Spark入门1-使用Spark-shell交互分析

朱金华 njzhujinhua 2017/12/03

本系列以Spark2.2.0版本为依据整理。

第一节通过Spark交互式shell熟悉其计算过程。

在2.0版本之前，Spark的主要编程接口是RDD（弹性分布式数据集），在2.0之后，则主推Dataset，他与RDD一样是强类型，但更加优化。RDD接口仍然支持，但为了更优性能考虑还是用Dataset的好。

在spark目录中运行bin/spark-shell，或将spark安装目录设为SPARK_HOME环境变量且将其$SPARK_HOME/bin加到PATH中，则以后可在任意目录执行spark-shell即可启动。

RDD可以从Hadoop的InputFormats文件（如hdfs文件）创建，也可读写本地文件，也可由其他RDD经转换而来。Dataset也具有这些性质。以读取文件为例，RDD时代可以在shell中通过sc.textFile(filename)直接读取，在Dataset则需要通过spark.read.textFile(filename)读取。

1 文件读取

1.1 读取RDD方式

scala> val rdd1=sc.textFile("README.md")rdd1: org.apache.spark.rdd.RDD[String] = README.md MapPartitionsRDD[72] at textFile at <console>:24

此处sc为SparkContext

顺便看下其textFile的实现，支持从HDFS文件系统，本地文件系统，及其他类hadoop文件系统的URI，并返回一个字符串的RDD。在程序中读取本地文件时要注意该文件在所有节点皆存在，因为任何executor的task都可能会执行此代码。

  /**   * Read a text file from HDFS, a local file system (available on all nodes), or any   * Hadoop-supported file system URI, and return it as an RDD of Strings.   * @param path path to the text file on a supported file system   * @param minPartitions suggested minimum number of partitions for the resulting RDD   * @return RDD of lines of the text file   */  def textFile(      path: String,      minPartitions: Int = defaultMinPartitions): RDD[String] = withScope {    assertNotStopped()    hadoopFile(path, classOf[TextInputFormat], classOf[LongWritable], classOf[Text],      minPartitions).map(pair => pair._2.toString).setName(path)  }

1.2 读取Dataset方式

scala> val dataset1=spark.read.textFile("README.md")org.apache.spark.sql.AnalysisException: Path does not exist: file:/home/hadoop/README.md;  at org.apache.spark.sql.execution.datasources.DataSource$$anonfun$14.apply(DataSource.scala:360)  at org.apache.spark.sql.execution.datasources.DataSource$$anonfun$14.apply(DataSource.scala:348)  at scala.collection.TraversableLike$$anonfun$flatMap$1.apply(TraversableLike.scala:241)  at scala.collection.TraversableLike$$anonfun$flatMap$1.apply(TraversableLike.scala:241)  at scala.collection.immutable.List.foreach(List.scala:381)  at scala.collection.TraversableLike$class.flatMap(TraversableLike.scala:241)  at scala.collection.immutable.List.flatMap(List.scala:344)  at org.apache.spark.sql.execution.datasources.DataSource.resolveRelation(DataSource.scala:348)  at org.apache.spark.sql.DataFrameReader.load(DataFrameReader.scala:178)  at org.apache.spark.sql.DataFrameReader.text(DataFrameReader.scala:623)  at org.apache.spark.sql.DataFrameReader.textFile(DataFrameReader.scala:657)  at org.apache.spark.sql.DataFrameReader.textFile(DataFrameReader.scala:632)  ... 48 elidedscala> val dataset1=spark.read.textFile("/webdev/hadoop/dev/spark/README.md")dataset1: org.apache.spark.sql.Dataset[String] = [value: string]

此处实验可见spark.read方式并没有读取环境变量到真正的SPARK_HOME下查找，而是当前目录下寻找但未果。
Spark SQL是查询结构化数据的一个模块，与基础的Spark RDD API不同，Spark SQL提供了查询结构化数据及计算结果等信息的接口。与Spark SQL进行交互的方式包括SQL和Dataset API。运行SQL的查询结果将以Dataset/DataFrame的形式返回。
从上述的实验返回结果页很明确的看出Dataset挂在spark.sql包下，而此处的spark类型为org.apache.spark.sql.SparkSession。

spark.read返回的实际是一个DataFrameReader

  /**   * Returns a [[DataFrameReader]] that can be used to read non-streaming data in as a   * `DataFrame`.   * {{{   *   sparkSession.read.parquet("/path/to/file.parquet")   *   sparkSession.read.schema(schema).json("/path/to/file.json")   * }}}   *   * @since 2.0.0   */  def read: DataFrameReader = new DataFrameReader(self)

DataFrameReader.textFile的实现为

  /**   * Loads text files and returns a [[Dataset]] of String. The underlying schema of the Dataset   * contains a single string column named "value".   *   * If the directory structure of the text files contains partitioning information, those are   * ignored in the resulting Dataset. To include partitioning information as columns, use `text`.   *   * Each line in the text files is a new element in the resulting Dataset. For example:   * {{{   *   // Scala:   *   spark.read.textFile("/path/to/spark/README.md")   *   *   // Java:   *   spark.read().textFile("/path/to/spark/README.md")   * }}}   *   * @param paths input path   * @since 2.0.0   */  @scala.annotation.varargs  def textFile(paths: String*): Dataset[String] = {    assertNoSpecifiedSchema("textFile")    text(paths : _*).select("value").as[String](sparkSession.implicits.newStringEncoder)  }

其加载文本文件并返回一个string的Dataset，这个dataset仅包含单个名为”value”的列。若文本文件的目录结构包含分区信息，在读到的dataset中也将被忽略，要想将这些分区信息作为schema列信息的话，需要用text API, 看textFile的实现，其也是用的text的特殊参数。

2. 查看文件

2.1 查看内容

查看内容可以使用

dataset1.collect().foreach(println)
或

dataset1.take(10).foreach(println)。

其中collect方式返回所有记录，take(n)返回n条记录

scala> dataset1.take(3)res13: Array[String] = Array(# Apache Spark, "", Spark is a fast and general cluster computing system for Big Data. It provides)

2.2 WordCount

演示将每行按空格拆分为单词的过程，这里假设用空格分隔，暂不考虑标点符号的分隔。
先如前所述看下前三行的内容，展示下其格式

scala> dataset1.take(3)res13: Array[String] = Array(# Apache Spark, "", Spark is a fast and general cluster computing system for Big Data. It provides)

然后利用flatMap按照一对多转换，即字符串line转为数组。

scala> val words=dataset1.flatMap(line=>line.split(" "))words: org.apache.spark.sql.Dataset[String] = [value: string]

查看结果的前十个元素

scala> words.take(10)res14: Array[String] = Array(#, Apache, Spark, "", Spark, is, a, fast, and, general)

进一步的可以利用count查看其个数

scala> words.countres47: Long = 566

Dataset没有reduceByKey接口，但可以使用groupByKey接口。我们实现wordcount功能可以使用：

scala> val groupwords=words.groupByKey(a=>a)groupwords: org.apache.spark.sql.KeyValueGroupedDataset[String,String] = org.apache.spark.sql.KeyValueGroupedDataset@59de6c1fscala> val cnt=groupwords.count()cnt: org.apache.spark.sql.Dataset[(String, Long)] = [value: string, count(1): bigint]scala> cnt.collect()res53: Array[(String, Long)] = Array((online,1), (graphs,1), (["Parallel,1), (["Building,1), (thread,1), (documentation,3), (command,,2), (abbreviated,1), (overview,1), (rich,1), (set,2), (-DskipTests,1), (name,1), (page](http://spark.apache.org/documentation.html).,1), (["Specifying,1), (stream,1), (run:,1), (not,1), (programs,2), (tests,2), (./dev/run-tests,1), (will,1), ([run,1), (particular,2), (option,1), (Alternatively,,1), (by,1), (must,1), (using,5), (you,4), (MLlib,1), (DataFrames,,1), (variable,1), (Note,1), (core,1), (more,1), (protocols,1), (guidance,2), (shell:,2), (can,7), (site,,1), (systems.,1), (Maven,1), ([building,1), (configure,1), (for,12), (README,1), (Interactive,2), (how,3), ([Configuration,1), (Hive,2), (system,1), (provides,1), (Hadoop-supported,1), (pre-built,...

以上连起来就是

scala> val ds=spark.read.textFile("README.md").flatMap(a=>a.split("\\s+")).groupByKey(a=>a).countds: org.apache.spark.sql.Dataset[(String, Long)] = [value: string, count(1): bigint]

如果Dataset再转为利用经典的rdd的方式则可以

scala> val ds2rdd1=spark.read.textFile("README.md").rdd.flatMap(a=>a.split("\\s+")).map(a=>(a,1)).reduceByKey(_+_)ds2rdd1: org.apache.spark.rdd.RDD[(String, Int)] = ShuffledRDD[128] at reduceByKey at <console>:23scala> val ds2rdd2=spark.read.textFile("README.md").flatMap(a=>a.split("\\s+")).map(a=>(a,1)).rdd.reduceByKey(_+_)ds2rdd2: org.apache.spark.rdd.RDD[(String, Int)] = ShuffledRDD[135] at reduceByKey at <console>:23