelFieldCollation(rel.getGroupSet().nth(i)));
        }
    }
    // 2) We propagate
    final RelNode child = dispatchAlign(rel.getInput(), propagateCollations.build());
    // 3) We annotate the Aggregate operator with this info
    final HiveAggregate newAggregate = (HiveAggregate) rel.copy(rel.getTraitSet(), ImmutableList.of(child));
    newAggregate.setAggregateColumnsOrder(aggregateColumnsOrder);
    return newAggregate;
}