Spark修炼之道（进阶篇）——Spark入门到精通：第十四节 Spark Streaming 缓存、Checkpoint机制

通过前面一系列的课程介绍，我们知道DStream是由一系列的RDD构成的，它同一般的RDD一样，也可以将流式数据持久化到内容当中，采用的同样是persisit方法，调用该方法后DStream将持久化所有的RDD数据。这对于一些需要重复计算多次或数据需要反复被使用的DStream特别有效。像reduceByWindow、reduceByKeyAndWindow等基于窗口操作的方法，它们默认都是有persisit操作的。reduceByKeyAndWindow方法源码具体如下：

val cleanedReduceFunc = ssc.sc.clean(reduceFunc) val cleanedInvReduceFunc = ssc.sc.clean(invReduceFunc) val cleanedFilterFunc = if (filterFunc != null) Some(ssc.sc.clean(filterFunc)) else None new ReducedWindowedDStream[K, V]( self, cleanedReduceFunc, cleanedInvReduceFunc, cleanedFilterFunc, windowDuration, slideDuration, partitioner }

从上面的方法来看，它最返回的是一个ReducedWindowedDStream对象，跳到该类的源码中可以看到在其主构造函数中包含下面两段代码：

//默认被缓存到内存当中 // Persist RDDs to memory by default as these RDDs are going to be reused. super.persist(StorageLevel.MEMORY_ONLY_SER) reducedStream.persist(StorageLevel.MEMORY_ONLY_SER) }

通过上面的代码我们可以看到，通过窗口操作产生的DStream不需要开发人员手动去调用persist方法，Spark会自动帮我们将数据缓存当内存当中。同一般的RDD类似，DStream支持的persisit级别为：

通过前期对Spark Streaming的理解，我们知道，Spark Streaming应用程序如果不手动停止，则将一直运行下去，在实际中应用程序一般是24小时*7天不间断运行的，因此Streaming必须对诸如系统错误、JVM出错等与程序逻辑无关的错误（failures ）具体很强的弹性，具备一定的非应用程序出错的容错性。Spark Streaming的Checkpoint机制便是为此设计的，它将足够多的信息checkpoint到某些具备容错性的存储系统如HDFS上，以便出错时能够迅速恢复。有两种数据可以chekpoint：

（1）Metadata checkpointing
将流式计算的信息保存到具备容错性的存储上如HDFS，Metadata Checkpointing适用于当streaming应用程序Driver所在的节点出错时能够恢复，元数据包括：
Configuration（配置信息） - 创建streaming应用程序的配置信息
DStream operations - 在streaming应用程序中定义的DStreaming操作
Incomplete batches - 在列队中没有处理完的作业

（2）Data checkpointing
将生成的RDD保存到外部可靠的存储当中，对于一些数据跨度为多个bactch的有状态tranformation操作来说，checkpoint非常有必要，因为在这些transformation操作生成的RDD对前一RDD有依赖，随着时间的增加，依赖链可能会非常长，checkpoint机制能够切断依赖链，将中间的RDD周期性地checkpoint到可靠存储当中，从而在出错时可以直接从checkpoint点恢复。

具体来说，metadata checkpointing主要还是从drvier失败中恢复，而Data Checkpoing用于对有状态的transformation操作进行checkpointing

Checkpointing具体的使用方式时通过下列方法：

//checkpointDirectory为checkpoint文件保存目录

streamingContext.checkpoint(checkpointDirectory)

import org.apache.spark.rdd.RDD import org.apache.spark.streaming.{Time, Seconds, StreamingContext} import org.apache.spark.util.IntParam * Counts words in text encoded with UTF8 received from the network every second. * Usage: RecoverableNetworkWordCount hostname port checkpoint-directory output-file * hostname and port describe the TCP server that Spark Streaming would connect to receive * data. checkpoint-directory directory to HDFS-compatible file system which checkpoint data * output-file file to which the word counts will be appended * checkpoint-directory and output-file must be absolute paths * To run this on your local machine, you need to first run a Netcat server * `$ nc -lk 9999` * and run the example as * `$ ./bin/run-example org.apache.spark.examples.streaming.RecoverableNetworkWordCount \ * localhost 9999 ~/checkpoint/ ~/out` * If the directory ~/checkpoint/ does not exist (e.g. running for the first time), it will create * a new StreamingContext (will print "Creating new context" to the console). Otherwise, if * checkpoint data exists in ~/checkpoint/, then it will create StreamingContext from * the checkpoint data. * Refer to the online documentation for more details. object RecoverableNetworkWordCount { def createContext(ip: String, port: Int, outputPath: String, checkpointDirectory: String) : StreamingContext = {
//程序第一运行时会创建该条语句，如果应用程序失败，则会从checkpoint中恢复，该条语句不会执行 println("Creating new context") val outputFile = new File(outputPath) if (outputFile.exists()) outputFile.delete() val sparkConf = new SparkConf().setAppName("RecoverableNetworkWordCount").setMaster("local[4]") // Create the context with a 1 second batch size val ssc = new StreamingContext(sparkConf, Seconds(1)) ssc.checkpoint(checkpointDirectory) //将socket作为数据源 val lines = ssc.socketTextStream(ip, port) val words = lines.flatMap(_.split(" ")) val wordCounts = words.map(x = (x, 1)).reduceByKey(_ + _) wordCounts.foreachRDD((rdd: RDD[(String, Int)], time: Time) = { val counts = "Counts at time " + time + " " + rdd.collect().mkString("[", ", ", "]") println(counts) println("Appending to " + outputFile.getAbsolutePath) Files.append(counts + "\n", outputFile, Charset.defaultCharset()) //将String转换成Int private object IntParam { def unapply(str: String): Option[Int] = { try { Some(str.toInt) } catch { case e: NumberFormatException = None def main(args: Array[String]) { if (args.length != 4) { System.err.println("You arguments were " + args.mkString("[", ", ", "]")) System.err.println( |Usage: RecoverableNetworkWordCount hostname port checkpoint-directory | output-file . hostname and port describe the TCP server that Spark | Streaming would connect to receive data. checkpoint-directory directory to | HDFS-compatible file system which checkpoint data output-file file to which the | word counts will be appended |In local mode, master should be local[n] with n 1 |Both checkpoint-directory and output-file must be absolute paths """.stripMargin System.exit(1) val Array(ip, IntParam(port), checkpointDirectory, outputPath) = args //getOrCreate方法，从checkpoint中重新创建StreamingContext对象或新创建一个StreamingContext对象 val ssc = StreamingContext.getOrCreate(checkpointDirectory, () = { createContext(ip, port, outputPath, checkpointDirectory) ssc.start() ssc.awaitTermination() }

输入参数配置如下：

运行状态图如下：

首次运行时：

Creating new context 15/11/30 07:20:32 WARN MetricsSystem: Using default name DAGScheduler for source because spark.app.id is not set. 15/11/30 07:20:33 WARN SizeEstimator: Failed to check whether UseCompressedOops is set; assuming yes Counts at time 1448896840000 ms [] Appending to /root/out2 15/11/30 07:20:47 WARN BlockManager: Block input-0-1448896847000 replicated to only 0 peer(s) instead of 1 peers Counts at time 1448896850000 ms [(Spark,1), (Context,1)]

手动将程序停止，然后重新运行

//这时从checkpoint目录中读取元数据信息，进行StreamingContext的恢复

Counts at time 1448897070000 ms []

Appending to /root/out2

Counts at time 1448897080000 ms []

Appending to /root/out2

Counts at time 1448897090000 ms []

Appending to /root/out2

15/11/30 07:24:58 WARN BlockManager: Block input-0-1448897098600 replicated to only 0 peer(s) instead of 1 peers

[Stage 8: (0 + 0) / 4]Counts at time 1448897100000 ms [(Spark,1), (Context,1)]

Appending to /root/out2

Spark RDD详解 —— RDD特性、lineage、缓存、checkpoint、依赖关系 RDD（Resilient Distributed Datasets）弹性的分布式数据集，又称Spark core，它代表一个只读的、不可变、可分区，里面的元素可分布式并行计算的数据集。