maropu commented on a change in pull request #32243: URL: https://github.com/apache/spark/pull/32243#discussion_r624230505
########## File path: sql/core/src/test/scala/org/apache/spark/sql/tpcdsDagagen.scala ########## @@ -0,0 +1,443 @@ +/* + * 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 java.util.concurrent.LinkedBlockingQueue + +import scala.collection.immutable.Stream +import scala.sys.process._ +import scala.util.Try + +import org.apache.spark.SparkContext +import org.apache.spark.internal.Logging +import org.apache.spark.rdd.RDD +import org.apache.spark.sql.functions.{col, rpad} +import org.apache.spark.sql.types.{CharType, StringType, StructField, StructType, VarcharType} + +// The classes in this file are basically moved from https://github.com/databricks/spark-sql-perf + +/** + * Using ProcessBuilder.lineStream produces a stream, that uses + * a LinkedBlockingQueue with a default capacity of Integer.MAX_VALUE. + * + * This causes OOM if the consumer cannot keep up with the producer. + * + * See scala.sys.process.ProcessBuilderImpl.lineStream + */ +object BlockingLineStream { + + // See scala.sys.process.Streamed + private final class BlockingStreamed[T]( + val process: T => Unit, + val done: Int => Unit, + val stream: () => Stream[T]) + + // See scala.sys.process.Streamed + private object BlockingStreamed { + // scala.process.sys.Streamed uses default of Integer.MAX_VALUE, + // which causes OOMs if the consumer cannot keep up with producer. + val maxQueueSize = 65536 + + def apply[T](nonzeroException: Boolean): BlockingStreamed[T] = { + val q = new LinkedBlockingQueue[Either[Int, T]](maxQueueSize) + + def next(): Stream[T] = q.take match { + case Left(0) => Stream.empty + case Left(code) => + if (nonzeroException) scala.sys.error("Nonzero exit code: " + code) else Stream.empty + case Right(s) => Stream.cons(s, next()) + } + + new BlockingStreamed((s: T) => q put Right(s), code => q put Left(code), () => next()) + } + } + + // See scala.sys.process.ProcessImpl.Spawn + private object Spawn { + def apply(f: => Unit): Thread = apply(f, daemon = false) + def apply(f: => Unit, daemon: Boolean): Thread = { + val thread = new Thread() { override def run() = { f } } + thread.setDaemon(daemon) + thread.start() + thread + } + } + + def apply(command: Seq[String]): Stream[String] = { + val streamed = BlockingStreamed[String](true) + val process = command.run(BasicIO(false, streamed.process, None)) + Spawn(streamed.done(process.exitValue())) + streamed.stream() + } +} + +class Dsdgen(dsdgenDir: String) extends Serializable { + private val dsdgen = s"$dsdgenDir/dsdgen" + + def generate( + sparkContext: SparkContext, + tableName: String, + partitions: Int, + scaleFactor: Int): RDD[String] = { + val generatedData = { + sparkContext.parallelize(1 to partitions, partitions).flatMap { i => + val localToolsDir = if (new java.io.File(dsdgen).exists) { + dsdgenDir + } else if (new java.io.File(s"/$dsdgen").exists) { + s"/$dsdgenDir" + } else { + throw new IllegalStateException( + s"Could not find dsdgen at $dsdgen or /$dsdgen. Run install") + } + + // NOTE: RNGSEED is the RNG seed used by the data generator. Right now, it is fixed to + // 19620718 that is used to generate `https://github.com/maropu/spark-tpcds-sf-1`. + val parallel = if (partitions > 1) s"-parallel $partitions -child $i" else "" + val commands = Seq( + "bash", "-c", + s"cd $localToolsDir && ./dsdgen -table $tableName -filter Y -scale $scaleFactor " + + s"-RNGSEED 19620718 $parallel") + BlockingLineStream(commands) + } + } + + generatedData.setName(s"$tableName, sf=$scaleFactor, strings") + generatedData + } +} + +class TPCDSTables(sqlContext: SQLContext, dsdgenDir: String, scaleFactor: Int) + extends TPCDSSchema with Logging with Serializable { + + private val dataGenerator = new Dsdgen(dsdgenDir) + + private def tables: Seq[Table] = tableColumns.map { case (tableName, schemaString) => + val partitionColumns = tablePartitionColumns.getOrElse(tableName, Nil) + .map(_.stripPrefix("`").stripSuffix("`")) + Table(tableName, partitionColumns, StructType.fromDDL(schemaString)) + }.toSeq + + private case class Table(name: String, partitionColumns: Seq[String], schema: StructType) { + def nonPartitioned: Table = { + Table(name, Nil, schema) + } + + private def df(numPartition: Int) = { + val generatedData = dataGenerator.generate( + sqlContext.sparkContext, name, numPartition, scaleFactor) + val rows = generatedData.mapPartitions { iter => + iter.map { l => + val values = l.split("\\|", -1).dropRight(1).map { v => + if (v.equals("")) { + // If the string value is an empty string, we turn it to a null + null + } else { + v + } + } + Row.fromSeq(values) + } + } + + val stringData = + sqlContext.createDataFrame( + rows, + StructType(schema.fields.map(f => StructField(f.name, StringType)))) + + val convertedData = { + val columns = schema.fields.map { f => + val c = f.dataType match { + // Needs right-padding for char types + case CharType(n) => rpad(Column(f.name), n, " ") + // Don't need a cast for varchar types + case _: VarcharType => col(f.name) + case _ => col(f.name).cast(f.dataType) + } + c.as(f.name) + } + stringData.select(columns: _*) + } + + convertedData + } + + def genData( + location: String, + format: String, + overwrite: Boolean, + clusterByPartitionColumns: Boolean, + filterOutNullPartitionValues: Boolean, + numPartitions: Int): Unit = { + val mode = if (overwrite) SaveMode.Overwrite else SaveMode.Ignore + + val data = df(numPartitions) + val tempTableName = s"${name}_text" + data.createOrReplaceTempView(tempTableName) + + val writer = if (partitionColumns.nonEmpty) { + if (clusterByPartitionColumns) { + val columnString = data.schema.fields.map { field => + field.name + }.mkString(",") + val partitionColumnString = partitionColumns.mkString(",") + val predicates = if (filterOutNullPartitionValues) { + partitionColumns.map(col => s"$col IS NOT NULL").mkString("WHERE ", " AND ", "") + } else { + "" + } + + val query = + s""" + |SELECT + | $columnString + |FROM + | $tempTableName + |$predicates + |DISTRIBUTE BY + | $partitionColumnString + """.stripMargin + val grouped = sqlContext.sql(query) + logInfo(s"Pre-clustering with partitioning columns with query $query.") + grouped.write + } else { + data.write + } + } else { + // treat non-partitioned tables as "one partition" that we want to coalesce + if (clusterByPartitionColumns) { + // in case data has more than maxRecordsPerFile, split into multiple writers to improve + // datagen speed files will be truncated to maxRecordsPerFile value, so the final + // result will be the same. + val numRows = data.count + val maxRecordPerFile = Try { + sqlContext.getConf("spark.sql.files.maxRecordsPerFile").toInt + }.getOrElse(0) + + if (maxRecordPerFile > 0 && numRows > maxRecordPerFile) { + val numFiles = (numRows.toDouble/maxRecordPerFile).ceil.toInt + logInfo(s"Coalescing into $numFiles files") + data.coalesce(numFiles).write + } else { + data.coalesce(1).write + } + } else { + data.write + } + } + writer.format(format).mode(mode) + if (partitionColumns.nonEmpty) { + writer.partitionBy(partitionColumns: _*) + } + logInfo(s"Generating table $name in database to $location with save mode $mode.") + writer.save(location) + sqlContext.dropTempTable(tempTableName) + } + } + + def genData( + location: String, + format: String, + overwrite: Boolean, + partitionTables: Boolean, + clusterByPartitionColumns: Boolean, + filterOutNullPartitionValues: Boolean, + tableFilter: String = "", + numPartitions: Int = 100): Unit = { + var tablesToBeGenerated = if (partitionTables) { + tables + } else { + tables.map(_.nonPartitioned) + } + + if (!tableFilter.isEmpty) { + tablesToBeGenerated = tablesToBeGenerated.filter(_.name == tableFilter) + if (tablesToBeGenerated.isEmpty) { + throw new RuntimeException("Bad table name filter: " + tableFilter) + } + } + + tablesToBeGenerated.foreach { table => + val tableLocation = s"$location/${table.name}" + table.genData(tableLocation, format, overwrite, clusterByPartitionColumns, + filterOutNullPartitionValues, numPartitions) + } + } +} + +class GenTPCDSDataConfig(args: Array[String]) { + var master: String = "local[*]" + var dsdgenDir: String = null + var location: String = null + var scaleFactor: Int = 1 + var format: String = "parquet" + var overwrite: Boolean = false + var partitionTables: Boolean = false + var clusterByPartitionColumns: Boolean = false + var filterOutNullPartitionValues: Boolean = false + var tableFilter: String = "" + var numPartitions: Int = 100 + + parseArgs(args.toList) + + private def parseArgs(inputArgs: List[String]): Unit = { + var args = inputArgs + + while (args.nonEmpty) { + args match { + case "--master" :: value :: tail => + master = value + args = tail + + case "--dsdgenDir" :: value :: tail => + dsdgenDir = value + args = tail + + case "--location" :: value :: tail => + location = value + args = tail + + case "--scaleFactor" :: value :: tail => + scaleFactor = toPositiveIntValue("Scale factor", value) + args = tail + + case "--format" :: value :: tail => + format = value + args = tail + + case "--overwrite" :: tail => + overwrite = true + args = tail + + case "--partitionTables" :: tail => + partitionTables = true + args = tail + + case "--clusterByPartitionColumns" :: tail => + clusterByPartitionColumns = true + args = tail + + case "--filterOutNullPartitionValues" :: tail => + filterOutNullPartitionValues = true + args = tail + + case "--tableFilter" :: value :: tail => + tableFilter = value + args = tail + + case "--numPartitions" :: value :: tail => + numPartitions = toPositiveIntValue("Number of partitions", value) + args = tail + + case "--help" :: tail => + printUsageAndExit(0) + + case _ => + // scalastyle:off println + System.err.println("Unknown/unsupported param " + args) + // scalastyle:on println + printUsageAndExit(1) + } + } + + checkRequiredArguments() + } + + private def printUsageAndExit(exitCode: Int): Unit = { + // scalastyle:off + System.err.println(""" + |build/sbt "test:runMain <this class> [Options]" + |Options: + | --master the Spark master to use, default to local[*] + | --dsdgenDir location of dsdgen + | --location root directory of location to generate data in + | --scaleFactor size of the dataset to generate (in GB) + | --format generated data format, Parquet, ORC ... + | --overwrite whether to overwrite the data that is already there + | --partitionTables whether to create the partitioned fact tables + | --clusterByPartitionColumns whether to shuffle to get partitions coalesced into single files + | --filterOutNullPartitionValues whether to filter out the partition with NULL key value + | --tableFilter comma-separated list of table names to generate (e.g., store_sales,store_returns), + | all the tables are generated by default + | --numPartitions how many dsdgen partitions to run - number of input tasks + """.stripMargin) + // scalastyle:on + System.exit(exitCode) + } + + private def toPositiveIntValue(name: String, v: String): Int = { + if (Try(v.toInt).getOrElse(-1) <= 0) { + // scalastyle:off println + System.err.println(s"$name must be a positive number") + // scalastyle:on println + printUsageAndExit(-1) + } + v.toInt + } + + private def checkRequiredArguments(): Unit = { + if (dsdgenDir == null) { + // scalastyle:off println + System.err.println("Must specify a dsdgen path") + // scalastyle:on println + printUsageAndExit(-1) + } + if (location == null) { + // scalastyle:off println + System.err.println("Must specify an output location") + // scalastyle:on println + printUsageAndExit(-1) + } + } +} + +/** + * This class generates TPCDS table data by using tpcds-kit: + * - https://github.com/databricks/tpcds-kit + * + * To run this: + * {{{ + * build/sbt "sql/test:runMain <this class> --dsdgenDir <path> --location <path> --scaleFactor 1" + * }}} + */ +object GenTPCDSData { + + def main(args: Array[String]): Unit = { + val config = new GenTPCDSDataConfig(args) + + val spark = SparkSession + .builder() + .appName(getClass.getName) + .master(config.master) + .getOrCreate() + + val tables = new TPCDSTables( + spark.sqlContext, + dsdgenDir = config.dsdgenDir, + scaleFactor = config.scaleFactor) + + tables.genData( + location = config.location, + format = config.format, + overwrite = config.overwrite, + partitionTables = config.partitionTables, + clusterByPartitionColumns = config.clusterByPartitionColumns, + filterOutNullPartitionValues = config.filterOutNullPartitionValues, + tableFilter = config.tableFilter, + numPartitions = config.numPartitions) Review comment: Oh, I forgot it. Thank you. -- 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]
