KurtYoung commented on a change in pull request #8092: [FLINK-12076] [table-planner-blink] Add support for generating optimized logical plan for simple group aggregate on batch URL: https://github.com/apache/flink/pull/8092#discussion_r271111334
########## File path: flink-table/flink-table-planner-blink/src/main/scala/org/apache/flink/table/plan/rules/physical/batch/BatchExecAggRuleBase.scala ########## @@ -0,0 +1,176 @@ +/* + * Licensed to the Apache Software Foundation (ASF) under one + * or more contributor license agreements. See the NOTICE file + * distributed with this work for additional information + * regarding copyright ownership. The ASF licenses this file + * to you under the Apache License, Version 2.0 (the + * "License"); you may not use this file except in compliance + * with the License. You may obtain a copy of the License at + * + * http://www.apache.org/licenses/LICENSE-2.0 + * + * Unless required by applicable law or agreed to in writing, software + * distributed under the License is distributed on an "AS IS" BASIS, + * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied. + * See the License for the specific language governing permissions and + * limitations under the License. + */ +package org.apache.flink.table.plan.rules.physical.batch + +import org.apache.flink.table.JArrayList +import org.apache.flink.table.`type`.{InternalType, TypeConverters} +import org.apache.flink.table.api.{AggPhaseEnforcer, PlannerConfigOptions, TableConfig, TableException} +import org.apache.flink.table.calcite.FlinkTypeFactory +import org.apache.flink.table.dataformat.BinaryRow +import org.apache.flink.table.functions.utils.UserDefinedFunctionUtils._ +import org.apache.flink.table.functions.{AggregateFunction, DeclarativeAggregateFunction, UserDefinedFunction} +import org.apache.flink.table.plan.util.{AggregateUtil, FlinkRelOptUtil, RelFieldCollationUtil} + +import org.apache.calcite.rel.`type`.RelDataType +import org.apache.calcite.rel.core.Aggregate +import org.apache.calcite.rel.{RelCollation, RelCollations, RelFieldCollation} +import org.apache.calcite.util.Util + +import scala.collection.JavaConversions._ + +trait BatchExecAggRuleBase { + + protected def inferLocalAggType( + inputRowType: RelDataType, + agg: Aggregate, + groupSet: Array[Int], + auxGroupSet: Array[Int], + aggFunctions: Array[UserDefinedFunction], + aggBufferTypes: Array[Array[InternalType]]): RelDataType = { + + val typeFactory = agg.getCluster.getTypeFactory.asInstanceOf[FlinkTypeFactory] + val aggCallNames = Util.skip( + agg.getRowType.getFieldNames, groupSet.length + auxGroupSet.length).toList.toArray[String] + inferLocalAggType( + inputRowType, typeFactory, aggCallNames, groupSet, auxGroupSet, aggFunctions, aggBufferTypes) + } + + protected def inferLocalAggType( + inputRowType: RelDataType, + typeFactory: FlinkTypeFactory, + aggCallNames: Array[String], + groupSet: Array[Int], + auxGroupSet: Array[Int], + aggFunctions: Array[UserDefinedFunction], + aggBufferTypes: Array[Array[InternalType]]): RelDataType = { + + val aggBufferFieldNames = new Array[Array[String]](aggFunctions.length) + var index = -1 + aggFunctions.zipWithIndex.foreach { + case (udf, aggIndex) => + aggBufferFieldNames(aggIndex) = udf match { + case _: AggregateFunction[_, _] => + Array(aggCallNames(aggIndex)) + case agf: DeclarativeAggregateFunction => + agf.aggBufferAttributes.map { attr => + index += 1 + s"${attr.getName}$$$index" + } + case _: UserDefinedFunction => + throw new TableException(s"Don't get localAgg merge name") + } + } + + // local agg output order: groupSet + auxGroupSet + aggCalls + val aggBufferSqlTypes = aggBufferTypes.flatten.map { t => + val nullable = !FlinkTypeFactory.isTimeIndicatorType(t) + typeFactory.createTypeFromInternalType(t, nullable) + } + + val localAggFieldTypes = ( + groupSet.map(inputRowType.getFieldList.get(_).getType) ++ // groupSet + auxGroupSet.map(inputRowType.getFieldList.get(_).getType) ++ // auxGroupSet + aggBufferSqlTypes // aggCalls + ).toList + + val localAggFieldNames = ( + groupSet.map(inputRowType.getFieldList.get(_).getName) ++ // groupSet + auxGroupSet.map(inputRowType.getFieldList.get(_).getName) ++ // auxGroupSet + aggBufferFieldNames.flatten.toArray[String] // aggCalls + ).toList + + typeFactory.createStructType(localAggFieldTypes, localAggFieldNames) + } + + protected def isTwoPhaseAggWorkable( + aggFunctions: Array[UserDefinedFunction], + tableConfig: TableConfig): Boolean = { + getAggEnforceStrategy(tableConfig) match { + case AggPhaseEnforcer.ONE_PHASE => false + case _ => doAllSupportMerge(aggFunctions) + } + } + + protected def isOnePhaseAggWorkable( + agg: Aggregate, + aggFunctions: Array[UserDefinedFunction], + tableConfig: TableConfig): Boolean = { + getAggEnforceStrategy(tableConfig) match { + case AggPhaseEnforcer.ONE_PHASE => true + case AggPhaseEnforcer.TWO_PHASE => !doAllSupportMerge(aggFunctions) + case AggPhaseEnforcer.NONE => + if (!doAllSupportMerge(aggFunctions)) { + true + } else { + // if ndv of group key in aggregate is Unknown and all aggFunctions are splittable, + // use two-phase agg. + // else whether choose one-phase agg or two-phase agg depends on CBO. + val mq = agg.getCluster.getMetadataQuery + mq.getDistinctRowCount(agg.getInput, agg.getGroupSet, null) != null + } + } + } + + protected def doAllSupportMerge(aggFunctions: Array[UserDefinedFunction]): Boolean = { + val supportLocalAgg = aggFunctions.forall { + case _: DeclarativeAggregateFunction => true + case a => ifMethodExistInFunction("merge", a) + } + //it means grouping without aggregate functions + aggFunctions.isEmpty || supportLocalAgg + } + + protected def isEnforceOnePhaseAgg(tableConfig: TableConfig): Boolean = { + getAggEnforceStrategy(tableConfig) == AggPhaseEnforcer.ONE_PHASE + } + + protected def isEnforceTwoPhaseAgg(tableConfig: TableConfig): Boolean = { + getAggEnforceStrategy(tableConfig) == AggPhaseEnforcer.TWO_PHASE + } + + protected def getAggEnforceStrategy(tableConfig: TableConfig): AggPhaseEnforcer.Value = { + val aggPrefConfig = tableConfig.getConf.getString( + PlannerConfigOptions.SQL_OPTIMIZER_AGG_PHASE_ENFORCER) + AggPhaseEnforcer.values.find(_.toString.equalsIgnoreCase(aggPrefConfig)) + .getOrElse(throw new IllegalArgumentException( + "Agg phase enforcer can only set to be: NONE, ONE_PHASE, TWO_PHASE!")) + } + + protected def isAggBufferFixedLength(agg: Aggregate): Boolean = { + val (_, aggCallsWithoutAuxGroupCalls) = FlinkRelOptUtil.checkAndSplitAggCalls(agg) + val (_, aggBufferTypes, _) = AggregateUtil.transformToBatchAggregateFunctions( + aggCallsWithoutAuxGroupCalls, agg.getInput.getRowType) + + isAggBufferFixedLength(aggBufferTypes.map(_.map(TypeConverters.createInternalTypeFromTypeInfo))) + } + + protected def isAggBufferFixedLength(aggBufferTypes: Array[Array[InternalType]]): Boolean = { + val aggBuffAttributesTypes = aggBufferTypes.flatten + val isAggBufferFixedLength = aggBuffAttributesTypes.forall(BinaryRow.isMutable) Review comment: add a TODO to check `Decimal` can be applied to hash agg, like `sum(decimal)` ---------------------------------------------------------------- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. For queries about this service, please contact Infrastructure at: us...@infra.apache.org With regards, Apache Git Services
