Search in sources :

Example 51 with AggregateCall

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project flink by apache.

the class BatchExecPythonOverAggregate method translateToPlanInternal.

@SuppressWarnings("unchecked")
@Override
protected Transformation<RowData> translateToPlanInternal(PlannerBase planner, ExecNodeConfig config) {
    final ExecEdge inputEdge = getInputEdges().get(0);
    final Transformation<RowData> inputTransform = (Transformation<RowData>) inputEdge.translateToPlan(planner);
    final RowType inputType = (RowType) inputEdge.getOutputType();
    List<OverSpec.GroupSpec> groups = overSpec.getGroups();
    boolean[] isRangeWindows = new boolean[groups.size()];
    for (int i = 0; i < groups.size(); i++) {
        OverSpec.GroupSpec group = groups.get(i);
        List<AggregateCall> groupAggCalls = group.getAggCalls();
        aggCalls.addAll(groupAggCalls);
        for (int j = 0; j < groupAggCalls.size(); j++) {
            aggWindowIndex.add(i);
        }
        OverWindowMode mode = inferGroupMode(group);
        if (mode == OverWindowMode.ROW) {
            isRangeWindows[i] = false;
            if (isUnboundedWindow(group)) {
                lowerBoundary.add(Long.MIN_VALUE);
                upperBoundary.add(Long.MAX_VALUE);
            } else if (isUnboundedPrecedingWindow(group)) {
                lowerBoundary.add(Long.MIN_VALUE);
                upperBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getUpperBound()));
            } else if (isUnboundedFollowingWindow(group)) {
                lowerBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getLowerBound()));
                upperBoundary.add(Long.MAX_VALUE);
            } else if (isSlidingWindow(group)) {
                lowerBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getLowerBound()));
                upperBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getUpperBound()));
            } else {
                throw new TableException("Unsupported row window group spec " + group);
            }
        } else {
            isRangeWindows[i] = true;
            if (isUnboundedWindow(group)) {
                lowerBoundary.add(Long.MIN_VALUE);
                upperBoundary.add(Long.MAX_VALUE);
            } else if (isUnboundedPrecedingWindow(group)) {
                lowerBoundary.add(Long.MIN_VALUE);
                upperBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getUpperBound()));
            } else if (isUnboundedFollowingWindow(group)) {
                lowerBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getLowerBound()));
                upperBoundary.add(Long.MAX_VALUE);
            } else if (isSlidingWindow(group)) {
                lowerBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getLowerBound()));
                upperBoundary.add(OverAggregateUtil.getLongBoundary(overSpec, group.getUpperBound()));
            } else {
                throw new TableException("Unsupported range window group spec " + group);
            }
        }
    }
    Configuration pythonConfig = CommonPythonUtil.getMergedConfig(planner.getExecEnv(), config.getTableConfig());
    OneInputTransformation<RowData, RowData> transform = createPythonOneInputTransformation(inputTransform, inputType, InternalTypeInfo.of(getOutputType()).toRowType(), isRangeWindows, pythonConfig, config);
    if (CommonPythonUtil.isPythonWorkerUsingManagedMemory(pythonConfig)) {
        transform.declareManagedMemoryUseCaseAtSlotScope(ManagedMemoryUseCase.PYTHON);
    }
    return transform;
}
Also used : OneInputTransformation(org.apache.flink.streaming.api.transformations.OneInputTransformation) Transformation(org.apache.flink.api.dag.Transformation) TableException(org.apache.flink.table.api.TableException) ExecEdge(org.apache.flink.table.planner.plan.nodes.exec.ExecEdge) Configuration(org.apache.flink.configuration.Configuration) RowType(org.apache.flink.table.types.logical.RowType) OverSpec(org.apache.flink.table.planner.plan.nodes.exec.spec.OverSpec) AggregateCall(org.apache.calcite.rel.core.AggregateCall) RowData(org.apache.flink.table.data.RowData)

Example 52 with AggregateCall

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project flink by apache.

the class StreamPhysicalPythonGroupTableAggregateRule method matches.

@Override
public boolean matches(RelOptRuleCall call) {
    FlinkLogicalTableAggregate agg = call.rel(0);
    List<AggregateCall> aggCalls = agg.getAggCallList();
    boolean existGeneralPythonFunction = aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.GENERAL));
    boolean existPandasFunction = aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.PANDAS));
    boolean existJavaUserDefinedFunction = aggCalls.stream().anyMatch(x -> !PythonUtil.isPythonAggregate(x, null) && !PythonUtil.isBuiltInAggregate(x));
    if (existPandasFunction || existGeneralPythonFunction) {
        if (existPandasFunction) {
            throw new TableException("Pandas UDAFs are not supported in streaming mode currently.");
        }
        if (existJavaUserDefinedFunction) {
            throw new TableException("Python UDAF and Java/Scala UDAF cannot be used together.");
        }
        return true;
    } else {
        return false;
    }
}
Also used : AggregateCall(org.apache.calcite.rel.core.AggregateCall) TableException(org.apache.flink.table.api.TableException) FlinkLogicalTableAggregate(org.apache.flink.table.planner.plan.nodes.logical.FlinkLogicalTableAggregate)

Example 53 with AggregateCall

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project flink by apache.

the class StreamPhysicalPythonOverAggregateRule method matches.

@Override
public boolean matches(RelOptRuleCall call) {
    FlinkLogicalOverAggregate logicWindow = call.rel(0);
    List<AggregateCall> aggCalls = logicWindow.groups.get(0).getAggregateCalls(logicWindow);
    boolean existGeneralPythonFunction = aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.GENERAL));
    boolean existPandasFunction = aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.PANDAS));
    boolean existJavaFunction = aggCalls.stream().anyMatch(x -> !PythonUtil.isPythonAggregate(x, null));
    if (existPandasFunction || existGeneralPythonFunction) {
        if (existGeneralPythonFunction) {
            throw new TableException("Non-Pandas Python UDAFs are not supported in stream mode currently.");
        }
        if (existJavaFunction) {
            throw new TableException("Python UDAF and Java/Scala UDAF cannot be used together.");
        }
        return true;
    } else {
        return false;
    }
}
Also used : AggregateCall(org.apache.calcite.rel.core.AggregateCall) TableException(org.apache.flink.table.api.TableException) FlinkLogicalOverAggregate(org.apache.flink.table.planner.plan.nodes.logical.FlinkLogicalOverAggregate)

Example 54 with AggregateCall

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project flink by apache.

the class PushLocalAggIntoScanRuleBase method pushLocalAggregateIntoScan.

protected void pushLocalAggregateIntoScan(RelOptRuleCall call, BatchPhysicalGroupAggregateBase localAgg, BatchPhysicalTableSourceScan oldScan, int[] calcRefFields) {
    RowType inputType = FlinkTypeFactory.toLogicalRowType(oldScan.getRowType());
    List<int[]> groupingSets = Collections.singletonList(ArrayUtils.addAll(localAgg.grouping(), localAgg.auxGrouping()));
    List<AggregateCall> aggCallList = JavaScalaConversionUtil.toJava(localAgg.getAggCallList());
    // map arg index in aggregate to field index in scan through referred fields by calc.
    if (calcRefFields != null) {
        groupingSets = translateGroupingArgIndex(groupingSets, calcRefFields);
        aggCallList = translateAggCallArgIndex(aggCallList, calcRefFields);
    }
    RowType producedType = FlinkTypeFactory.toLogicalRowType(localAgg.getRowType());
    TableSourceTable oldTableSourceTable = oldScan.tableSourceTable();
    DynamicTableSource newTableSource = oldScan.tableSource().copy();
    boolean isPushDownSuccess = AggregatePushDownSpec.apply(inputType, groupingSets, aggCallList, producedType, newTableSource, SourceAbilityContext.from(oldScan));
    if (!isPushDownSuccess) {
        // aggregate push down failed, just return without changing any nodes.
        return;
    }
    // create new source table with new spec and statistic.
    AggregatePushDownSpec aggregatePushDownSpec = new AggregatePushDownSpec(inputType, groupingSets, aggCallList, producedType);
    TableSourceTable newTableSourceTable = oldTableSourceTable.copy(newTableSource, localAgg.getRowType(), new SourceAbilitySpec[] { aggregatePushDownSpec }).copy(FlinkStatistic.UNKNOWN());
    // transform to new nodes.
    BatchPhysicalTableSourceScan newScan = oldScan.copy(oldScan.getTraitSet(), newTableSourceTable);
    BatchPhysicalExchange oldExchange = call.rel(0);
    BatchPhysicalExchange newExchange = oldExchange.copy(oldExchange.getTraitSet(), newScan, oldExchange.getDistribution());
    call.transformTo(newExchange);
}
Also used : AggregateCall(org.apache.calcite.rel.core.AggregateCall) SourceAbilitySpec(org.apache.flink.table.planner.plan.abilities.source.SourceAbilitySpec) AggregatePushDownSpec(org.apache.flink.table.planner.plan.abilities.source.AggregatePushDownSpec) RowType(org.apache.flink.table.types.logical.RowType) TableSourceTable(org.apache.flink.table.planner.plan.schema.TableSourceTable) BatchPhysicalExchange(org.apache.flink.table.planner.plan.nodes.physical.batch.BatchPhysicalExchange) DynamicTableSource(org.apache.flink.table.connector.source.DynamicTableSource) BatchPhysicalTableSourceScan(org.apache.flink.table.planner.plan.nodes.physical.batch.BatchPhysicalTableSourceScan)

Example 55 with AggregateCall

use of org.apache.beam.vendor.calcite.v1_28_0.org.apache.calcite.rel.core.AggregateCall in project flink by apache.

the class StreamPhysicalPythonGroupWindowAggregateRule method matches.

@Override
public boolean matches(RelOptRuleCall call) {
    FlinkLogicalWindowAggregate agg = call.rel(0);
    List<AggregateCall> aggCalls = agg.getAggCallList();
    // check if we have grouping sets
    if (agg.getGroupType() != Aggregate.Group.SIMPLE || agg.indicator) {
        throw new TableException("GROUPING SETS are currently not supported.");
    }
    boolean existGeneralPythonFunction = aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.GENERAL));
    boolean existPandasFunction = aggCalls.stream().anyMatch(x -> PythonUtil.isPythonAggregate(x, PythonFunctionKind.PANDAS));
    boolean existJavaFunction = aggCalls.stream().anyMatch(x -> !PythonUtil.isPythonAggregate(x, null) && !PythonUtil.isBuiltInAggregate(x));
    if (existPandasFunction && existGeneralPythonFunction) {
        throw new TableException("Pandas UDAFs and General Python UDAFs are not supported in used together currently.");
    }
    if (existPandasFunction || existGeneralPythonFunction) {
        if (existJavaFunction) {
            throw new TableException("Python UDAF and Java/Scala UDAF cannot be used together.");
        }
        return true;
    } else {
        return false;
    }
}
Also used : AggregateCall(org.apache.calcite.rel.core.AggregateCall) TableException(org.apache.flink.table.api.TableException) FlinkLogicalWindowAggregate(org.apache.flink.table.planner.plan.nodes.logical.FlinkLogicalWindowAggregate)

Aggregations

AggregateCall (org.apache.calcite.rel.core.AggregateCall)158 ArrayList (java.util.ArrayList)82 RexNode (org.apache.calcite.rex.RexNode)78 ImmutableBitSet (org.apache.calcite.util.ImmutableBitSet)57 RelNode (org.apache.calcite.rel.RelNode)54 RexBuilder (org.apache.calcite.rex.RexBuilder)52 RelDataType (org.apache.calcite.rel.type.RelDataType)42 Aggregate (org.apache.calcite.rel.core.Aggregate)37 RelDataTypeField (org.apache.calcite.rel.type.RelDataTypeField)36 RexInputRef (org.apache.calcite.rex.RexInputRef)33 RelBuilder (org.apache.calcite.tools.RelBuilder)29 HashMap (java.util.HashMap)28 SqlAggFunction (org.apache.calcite.sql.SqlAggFunction)28 List (java.util.List)27 RexLiteral (org.apache.calcite.rex.RexLiteral)23 Pair (org.apache.calcite.util.Pair)20 ImmutableList (com.google.common.collect.ImmutableList)19 Project (org.apache.calcite.rel.core.Project)17 RelDataTypeFactory (org.apache.calcite.rel.type.RelDataTypeFactory)17 LogicalAggregate (org.apache.calcite.rel.logical.LogicalAggregate)16