use of org.apache.calcite.rel.type.RelDataTypeField in project hive by apache.
the class HiveAggregate method deriveRowType.
public static RelDataType deriveRowType(RelDataTypeFactory typeFactory, final RelDataType inputRowType, boolean indicator, ImmutableBitSet groupSet, List<ImmutableBitSet> groupSets, final List<AggregateCall> aggCalls) {
final List<Integer> groupList = groupSet.asList();
assert groupList.size() == groupSet.cardinality();
final RelDataTypeFactory.FieldInfoBuilder builder = typeFactory.builder();
final List<RelDataTypeField> fieldList = inputRowType.getFieldList();
final Set<String> containedNames = Sets.newHashSet();
for (int groupKey : groupList) {
containedNames.add(fieldList.get(groupKey).getName());
builder.add(fieldList.get(groupKey));
}
if (indicator) {
for (int groupKey : groupList) {
final RelDataType booleanType = typeFactory.createTypeWithNullability(typeFactory.createSqlType(SqlTypeName.BOOLEAN), false);
String name = "i$" + fieldList.get(groupKey).getName();
int i = 0;
while (containedNames.contains(name)) {
name += "_" + i++;
}
containedNames.add(name);
builder.add(name, booleanType);
}
}
for (Ord<AggregateCall> aggCall : Ord.zip(aggCalls)) {
String name;
if (aggCall.e.name != null) {
name = aggCall.e.name;
} else {
name = "$f" + (groupList.size() + aggCall.i);
}
int i = 0;
while (containedNames.contains(name)) {
name += "_" + i++;
}
containedNames.add(name);
builder.add(name, aggCall.e.type);
}
return builder.build();
}
use of org.apache.calcite.rel.type.RelDataTypeField in project hive by apache.
the class HiveJoinToMultiJoinRule method mergeJoin.
// This method tries to merge the join with its left child. The left
// child should be a join for this to happen.
private static RelNode mergeJoin(HiveJoin join, RelNode left, RelNode right) {
final RexBuilder rexBuilder = join.getCluster().getRexBuilder();
// We check whether the join can be combined with any of its children
final List<RelNode> newInputs = Lists.newArrayList();
final List<RexNode> newJoinCondition = Lists.newArrayList();
final List<Pair<Integer, Integer>> joinInputs = Lists.newArrayList();
final List<JoinRelType> joinTypes = Lists.newArrayList();
final List<RexNode> joinFilters = Lists.newArrayList();
// Left child
if (left instanceof HiveJoin || left instanceof HiveMultiJoin) {
final RexNode leftCondition;
final List<Pair<Integer, Integer>> leftJoinInputs;
final List<JoinRelType> leftJoinTypes;
final List<RexNode> leftJoinFilters;
boolean combinable;
if (left instanceof HiveJoin) {
HiveJoin hj = (HiveJoin) left;
leftCondition = hj.getCondition();
leftJoinInputs = ImmutableList.of(Pair.of(0, 1));
leftJoinTypes = ImmutableList.of(hj.getJoinType());
leftJoinFilters = ImmutableList.of(hj.getJoinFilter());
try {
combinable = isCombinableJoin(join, hj);
} catch (CalciteSemanticException e) {
LOG.trace("Failed to merge join-join", e);
combinable = false;
}
} else {
HiveMultiJoin hmj = (HiveMultiJoin) left;
leftCondition = hmj.getCondition();
leftJoinInputs = hmj.getJoinInputs();
leftJoinTypes = hmj.getJoinTypes();
leftJoinFilters = hmj.getJoinFilters();
try {
combinable = isCombinableJoin(join, hmj);
} catch (CalciteSemanticException e) {
LOG.trace("Failed to merge join-multijoin", e);
combinable = false;
}
}
if (combinable) {
newJoinCondition.add(leftCondition);
for (int i = 0; i < leftJoinInputs.size(); i++) {
joinInputs.add(leftJoinInputs.get(i));
joinTypes.add(leftJoinTypes.get(i));
joinFilters.add(leftJoinFilters.get(i));
}
newInputs.addAll(left.getInputs());
} else {
// The join operation in the child is not on the same keys
return null;
}
} else {
// The left child is not a join or multijoin operator
return null;
}
final int numberLeftInputs = newInputs.size();
// Right child
newInputs.add(right);
// If we cannot combine any of the children, we bail out
newJoinCondition.add(join.getCondition());
if (newJoinCondition.size() == 1) {
return null;
}
final List<RelDataTypeField> systemFieldList = ImmutableList.of();
List<List<RexNode>> joinKeyExprs = new ArrayList<List<RexNode>>();
List<Integer> filterNulls = new ArrayList<Integer>();
for (int i = 0; i < newInputs.size(); i++) {
joinKeyExprs.add(new ArrayList<RexNode>());
}
RexNode filters;
try {
filters = HiveRelOptUtil.splitHiveJoinCondition(systemFieldList, newInputs, join.getCondition(), joinKeyExprs, filterNulls, null);
} catch (CalciteSemanticException e) {
LOG.trace("Failed to merge joins", e);
return null;
}
ImmutableBitSet.Builder keysInInputsBuilder = ImmutableBitSet.builder();
for (int i = 0; i < newInputs.size(); i++) {
List<RexNode> partialCondition = joinKeyExprs.get(i);
if (!partialCondition.isEmpty()) {
keysInInputsBuilder.set(i);
}
}
// If we cannot merge, we bail out
ImmutableBitSet keysInInputs = keysInInputsBuilder.build();
ImmutableBitSet leftReferencedInputs = keysInInputs.intersect(ImmutableBitSet.range(numberLeftInputs));
ImmutableBitSet rightReferencedInputs = keysInInputs.intersect(ImmutableBitSet.range(numberLeftInputs, newInputs.size()));
if (join.getJoinType() != JoinRelType.INNER && (leftReferencedInputs.cardinality() > 1 || rightReferencedInputs.cardinality() > 1)) {
return null;
}
// Otherwise, we add to the join specs
if (join.getJoinType() != JoinRelType.INNER) {
int leftInput = keysInInputs.nextSetBit(0);
int rightInput = keysInInputs.nextSetBit(numberLeftInputs);
joinInputs.add(Pair.of(leftInput, rightInput));
joinTypes.add(join.getJoinType());
joinFilters.add(filters);
} else {
for (int i : leftReferencedInputs) {
for (int j : rightReferencedInputs) {
joinInputs.add(Pair.of(i, j));
joinTypes.add(join.getJoinType());
joinFilters.add(filters);
}
}
}
// We can now create a multijoin operator
RexNode newCondition = RexUtil.flatten(rexBuilder, RexUtil.composeConjunction(rexBuilder, newJoinCondition, false));
List<RelNode> newInputsArray = Lists.newArrayList(newInputs);
JoinPredicateInfo joinPredInfo = null;
try {
joinPredInfo = HiveCalciteUtil.JoinPredicateInfo.constructJoinPredicateInfo(newInputsArray, systemFieldList, newCondition);
} catch (CalciteSemanticException e) {
throw new RuntimeException(e);
}
// If the number of joins < number of input tables-1, this is not a star join.
if (joinPredInfo.getEquiJoinPredicateElements().size() < newInputs.size() - 1) {
return null;
}
// Validate that the multi-join is a valid star join before returning it.
for (int i = 0; i < newInputs.size(); i++) {
List<RexNode> joinKeys = null;
for (int j = 0; j < joinPredInfo.getEquiJoinPredicateElements().size(); j++) {
List<RexNode> currJoinKeys = joinPredInfo.getEquiJoinPredicateElements().get(j).getJoinExprs(i);
if (currJoinKeys.isEmpty()) {
continue;
}
if (joinKeys == null) {
joinKeys = currJoinKeys;
} else {
// Bail out if this is the case.
if (!joinKeys.containsAll(currJoinKeys) || !currJoinKeys.containsAll(joinKeys)) {
return null;
}
}
}
}
return new HiveMultiJoin(join.getCluster(), newInputsArray, newCondition, join.getRowType(), joinInputs, joinTypes, joinFilters, joinPredInfo);
}
use of org.apache.calcite.rel.type.RelDataTypeField in project hive by apache.
the class HiveProject method projectMapping.
/**
* Creates a relational expression which projects the output fields of a
* relational expression according to a partial mapping.
*
* <p>
* A partial mapping is weaker than a permutation: every target has one
* source, but a source may have 0, 1 or more than one targets. Usually the
* result will have fewer fields than the source, unless some source fields
* are projected multiple times.
*
* <p>
* This method could optimize the result as {@link #permute} does, but does
* not at present.
*
* @param rel
* Relational expression
* @param mapping
* Mapping from source fields to target fields. The mapping type must
* obey the constraints {@link MappingType#isMandatorySource()} and
* {@link MappingType#isSingleSource()}, as does
* {@link MappingType#INVERSE_FUNCTION}.
* @param fieldNames
* Field names; if null, or if a particular entry is null, the name
* of the permuted field is used
* @return relational expression which projects a subset of the input fields
* @throws CalciteSemanticException
*/
public static RelNode projectMapping(RelNode rel, Mapping mapping, List<String> fieldNames) throws CalciteSemanticException {
assert mapping.getMappingType().isSingleSource();
assert mapping.getMappingType().isMandatorySource();
if (mapping.isIdentity()) {
return rel;
}
final List<String> outputNameList = new ArrayList<String>();
final List<RexNode> outputProjList = new ArrayList<RexNode>();
final List<RelDataTypeField> fields = rel.getRowType().getFieldList();
final RexBuilder rexBuilder = rel.getCluster().getRexBuilder();
for (int i = 0; i < mapping.getTargetCount(); i++) {
int source = mapping.getSource(i);
final RelDataTypeField sourceField = fields.get(source);
outputNameList.add(((fieldNames == null) || (fieldNames.size() <= i) || (fieldNames.get(i) == null)) ? sourceField.getName() : fieldNames.get(i));
outputProjList.add(rexBuilder.makeInputRef(rel, source));
}
return create(rel, outputProjList, outputNameList);
}
use of org.apache.calcite.rel.type.RelDataTypeField in project hive by apache.
the class HiveCalciteUtil method createUDTFForSetOp.
public static HiveTableFunctionScan createUDTFForSetOp(RelOptCluster cluster, RelNode input) throws SemanticException {
RelTraitSet traitSet = TraitsUtil.getDefaultTraitSet(cluster);
List<RexNode> originalInputRefs = Lists.transform(input.getRowType().getFieldList(), new Function<RelDataTypeField, RexNode>() {
@Override
public RexNode apply(RelDataTypeField input) {
return new RexInputRef(input.getIndex(), input.getType());
}
});
ImmutableList.Builder<RelDataType> argTypeBldr = ImmutableList.<RelDataType>builder();
for (int i = 0; i < originalInputRefs.size(); i++) {
argTypeBldr.add(originalInputRefs.get(i).getType());
}
RelDataType retType = input.getRowType();
String funcName = "replicate_rows";
FunctionInfo fi = FunctionRegistry.getFunctionInfo(funcName);
SqlOperator calciteOp = SqlFunctionConverter.getCalciteOperator(funcName, fi.getGenericUDTF(), argTypeBldr.build(), retType);
// Hive UDTF only has a single input
List<RelNode> list = new ArrayList<>();
list.add(input);
RexNode rexNode = cluster.getRexBuilder().makeCall(calciteOp, originalInputRefs);
return HiveTableFunctionScan.create(cluster, traitSet, list, rexNode, null, retType, null);
}
use of org.apache.calcite.rel.type.RelDataTypeField in project druid by druid-io.
the class QueryMaker method executeTopN.
private Sequence<Object[]> executeTopN(final DruidQueryBuilder queryBuilder, final TopNQuery query) {
final List<RelDataTypeField> fieldList = queryBuilder.getRowType().getFieldList();
Hook.QUERY_PLAN.run(query);
return Sequences.concat(Sequences.map(query.run(walker, Maps.<String, Object>newHashMap()), new Function<Result<TopNResultValue>, Sequence<Object[]>>() {
@Override
public Sequence<Object[]> apply(final Result<TopNResultValue> result) {
final List<DimensionAndMetricValueExtractor> rows = result.getValue().getValue();
final List<Object[]> retVals = new ArrayList<>(rows.size());
for (DimensionAndMetricValueExtractor row : rows) {
final Object[] retVal = new Object[fieldList.size()];
for (final RelDataTypeField field : fieldList) {
final String outputName = queryBuilder.getRowOrder().get(field.getIndex());
retVal[field.getIndex()] = coerce(row.getMetric(outputName), field.getType().getSqlTypeName());
}
retVals.add(retVal);
}
return Sequences.simple(retVals);
}
}));
}
Aggregations