use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project drill by axbaretto.
the class AggPrelBase method createKeysAndExprs.
protected void createKeysAndExprs() {
final List<String> childFields = getInput().getRowType().getFieldNames();
final List<String> fields = getRowType().getFieldNames();
for (int group : BitSets.toIter(groupSet)) {
FieldReference fr = FieldReference.getWithQuotedRef(childFields.get(group));
keys.add(new NamedExpression(fr, fr));
}
for (Ord<AggregateCall> aggCall : Ord.zip(aggCalls)) {
int aggExprOrdinal = groupSet.cardinality() + aggCall.i;
FieldReference ref = FieldReference.getWithQuotedRef(fields.get(aggExprOrdinal));
LogicalExpression expr = toDrill(aggCall.e, childFields);
NamedExpression ne = new NamedExpression(expr, ref);
aggExprs.add(ne);
if (getOperatorPhase() == OperatorPhase.PHASE_1of2) {
if (aggCall.e.getAggregation().getName().equals("COUNT")) {
// If we are doing a COUNT aggregate in Phase1of2, then in Phase2of2 we should SUM the COUNTs,
SqlAggFunction sumAggFun = new SqlSumCountAggFunction(aggCall.e.getType());
AggregateCall newAggCall = AggregateCall.create(sumAggFun, aggCall.e.isDistinct(), aggCall.e.isApproximate(), Collections.singletonList(aggExprOrdinal), aggCall.e.filterArg, aggCall.e.getType(), aggCall.e.getName());
phase2AggCallList.add(newAggCall);
} else {
AggregateCall newAggCall = AggregateCall.create(aggCall.e.getAggregation(), aggCall.e.isDistinct(), aggCall.e.isApproximate(), Collections.singletonList(aggExprOrdinal), aggCall.e.filterArg, aggCall.e.getType(), aggCall.e.getName());
phase2AggCallList.add(newAggCall);
}
}
}
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project drill by axbaretto.
the class AggPruleBase method create2PhasePlan.
// Create 2 phase aggr plan for aggregates such as SUM, MIN, MAX
// If any of the aggregate functions are not one of these, then we
// currently won't generate a 2 phase plan.
protected boolean create2PhasePlan(RelOptRuleCall call, DrillAggregateRel aggregate) {
PlannerSettings settings = PrelUtil.getPlannerSettings(call.getPlanner());
RelNode child = call.rel(0).getInputs().get(0);
boolean smallInput = child.estimateRowCount(child.getCluster().getMetadataQuery()) < settings.getSliceTarget();
if (!settings.isMultiPhaseAggEnabled() || settings.isSingleMode() || // Can override a small child - e.g., for testing with a small table
(smallInput && !settings.isForce2phaseAggr())) {
return false;
}
for (AggregateCall aggCall : aggregate.getAggCallList()) {
String name = aggCall.getAggregation().getName();
if (!(name.equals(SqlKind.SUM.name()) || name.equals(SqlKind.MIN.name()) || name.equals(SqlKind.MAX.name()) || name.equals(SqlKind.COUNT.name()) || name.equals("$SUM0"))) {
return false;
}
}
return true;
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project drill by axbaretto.
the class ConvertCountToDirectScan method collectCounts.
/**
* Collects counts for each aggregation call.
* Will return empty result map if was not able to determine count for at least one aggregation call,
*
* For each aggregate call will determine if count can be calculated. Collects counts only for COUNT function.
* For star, not null expressions and implicit columns sets count to total record number.
* For other cases obtains counts from group scan operator. Also count can not be calculated for parition columns.
*
* @param agg aggregate relational expression
* @param scan scan relational expression
* @param project project relational expression
* @return result map where key is count column name, value is count value
*/
private Map<String, Long> collectCounts(PlannerSettings settings, DrillAggregateRel agg, DrillScanRel scan, DrillProjectRel project) {
final Set<String> implicitColumnsNames = ColumnExplorer.initImplicitFileColumns(settings.getOptions()).keySet();
final GroupScan oldGrpScan = scan.getGroupScan();
final long totalRecordCount = oldGrpScan.getScanStats(settings).getRecordCount();
final LinkedHashMap<String, Long> result = new LinkedHashMap<>();
for (int i = 0; i < agg.getAggCallList().size(); i++) {
AggregateCall aggCall = agg.getAggCallList().get(i);
// for (AggregateCall aggCall : agg.getAggCallList()) {
long cnt;
// rule can be applied only for count function, return empty counts
if (!"count".equalsIgnoreCase(aggCall.getAggregation().getName())) {
return ImmutableMap.of();
}
if (containsStarOrNotNullInput(aggCall, agg)) {
cnt = totalRecordCount;
} else if (aggCall.getArgList().size() == 1) {
// count(columnName) ==> Agg ( Scan )) ==> columnValueCount
int index = aggCall.getArgList().get(0);
if (project != null) {
// return count of "col2" in Scan's metadata, if found.
if (!(project.getProjects().get(index) instanceof RexInputRef)) {
// do not apply for all other cases.
return ImmutableMap.of();
}
index = ((RexInputRef) project.getProjects().get(index)).getIndex();
}
String columnName = scan.getRowType().getFieldNames().get(index).toLowerCase();
// for implicit column count will the same as total record count
if (implicitColumnsNames.contains(columnName)) {
cnt = totalRecordCount;
} else {
SchemaPath simplePath = SchemaPath.getSimplePath(columnName);
if (ColumnExplorer.isPartitionColumn(settings.getOptions(), simplePath)) {
return ImmutableMap.of();
}
cnt = oldGrpScan.getColumnValueCount(simplePath);
if (cnt == GroupScan.NO_COLUMN_STATS) {
// if column stats is not available don't apply this rule, return empty counts
return ImmutableMap.of();
}
}
} else {
return ImmutableMap.of();
}
String name = "count" + i + "$" + (aggCall.getName() == null ? aggCall.toString() : aggCall.getName());
result.put(name, cnt);
}
return ImmutableMap.copyOf(result);
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project drill by axbaretto.
the class DrillReduceAggregatesRule method reduceSum.
private RexNode reduceSum(Aggregate oldAggRel, AggregateCall oldCall, List<AggregateCall> newCalls, Map<AggregateCall, RexNode> aggCallMapping) {
final PlannerSettings plannerSettings = (PlannerSettings) oldAggRel.getCluster().getPlanner().getContext();
final boolean isInferenceEnabled = plannerSettings.isTypeInferenceEnabled();
final int nGroups = oldAggRel.getGroupCount();
RelDataTypeFactory typeFactory = oldAggRel.getCluster().getTypeFactory();
RexBuilder rexBuilder = oldAggRel.getCluster().getRexBuilder();
int arg = oldCall.getArgList().get(0);
RelDataType argType = getFieldType(oldAggRel.getInput(), arg);
final RelDataType sumType;
final SqlAggFunction sumZeroAgg;
if (isInferenceEnabled) {
sumType = oldCall.getType();
} else {
sumType = typeFactory.createTypeWithNullability(oldCall.getType(), argType.isNullable());
}
sumZeroAgg = new DrillCalciteSqlAggFunctionWrapper(new SqlSumEmptyIsZeroAggFunction(), sumType);
AggregateCall sumZeroCall = AggregateCall.create(sumZeroAgg, oldCall.isDistinct(), oldCall.isApproximate(), oldCall.getArgList(), -1, sumType, null);
final SqlCountAggFunction countAgg = (SqlCountAggFunction) SqlStdOperatorTable.COUNT;
final RelDataType countType = countAgg.getReturnType(typeFactory);
AggregateCall countCall = AggregateCall.create(countAgg, oldCall.isDistinct(), oldCall.isApproximate(), oldCall.getArgList(), -1, countType, null);
// NOTE: these references are with respect to the output
// of newAggRel
RexNode sumZeroRef = rexBuilder.addAggCall(sumZeroCall, nGroups, oldAggRel.indicator, newCalls, aggCallMapping, ImmutableList.of(argType));
if (!oldCall.getType().isNullable()) {
// null). Therefore we translate to SUM0(x).
return sumZeroRef;
}
RexNode countRef = rexBuilder.addAggCall(countCall, nGroups, oldAggRel.indicator, newCalls, aggCallMapping, ImmutableList.of(argType));
return rexBuilder.makeCall(SqlStdOperatorTable.CASE, rexBuilder.makeCall(SqlStdOperatorTable.EQUALS, countRef, rexBuilder.makeExactLiteral(BigDecimal.ZERO)), rexBuilder.constantNull(), sumZeroRef);
}
use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project drill by axbaretto.
the class DrillWindowRel method implement.
@Override
public LogicalOperator implement(DrillImplementor implementor) {
final LogicalOperator inputOp = implementor.visitChild(this, 0, getInput());
org.apache.drill.common.logical.data.Window.Builder builder = new org.apache.drill.common.logical.data.Window.Builder();
final List<String> fields = getRowType().getFieldNames();
final List<String> childFields = getInput().getRowType().getFieldNames();
for (Group window : groups) {
for (RelFieldCollation orderKey : window.orderKeys.getFieldCollations()) {
builder.addOrdering(new Order.Ordering(orderKey.getDirection(), new FieldReference(fields.get(orderKey.getFieldIndex()))));
}
for (int group : BitSets.toIter(window.keys)) {
FieldReference fr = new FieldReference(childFields.get(group), ExpressionPosition.UNKNOWN);
builder.addWithin(fr, fr);
}
int groupCardinality = window.keys.cardinality();
for (Ord<AggregateCall> aggCall : Ord.zip(window.getAggregateCalls(this))) {
FieldReference ref = new FieldReference(fields.get(groupCardinality + aggCall.i));
LogicalExpression expr = toDrill(aggCall.e, childFields);
builder.addAggregation(ref, expr);
}
}
builder.setInput(inputOp);
org.apache.drill.common.logical.data.Window frame = builder.build();
return frame;
}
Aggregations