spark FileSourceStrategy 源码

spark FileSourceStrategy 代码

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

import java.util.Locale

import org.apache.spark.internal.Logging
import org.apache.spark.sql._
import org.apache.spark.sql.catalyst.catalog.BucketSpec
import org.apache.spark.sql.catalyst.expressions
import org.apache.spark.sql.catalyst.expressions._
import org.apache.spark.sql.catalyst.planning.PhysicalOperation
import org.apache.spark.sql.catalyst.plans.logical.LogicalPlan
import org.apache.spark.sql.execution.{FileSourceScanExec, SparkPlan}
import org.apache.spark.sql.execution.datasources.FileFormat.METADATA_NAME
import org.apache.spark.sql.types.{DoubleType, FloatType, LongType, StructType}
import org.apache.spark.util.collection.BitSet

/**
 * A strategy for planning scans over collections of files that might be partitioned or bucketed
 * by user specified columns.
 *
 * At a high level planning occurs in several phases:
 *  - Split filters by when they need to be evaluated.
 *  - Prune the schema of the data requested based on any projections present. Today this pruning
 *    is only done on top level columns, but formats should support pruning of nested columns as
 *    well.
 *  - Construct a reader function by passing filters and the schema into the FileFormat.
 *  - Using a partition pruning predicates, enumerate the list of files that should be read.
 *  - Split the files into tasks and construct a FileScanRDD.
 *  - Add any projection or filters that must be evaluated after the scan.
 *
 * Files are assigned into tasks using the following algorithm:
 *  - If the table is bucketed, group files by bucket id into the correct number of partitions.
 *  - If the table is not bucketed or bucketing is turned off:
 *   - If any file is larger than the threshold, split it into pieces based on that threshold
 *   - Sort the files by decreasing file size.
 *   - Assign the ordered files to buckets using the following algorithm.  If the current partition
 *     is under the threshold with the addition of the next file, add it.  If not, open a new bucket
 *     and add it.  Proceed to the next file.
 */
object FileSourceStrategy extends Strategy with PredicateHelper with Logging {

  // should prune buckets iff num buckets is greater than 1 and there is only one bucket column
  private def shouldPruneBuckets(bucketSpec: Option[BucketSpec]): Boolean = {
    bucketSpec match {
      case Some(spec) => spec.bucketColumnNames.length == 1 && spec.numBuckets > 1
      case None => false
    }
  }

  private def getExpressionBuckets(
      expr: Expression,
      bucketColumnName: String,
      numBuckets: Int): BitSet = {

    def getBucketNumber(attr: Attribute, v: Any): Int = {
      BucketingUtils.getBucketIdFromValue(attr, numBuckets, v)
    }

    def getBucketSetFromIterable(attr: Attribute, iter: Iterable[Any]): BitSet = {
      val matchedBuckets = new BitSet(numBuckets)
      iter
        .map(v => getBucketNumber(attr, v))
        .foreach(bucketNum => matchedBuckets.set(bucketNum))
      matchedBuckets
    }

    def getBucketSetFromValue(attr: Attribute, v: Any): BitSet = {
      val matchedBuckets = new BitSet(numBuckets)
      matchedBuckets.set(getBucketNumber(attr, v))
      matchedBuckets
    }

    expr match {
      case expressions.Equality(a: Attribute, Literal(v, _)) if a.name == bucketColumnName =>
        getBucketSetFromValue(a, v)
      case expressions.In(a: Attribute, list)
        if list.forall(_.isInstanceOf[Literal]) && a.name == bucketColumnName =>
        getBucketSetFromIterable(a, list.map(e => e.eval(EmptyRow)))
      case expressions.InSet(a: Attribute, hset) if a.name == bucketColumnName =>
        getBucketSetFromIterable(a, hset)
      case expressions.IsNull(a: Attribute) if a.name == bucketColumnName =>
        getBucketSetFromValue(a, null)
      case expressions.IsNaN(a: Attribute)
        if a.name == bucketColumnName && a.dataType == FloatType =>
        getBucketSetFromValue(a, Float.NaN)
      case expressions.IsNaN(a: Attribute)
        if a.name == bucketColumnName && a.dataType == DoubleType =>
        getBucketSetFromValue(a, Double.NaN)
      case expressions.And(left, right) =>
        getExpressionBuckets(left, bucketColumnName, numBuckets) &
          getExpressionBuckets(right, bucketColumnName, numBuckets)
      case expressions.Or(left, right) =>
        getExpressionBuckets(left, bucketColumnName, numBuckets) |
        getExpressionBuckets(right, bucketColumnName, numBuckets)
      case _ =>
        val matchedBuckets = new BitSet(numBuckets)
        matchedBuckets.setUntil(numBuckets)
        matchedBuckets
    }
  }

  private def genBucketSet(
      normalizedFilters: Seq[Expression],
      bucketSpec: BucketSpec): Option[BitSet] = {
    if (normalizedFilters.isEmpty) {
      return None
    }

    val bucketColumnName = bucketSpec.bucketColumnNames.head
    val numBuckets = bucketSpec.numBuckets

    val normalizedFiltersAndExpr = normalizedFilters
      .reduce(expressions.And)
    val matchedBuckets = getExpressionBuckets(normalizedFiltersAndExpr, bucketColumnName,
      numBuckets)

    val numBucketsSelected = matchedBuckets.cardinality()

    logInfo {
      s"Pruned ${numBuckets - numBucketsSelected} out of $numBuckets buckets."
    }

    // None means all the buckets need to be scanned
    if (numBucketsSelected == numBuckets) {
      None
    } else {
      Some(matchedBuckets)
    }
  }

  def apply(plan: LogicalPlan): Seq[SparkPlan] = plan match {
    case PhysicalOperation(projects, filters,
      l @ LogicalRelation(fsRelation: HadoopFsRelation, _, table, _)) =>
      // Filters on this relation fall into four categories based on where we can use them to avoid
      // reading unneeded data:
      //  - partition keys only - used to prune directories to read
      //  - bucket keys only - optionally used to prune files to read
      //  - keys stored in the data only - optionally used to skip groups of data in files
      //  - filters that need to be evaluated again after the scan
      val filterSet = ExpressionSet(filters)

      val normalizedFilters = DataSourceStrategy.normalizeExprs(
        filters.filter(_.deterministic), l.output)

      val partitionColumns =
        l.resolve(
          fsRelation.partitionSchema, fsRelation.sparkSession.sessionState.analyzer.resolver)
      val partitionSet = AttributeSet(partitionColumns)

      // this partitionKeyFilters should be the same with the ones being executed in
      // PruneFileSourcePartitions
      val partitionKeyFilters = DataSourceStrategy.getPushedDownFilters(partitionColumns,
        normalizedFilters)

      // subquery expressions are filtered out because they can't be used to prune buckets or pushed
      // down as data filters, yet they would be executed
      val normalizedFiltersWithoutSubqueries =
        normalizedFilters.filterNot(SubqueryExpression.hasSubquery)

      val bucketSpec: Option[BucketSpec] = fsRelation.bucketSpec
      val bucketSet = if (shouldPruneBuckets(bucketSpec)) {
        genBucketSet(normalizedFiltersWithoutSubqueries, bucketSpec.get)
      } else {
        None
      }

      val dataColumns =
        l.resolve(fsRelation.dataSchema, fsRelation.sparkSession.sessionState.analyzer.resolver)

      // Partition keys are not available in the statistics of the files.
      // `dataColumns` might have partition columns, we need to filter them out.
      val dataColumnsWithoutPartitionCols = dataColumns.filterNot(partitionSet.contains)
      val dataFilters = normalizedFiltersWithoutSubqueries.flatMap { f =>
        if (f.references.intersect(partitionSet).nonEmpty) {
          extractPredicatesWithinOutputSet(f, AttributeSet(dataColumnsWithoutPartitionCols))
        } else {
          Some(f)
        }
      }
      val supportNestedPredicatePushdown =
        DataSourceUtils.supportNestedPredicatePushdown(fsRelation)
      val pushedFilters = dataFilters
        .flatMap(DataSourceStrategy.translateFilter(_, supportNestedPredicatePushdown))
      logInfo(s"Pushed Filters: ${pushedFilters.mkString(",")}")

      // Predicates with both partition keys and attributes need to be evaluated after the scan.
      val afterScanFilters = filterSet -- partitionKeyFilters.filter(_.references.nonEmpty)
      logInfo(s"Post-Scan Filters: ${afterScanFilters.mkString(",")}")

      val filterAttributes = AttributeSet(afterScanFilters)
      val requiredExpressions: Seq[NamedExpression] = filterAttributes.toSeq ++ projects
      val requiredAttributes = AttributeSet(requiredExpressions)

      val metadataStructOpt = l.output.collectFirst {
        case FileSourceMetadataAttribute(attr) => attr
      }

      val metadataColumns = metadataStructOpt.map { metadataStruct =>
        metadataStruct.dataType.asInstanceOf[StructType].fields.map { field =>
          FileSourceMetadataAttribute(field.name, field.dataType)
        }.toSeq
      }.getOrElse(Seq.empty)

      val fileConstantMetadataColumns: Seq[Attribute] =
        metadataColumns.filter(_.name != FileFormat.ROW_INDEX)

      val readDataColumns = dataColumns
          .filter(requiredAttributes.contains)
          .filterNot(partitionColumns.contains)

      val fileFormatReaderGeneratedMetadataColumns: Seq[Attribute] =
        metadataColumns.map(_.name).flatMap {
          case FileFormat.ROW_INDEX =>
            if ((readDataColumns ++ partitionColumns).map(_.name.toLowerCase(Locale.ROOT))
                .contains(FileFormat.ROW_INDEX_TEMPORARY_COLUMN_NAME)) {
              throw new AnalysisException(FileFormat.ROW_INDEX_TEMPORARY_COLUMN_NAME +
                " is a reserved column name that cannot be read in combination with " +
                s"${FileFormat.METADATA_NAME}.${FileFormat.ROW_INDEX} column.")
            }
            Some(AttributeReference(FileFormat.ROW_INDEX_TEMPORARY_COLUMN_NAME, LongType)())
          case _ => None
        }

      val outputDataSchema = (readDataColumns ++ fileFormatReaderGeneratedMetadataColumns)
        .toStructType

      // The output rows will be produced during file scan operation in three steps:
      //  (1) File format reader populates a `Row` with `readDataColumns` and
      //      `fileFormatReaderGeneratedMetadataColumns`
      //  (2) Then, a row containing `partitionColumns` is joined at the end.
      //  (3) Finally, a row containing `fileConstantMetadataColumns` is also joined at the end.
      // By placing `fileFormatReaderGeneratedMetadataColumns` before `partitionColumns` and
      // `fileConstantMetadataColumns` in the `outputAttributes` we make these row operations
      // simpler and more efficient.
      val outputAttributes = readDataColumns ++ fileFormatReaderGeneratedMetadataColumns ++
        partitionColumns ++ fileConstantMetadataColumns

      val scan =
        FileSourceScanExec(
          fsRelation,
          outputAttributes,
          outputDataSchema,
          partitionKeyFilters.toSeq,
          bucketSet,
          None,
          dataFilters,
          table.map(_.identifier))

      // extra Project node: wrap flat metadata columns to a metadata struct
      val withMetadataProjections = metadataStructOpt.map { metadataStruct =>
        val structColumns = metadataColumns.map { col => col.name match {
            case FileFormat.FILE_PATH | FileFormat.FILE_NAME | FileFormat.FILE_SIZE |
                 FileFormat.FILE_MODIFICATION_TIME =>
              col
            case FileFormat.ROW_INDEX =>
              fileFormatReaderGeneratedMetadataColumns
                .find(_.name == FileFormat.ROW_INDEX_TEMPORARY_COLUMN_NAME)
                .get.withName(FileFormat.ROW_INDEX)
          }
        }
        val metadataAlias =
          Alias(CreateStruct(structColumns), METADATA_NAME)(exprId = metadataStruct.exprId)
        execution.ProjectExec(
          readDataColumns ++ partitionColumns :+ metadataAlias, scan)
      }.getOrElse(scan)

      val afterScanFilter = afterScanFilters.toSeq.reduceOption(expressions.And)
      val withFilter = afterScanFilter
        .map(execution.FilterExec(_, withMetadataProjections))
        .getOrElse(withMetadataProjections)
      val withProjections = if (projects == withFilter.output) {
        withFilter
      } else {
        execution.ProjectExec(projects, withFilter)
      }

      withProjections :: Nil

    case _ => Nil
  }
}

相关信息

spark 源码目录

相关文章

spark AggregatePushDownUtils 源码

spark ApplyCharTypePadding 源码

spark BasicWriteStatsTracker 源码

spark BucketingUtils 源码

spark CatalogFileIndex 源码

spark CodecStreams 源码

spark DataSource 源码

spark DataSourceStrategy 源码

spark DataSourceUtils 源码

spark DaysWritable 源码