cloud-fan commented on a change in pull request #29795: URL: https://github.com/apache/spark/pull/29795#discussion_r494764849
########## File path: sql/core/src/test/scala/org/apache/spark/sql/UpdateFieldsBenchmark.scala ########## @@ -0,0 +1,310 @@ +/* + * 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 + +import org.apache.spark.benchmark.Benchmark +import org.apache.spark.sql.execution.benchmark.SqlBasedBenchmark +import org.apache.spark.sql.functions.{col, lit} +import org.apache.spark.sql.test.SharedSparkSession +import org.apache.spark.sql.types.{IntegerType, StructField, StructType} + +/** + * Benchmark to measure Spark's performance analyzing and optimizing long UpdateFields chains. + * + * {{{ + * To run this benchmark: + * 1. without sbt: + * bin/spark-submit --class <this class> <spark sql test jar> + * 2. with sbt: + * build/sbt "sql/test:runMain <this class>" + * 3. generate result: + * SPARK_GENERATE_BENCHMARK_FILES=1 build/sbt "sql/test:runMain <this class>" + * Results will be written to "benchmarks/UpdateFieldsBenchmark-results.txt". + * }}} + */ +object UpdateFieldsBenchmark extends SqlBasedBenchmark { + + private def nestedColName(d: Int, colNum: Int): String = s"nested${d}Col$colNum" + + private def nestedStructType( + colNums: Seq[Int], + nullable: Boolean, + maxDepth: Int, + currDepth: Int = 1): StructType = { + + if (currDepth == maxDepth) { + val fields = colNums.map { colNum => + val name = nestedColName(currDepth, colNum) + StructField(name, IntegerType, nullable = false) + } + StructType(fields) + } else { + val fields = colNums.foldLeft(Seq.empty[StructField]) { + case (structFields, colNum) if colNum == 0 => + val nested = nestedStructType(colNums, nullable, maxDepth, currDepth + 1) + structFields :+ StructField(nestedColName(currDepth, colNum), nested, nullable) + case (structFields, colNum) => + val name = nestedColName(currDepth, colNum) + structFields :+ StructField(name, IntegerType, nullable = false) + } + StructType(fields) + } + } + + private def nestedRow(colNums: Seq[Int], maxDepth: Int, currDepth: Int = 1): Row = { + if (currDepth == maxDepth) { + Row.fromSeq(colNums) + } else { + val values = colNums.foldLeft(Seq.empty[Any]) { + case (values, colNum) if colNum == 0 => + values :+ nestedRow(colNums, maxDepth, currDepth + 1) + case (values, colNum) => + values :+ colNum + } + Row.fromSeq(values) + } + } + + /** + * Utility function for generating a DataFrame with nested columns. + * + * @param maxDepth: The depth to which to create nested columns. + * @param numColsAtEachDepth: The number of columns to create at each depth. The value of each + * column will be the same as its index (IntegerType) at that depth + * unless the index = 0, in which case it may be a StructType which + * represents the next depth. + * @param nullable: This value is used to set the nullability of StructType columns. + */ + def nestedDf(maxDepth: Int, numColsAtEachDepth: Int, nullable: Boolean): DataFrame = { + require(maxDepth > 0) + require(numColsAtEachDepth > 0) + + val colNums = 0 until numColsAtEachDepth + val nestedColumn = nestedRow(colNums, maxDepth) + val nestedColumnDataType = nestedStructType(colNums, nullable, maxDepth) + + spark.createDataFrame( + spark.sparkContext.parallelize(Row(nestedColumn) :: Nil), + StructType(Seq(StructField(nestedColName(0, 0), nestedColumnDataType, nullable)))) + } + + // simulates how a user would add/drop nested fields in a performant manner + def modifyNestedColumns( + column: Column, + numsToAdd: Seq[Int], + numsToDrop: Seq[Int], + maxDepth: Int, + currDepth: Int = 1): Column = { + + // drop columns at the current depth + val dropped = if (numsToDrop.nonEmpty) { + column.dropFields(numsToDrop.map(num => nestedColName(currDepth, num)): _*) + } else column + + // add columns at the current depth + val added = numsToAdd.foldLeft(dropped) { + (res, num) => res.withField(nestedColName(currDepth, num), lit(num)) + } + + if (currDepth == maxDepth) { + added + } else { + // add/drop columns at the next depth + val newValue = modifyNestedColumns( + column = col((0 to currDepth).map(d => nestedColName(d, 0)).mkString(".")), + numsToAdd = numsToAdd, + numsToDrop = numsToDrop, + currDepth = currDepth + 1, + maxDepth = maxDepth) + added.withField(nestedColName(currDepth, 0), newValue) + } + } + + def updateFieldsBenchmark( + maxDepth: Int, + initialNumberOfColumns: Int, + numsToAdd: Seq[Int] = Seq.empty, + numsToDrop: Seq[Int] = Seq.empty): Unit = { + + val name = s"Add ${numsToAdd.length} columns and drop ${numsToDrop.length} columns " + + s"at $maxDepth different depths of nesting" + + runBenchmark(name) { + val benchmark = new Benchmark( + name = name, + // Because the point of this benchmark is only to ensure Spark is able to analyze and + // optimize long UpdateFields chains quickly, this benchmark operates only over 1 row of + // data. + valuesPerIteration = 1, + output = output) + + val columnFunc = modifyNestedColumns( + col(nestedColName(0, 0)), + numsToAdd, + numsToDrop, + maxDepth + ).as(nestedColName(0, 0)) + + val nonNullableInputDf = nestedDf(maxDepth, initialNumberOfColumns, nullable = false) + val nullableInputDf = nestedDf(maxDepth, initialNumberOfColumns, nullable = true) + + benchmark.addCase("Non-Nullable StructTypes") { _ => + nonNullableInputDf.select(columnFunc).noop() Review comment: if we only care about analyzer/optimizer performance, we can probably do `nonNullableInputDf.select(columnFunc).queryExecution.optimizedPlan` ---------------------------------------------------------------- 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: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
