Github user cloud-fan commented on a diff in the pull request: https://github.com/apache/spark/pull/13494#discussion_r68580134 --- Diff: sql/core/src/main/scala/org/apache/spark/sql/execution/MetadataOnlyOptimizer.scala --- @@ -0,0 +1,190 @@ +/* + * 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.spark.sql.execution + +import org.apache.spark.sql.{AnalysisException, SparkSession} +import org.apache.spark.sql.catalyst.InternalRow +import org.apache.spark.sql.catalyst.catalog.{CatalogRelation, SessionCatalog} +import org.apache.spark.sql.catalyst.expressions._ +import org.apache.spark.sql.catalyst.expressions.aggregate._ +import org.apache.spark.sql.catalyst.plans.logical._ +import org.apache.spark.sql.catalyst.rules.Rule +import org.apache.spark.sql.execution.datasources.{HadoopFsRelation, LogicalRelation} + +/** + * When scanning only partition columns, get results based on metadata without scanning files. + * It is used for distinct, distinct aggregations or distinct-like aggregations(example: Max/Min). + * Example: select Max(partition) from table. + */ +case class MetadataOnlyOptimizer( + sparkSession: SparkSession, + catalog: SessionCatalog) extends Rule[LogicalPlan] { + + private def canSupportMetadataOnly(a: Aggregate): Boolean = { + val aggregateExpressions = a.aggregateExpressions.flatMap { expr => + expr.collect { + case agg: AggregateExpression => agg + } + }.distinct + if (aggregateExpressions.isEmpty) { + // Cannot support for aggregate that has no aggregateFunction. + // example: select col1 from table group by col1. + false + } else { + aggregateExpressions.forall { agg => + if (agg.isDistinct) { + true + } else { + // If function can be evaluated on just the distinct values of a column, it can be used + // by metadata-only optimizer. + agg.aggregateFunction match { + case max: Max => true + case min: Min => true + case hyperLog: HyperLogLogPlusPlus => true + case _ => false + } + } + } + } + } + + private def collectAliases(fields: Seq[Expression]): Map[ExprId, Expression] = fields.collect { + case a @ Alias(child, _) => a.toAttribute.exprId -> child + }.toMap + + private def substitute(aliases: Map[ExprId, Expression])(expr: Expression): Expression = { + expr.transform { + case a @ Alias(ref: AttributeReference, name) => + aliases.get(ref.exprId) + .map(Alias(_, name)(a.exprId, a.qualifier, isGenerated = a.isGenerated)) + .getOrElse(a) + + case a: AttributeReference => + aliases.get(a.exprId) + .map(Alias(_, a.name)(a.exprId, a.qualifier, isGenerated = a.isGenerated)).getOrElse(a) + } + } + + private def findRelation(plan: LogicalPlan) + : (Option[LogicalPlan], Seq[NamedExpression], Seq[Expression], Map[ExprId, Expression]) = { + plan match { + case relation @ LogicalRelation(files: HadoopFsRelation, _, table) + if files.partitionSchema.nonEmpty => + (Some(relation), Seq.empty[NamedExpression], Seq.empty[Expression], Map.empty) + + case relation: CatalogRelation if relation.catalogTable.partitionColumnNames.nonEmpty => + (Some(relation), Seq.empty[NamedExpression], Seq.empty[Expression], Map.empty) + + case p @ Project(fields, child) if fields.forall(_.deterministic) => + val (plan, _, filters, aliases) = findRelation(child) + val substitutedFields = fields.map(substitute(aliases)).asInstanceOf[Seq[NamedExpression]] + (plan, substitutedFields, filters, collectAliases(substitutedFields)) + + case f @ Filter(condition, child) if condition.deterministic => + val (plan, fields, filters, aliases) = findRelation(child) + val substitutedCondition = substitute(aliases)(condition) + (plan, fields, filters ++ Seq(substitutedCondition), aliases) + + case _ => (None, Seq.empty[NamedExpression], Seq.empty[Expression], Map.empty) + } + } + + private def convertToMetadataOnlyPlan( + parent: LogicalPlan, + projectList: Seq[NamedExpression], + filters: Seq[Expression], + relation: LogicalPlan): LogicalPlan = relation match { + case l @ LogicalRelation(files: HadoopFsRelation, _, _) => + val attributeMap = l.output.map(attr => (attr.name, attr)).toMap + val partitionColumns = files.partitionSchema.map { field => + attributeMap.getOrElse(field.name, throw new AnalysisException( + s"Unable to resolve ${field.name} given [${l.output.map(_.name).mkString(", ")}]")) + } + val filterColumns = filters.flatMap(_.references) + val projectSet = parent.references ++ AttributeSet(filterColumns) + if (projectSet.subsetOf(AttributeSet(partitionColumns))) { + val selectedPartitions = files.location.listFiles(filters) + val partitionValues = selectedPartitions.map(_.values) + val valuesRdd = sparkSession.sparkContext.parallelize(partitionValues, 1) + val valuesPlan = LogicalRDD(partitionColumns, valuesRdd)(sparkSession) + if (projectList.nonEmpty) { + parent.withNewChildren(Project(projectList, valuesPlan) :: Nil) + } else { + parent.withNewChildren(valuesPlan :: Nil) + } + } else { + parent + } + + case relation: CatalogRelation => + val attributeMap = relation.output.map(attr => (attr.name, attr)).toMap + val partitionColumns = relation.catalogTable.partitionColumnNames.map { column => + attributeMap.getOrElse(column, throw new AnalysisException( + s"Unable to resolve ${column} given [${relation.output.map(_.name).mkString(", ")}]")) + } + val filterColumns = filters.flatMap(_.references) + val projectSet = parent.references ++ AttributeSet(filterColumns) + if (projectSet.subsetOf(AttributeSet(partitionColumns))) { + val partitionColumnDataTypes = partitionColumns.map(_.dataType) + val partitionValues = catalog.getPartitionsByFilter(relation.catalogTable, filters) --- End diff -- for the sake of simplicity, how about we remove this optimization and always apply filters after getting the partition values? We can implement it in follow-ups.
--- If your project is set up for it, you can reply to this email and have your reply appear on GitHub as well. If your project does not have this feature enabled and wishes so, or if the feature is enabled but not working, please contact infrastructure at infrastruct...@apache.org or file a JIRA ticket with INFRA. --- --------------------------------------------------------------------- To unsubscribe, e-mail: reviews-unsubscr...@spark.apache.org For additional commands, e-mail: reviews-h...@spark.apache.org