spark RemoveRedundantProjects 源码

spark RemoveRedundantProjects 代码

文件路径：/sql/core/src/main/scala/org/apache/spark/sql/execution/RemoveRedundantProjects.scala

/*
 * 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.catalyst.expressions.Attribute
import org.apache.spark.sql.catalyst.expressions.aggregate.{Final, PartialMerge}
import org.apache.spark.sql.catalyst.rules.Rule
import org.apache.spark.sql.execution.aggregate.BaseAggregateExec
import org.apache.spark.sql.execution.datasources.v2.DataSourceV2ScanExecBase
import org.apache.spark.sql.execution.joins.BaseJoinExec
import org.apache.spark.sql.execution.window.WindowExec
import org.apache.spark.sql.internal.SQLConf

/**
 * Remove redundant ProjectExec node from the spark plan. A ProjectExec node is redundant when
 * - It has the same output attributes and orders as its child's output and the ordering of
 *   the attributes is required.
 * - It has the same output attributes as its child's output when attribute output ordering
 *   is not required.
 * This rule needs to be a physical rule because project nodes are useful during logical
 * optimization to prune data. During physical planning, redundant project nodes can be removed
 * to simplify the query plan.
 */
object RemoveRedundantProjects extends Rule[SparkPlan] {
  def apply(plan: SparkPlan): SparkPlan = {
    if (!conf.getConf(SQLConf.REMOVE_REDUNDANT_PROJECTS_ENABLED)) {
      plan
    } else {
      removeProject(plan, true)
    }
  }

  private def removeProject(plan: SparkPlan, requireOrdering: Boolean): SparkPlan = {
    plan match {
      case p @ ProjectExec(_, child) =>
        if (isRedundant(p, child, requireOrdering) && canRemove(p, child)) {
          val newPlan = removeProject(child, requireOrdering)
          // The `newPlan` should retain the logical plan link already. We call `setLogicalLink`
          // here to make sure the `newPlan` sets the `LOGICAL_PLAN_TAG` tag.
          newPlan.setLogicalLink(child.logicalLink.get)
          newPlan
        } else {
          p.mapChildren(removeProject(_, false))
        }
      case op: TakeOrderedAndProjectExec =>
        op.mapChildren(removeProject(_, false))
      case a: BaseAggregateExec =>
        // BaseAggregateExec require specific column ordering when mode is Final or PartialMerge.
        // See comments in BaseAggregateExec inputAttributes method.
        val keepOrdering = a.aggregateExpressions
          .exists(ae => ae.mode.equals(Final) || ae.mode.equals(PartialMerge))
        a.mapChildren(removeProject(_, keepOrdering))
      case o =>
        val required = if (canPassThrough(o)) requireOrdering else true
        o.mapChildren(removeProject(_, requireOrdering = required))
    }
  }

  /**
   * Check if the given node can pass the ordering requirement from its parent.
   */
  private def canPassThrough(plan: SparkPlan): Boolean = plan match {
    case _: FilterExec => true
    // JoinExec ordering requirement should inherit from its parent. If there is no ProjectExec in
    // its ancestors, JoinExec should require output columns to be ordered, and vice versa.
    case _: BaseJoinExec => true
    case _: WindowExec => true
    case _: ExpandExec => true
    case _ => false
  }

  /**
   * Check if the nullability change is positive. It catches the case when the project output
   * attribute is not nullable, but the child output attribute is nullable.
   */
  private def checkNullability(output: Seq[Attribute], childOutput: Seq[Attribute]): Boolean =
    output.zip(childOutput).forall { case (attr1, attr2) => attr1.nullable || !attr2.nullable }

  private def isRedundant(
      project: ProjectExec,
      child: SparkPlan,
      requireOrdering: Boolean): Boolean = {
    child match {
      // If a DataSourceV2ScanExec node does not support columnar, a ProjectExec node is required
      // to convert the rows to UnsafeRow. See DataSourceV2Strategy for more details.
      case d: DataSourceV2ScanExecBase if !d.supportsColumnar => false
      case FilterExec(_, d: DataSourceV2ScanExecBase) if !d.supportsColumnar => false
      case _ =>
        if (requireOrdering) {
          project.output.map(_.exprId.id) == child.output.map(_.exprId.id) &&
            checkNullability(project.output, child.output)
        } else {
          val orderedProjectOutput = project.output.sortBy(_.exprId.id)
          val orderedChildOutput = child.output.sortBy(_.exprId.id)
          orderedProjectOutput.map(_.exprId.id) == orderedChildOutput.map(_.exprId.id) &&
            checkNullability(orderedProjectOutput, orderedChildOutput)
        }
    }
  }

  // SPARK-36020: Currently a project can only be removed if (1) its logical link is empty or (2)
  // its logical link is the same as the child's logical link. This is to ensure the physical
  // plan node can correctly map to its logical plan node in AQE.
  private def canRemove(project: ProjectExec, child: SparkPlan): Boolean = {
    project.logicalLink.isEmpty || project.logicalLink.exists(child.logicalLink.contains)
  }
}

相关信息

spark 源码目录

相关文章

spark AggregatingAccumulator 源码

spark AliasAwareOutputExpression 源码

spark BaseScriptTransformationExec 源码

spark CacheManager 源码

spark CoGroupedIterator 源码

spark CollectMetricsExec 源码

spark Columnar 源码

spark CommandResultExec 源码

spark DataSourceScanExec 源码

spark ExistingRDD 源码