diff --git a/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/DataSourceWriteBenchmark.scala b/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/DataSourceWriteBenchmark.scala new file mode 100644 index 0000000000000..2d2cdebd067c1 --- /dev/null +++ b/sql/core/src/test/scala/org/apache/spark/sql/execution/benchmark/DataSourceWriteBenchmark.scala @@ -0,0 +1,149 @@ +/* + * 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.benchmark + +import org.apache.spark.SparkConf +import org.apache.spark.sql.SparkSession +import org.apache.spark.sql.internal.SQLConf +import org.apache.spark.util.Benchmark + +/** + * Benchmark to measure data source write performance. + * By default it measures 4 data source format: Parquet, ORC, JSON, CSV: + * spark-submit --class + * To measure specified formats, run it with arguments: + * spark-submit --class format1 [format2] [...] + */ +object DataSourceWriteBenchmark { + val conf = new SparkConf() + .setAppName("DataSourceWriteBenchmark") + .setIfMissing("spark.master", "local[1]") + .set("spark.sql.parquet.compression.codec", "snappy") + .set("spark.sql.orc.compression.codec", "snappy") + + val spark = SparkSession.builder.config(conf).getOrCreate() + + // Set default configs. Individual cases will change them if necessary. + spark.conf.set(SQLConf.WHOLESTAGE_CODEGEN_ENABLED.key, "true") + + val tempTable = "temp" + val numRows = 1024 * 1024 * 15 + + def withTempTable(tableNames: String*)(f: => Unit): Unit = { + try f finally tableNames.foreach(spark.catalog.dropTempView) + } + + def withTable(tableNames: String*)(f: => Unit): Unit = { + try f finally { + tableNames.foreach { name => + spark.sql(s"DROP TABLE IF EXISTS $name") + } + } + } + + def writeNumeric(table: String, format: String, benchmark: Benchmark, dataType: String): Unit = { + spark.sql(s"create table $table(id $dataType) using $format") + benchmark.addCase(s"Output Single $dataType Column") { _ => + spark.sql(s"INSERT OVERWRITE TABLE $table SELECT CAST(id AS $dataType) AS c1 FROM $tempTable") + } + } + + def writeIntString(table: String, format: String, benchmark: Benchmark): Unit = { + spark.sql(s"CREATE TABLE $table(c1 INT, c2 STRING) USING $format") + benchmark.addCase("Output Int and String Column") { _ => + spark.sql(s"INSERT OVERWRITE TABLE $table SELECT CAST(id AS INT) AS " + + s"c1, CAST(id AS STRING) AS c2 FROM $tempTable") + } + } + + def writePartition(table: String, format: String, benchmark: Benchmark): Unit = { + spark.sql(s"CREATE TABLE $table(p INT, id INT) USING $format PARTITIONED BY (p)") + benchmark.addCase("Output Partitions") { _ => + spark.sql(s"INSERT OVERWRITE TABLE $table SELECT CAST(id AS INT) AS id," + + s" CAST(id % 2 AS INT) AS p FROM $tempTable") + } + } + + def writeBucket(table: String, format: String, benchmark: Benchmark): Unit = { + spark.sql(s"CREATE TABLE $table(c1 INT, c2 INT) USING $format CLUSTERED BY (c2) INTO 2 BUCKETS") + benchmark.addCase("Output Buckets") { _ => + spark.sql(s"INSERT OVERWRITE TABLE $table SELECT CAST(id AS INT) AS " + + s"c1, CAST(id AS INT) AS c2 FROM $tempTable") + } + } + + def main(args: Array[String]): Unit = { + val tableInt = "tableInt" + val tableDouble = "tableDouble" + val tableIntString = "tableIntString" + val tablePartition = "tablePartition" + val tableBucket = "tableBucket" + val formats: Seq[String] = if (args.isEmpty) { + Seq("Parquet", "ORC", "JSON", "CSV") + } else { + args + } + /* + Intel(R) Core(TM) i7-6920HQ CPU @ 2.90GHz + Parquet writer benchmark: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative + ------------------------------------------------------------------------------------------------ + Output Single Int Column 1815 / 1932 8.7 115.4 1.0X + Output Single Double Column 1877 / 1878 8.4 119.3 1.0X + Output Int and String Column 6265 / 6543 2.5 398.3 0.3X + Output Partitions 4067 / 4457 3.9 258.6 0.4X + Output Buckets 5608 / 5820 2.8 356.6 0.3X + + ORC writer benchmark: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative + ------------------------------------------------------------------------------------------------ + Output Single Int Column 1201 / 1239 13.1 76.3 1.0X + Output Single Double Column 1542 / 1600 10.2 98.0 0.8X + Output Int and String Column 6495 / 6580 2.4 412.9 0.2X + Output Partitions 3648 / 3842 4.3 231.9 0.3X + Output Buckets 5022 / 5145 3.1 319.3 0.2X + + JSON writer benchmark: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative + ------------------------------------------------------------------------------------------------ + Output Single Int Column 1988 / 2093 7.9 126.4 1.0X + Output Single Double Column 2854 / 2911 5.5 181.4 0.7X + Output Int and String Column 6467 / 6653 2.4 411.1 0.3X + Output Partitions 4548 / 5055 3.5 289.1 0.4X + Output Buckets 5664 / 5765 2.8 360.1 0.4X + + CSV writer benchmark: Best/Avg Time(ms) Rate(M/s) Per Row(ns) Relative + ------------------------------------------------------------------------------------------------ + Output Single Int Column 3025 / 3190 5.2 192.3 1.0X + Output Single Double Column 3575 / 3634 4.4 227.3 0.8X + Output Int and String Column 7313 / 7399 2.2 464.9 0.4X + Output Partitions 5105 / 5190 3.1 324.6 0.6X + Output Buckets 6986 / 6992 2.3 444.1 0.4X + */ + withTempTable(tempTable) { + spark.range(numRows).createOrReplaceTempView(tempTable) + formats.foreach { format => + withTable(tableInt, tableDouble, tableIntString, tablePartition, tableBucket) { + val benchmark = new Benchmark(s"$format writer benchmark", numRows) + writeNumeric(tableInt, format, benchmark, "Int") + writeNumeric(tableDouble, format, benchmark, "Double") + writeIntString(tableIntString, format, benchmark) + writePartition(tablePartition, format, benchmark) + writeBucket(tableBucket, format, benchmark) + benchmark.run() + } + } + } + } +}