外部命令 s3-dist-cp 通过 scala.sys.process API 在 spark-scala 中执行

Question

>当我在 unix shell/终端中运行所有这 3 个命令时，它们都工作正常，返回退出状态为 0

unix_shell> ls -la
unix_shell> hadoop fs -ls /user/hadoop/temp
unix_shell> s3-dist-cp --src ./abc.txt --dest s3://bucket/folder/

现在我尝试通过 scala 进程 api 作为外部进程运行这些相同的命令，示例代码如下所示：

import scala.sys.process._
val cmd_1 = "ls -la"
val cmd_2 = "hadoop fs -ls /user/hadoop/temp/"
val cmd_3 = "/usr/bin/s3-dist-cp --src /tmp/sample.txt --dest s3://bucket/folder/"
val cmd_4 = "s3-dist-cp --src /tmp/sample.txt --dest s3://bucket/folder/"
val exitCode_1 = (stringToProcess(cmd_1)).! // works fine and produces result
val exitCode_2 = (stringToProcess(cmd_2)).! // works fine and produces result
val exitCode_3 = (stringToProcess(cmd_3)).! // **it just hangs, yielding nothing**
val exitCode_4 = (stringToProcess(cmd_4)).! // **it just hangs, yielding nothing**

上述cmd_3和cmd_4之间的区别只是绝对路径。我在 spark-submit 脚本中明确传递相关依赖项，如下所示

--jars hdfs:///user/hadoop/s3-dist-cp.jar

您的意见/建议将有所帮助。谢谢！

Answer 1

实际上 scala 进程在 spark 上下文之外工作，所以为了成功运行该 s3-dist-cp 命令，

我所做的只是在启动包装 s3-dist-cp 命令的 scala 进程之前停止 spark 上下文，完整的工作代码如下：

    logger.info("Moving ORC files from HDFS to S3 !!")
    import scala.sys.process._
    logger.info("stopping spark context..##")
    val spark = IngestionContext.sparkSession
    spark.stop()
    logger.info("spark context stopped..##")
    logger.info("sleeping for 10 secs")
    Thread.sleep(10000) // this sleep is not required, this was just for debugging purpose, you can remove this in your final code.
    logger.info("woke up after sleeping for 10 secs")
    try {
      /**
       * following is the java version, off course you need take care of few imports
       */
      //val pb = new java.lang.ProcessBuilder("s3-dist-cp", "--src", INGESTED_ORC_DIR, "--dest", "s3:/" + paramMap(Storage_Output_Path).substring(4) + "_temp", "--srcPattern", ".*\.orc")
      //val pb = new java.lang.ProcessBuilder("hadoop", "jar", "/usr/share/aws/emr/s3-dist-cp/lib/s3-dist-cp.jar", "--src", INGESTED_ORC_DIR, "--dest", "s3:/" + paramMap(Storage_Output_Path).substring(4) + "_temp", "--srcPattern", ".*\.orc")
      //pb.directory(new File("/tmp"))
      //pb.inheritIO()
      //pb.redirectErrorStream(true)
      //val process = pb.start()
      //val is = process.getInputStream()
      //val isr = new InputStreamReader(is)
      //val br = new BufferedReader(isr)
      //var line = ""
      //logger.info("printling lines:")
      //while (line != null) {
      //  line = br.readLine()
      //  logger.info("line=[{}]", line)
      //}
      //logger.info("process goes into waiting state")
      //logger.info("Waited for: " + process.waitFor())
      //logger.info("Program terminated!")
      /**
       * following is the scala version
       */
      val S3_DIST_CP = "s3-dist-cp"
      val INGESTED_ORC_DIR = S3Util.getSaveOrcPath()
      // listing out all the files
      //val s3DistCpCmd = S3_DIST_CP + " --src " + INGESTED_ORC_DIR + " --dest " + paramMap(Storage_Output_Path).substring(4) + "_temp --srcPattern .*\.orc"
      //-Dmapred.child.java.opts=-Xmx1024m -Dmapreduce.job.reduces=2
      val cmd = S3_DIST_CP + " --src " + INGESTED_ORC_DIR + " --dest " + "s3:/" + paramMap(Storage_Output_Path).substring(4) + "_temp --srcPattern .*\.orc"
      //val cmd = "hdfs dfs -cp -f " + INGESTED_ORC_DIR + "/* " + "s3:/" + paramMap(Storage_Output_Path).substring(4) + "_temp/"
      //val cmd = "hadoop distcp " + INGESTED_ORC_DIR + "/ s3:/" + paramMap(Storage_Output_Path).substring(4) + "_temp_2/"
      logger.info("full hdfs to s3 command : [{}]", cmd)
      // command execution
      val exitCode = (stringToProcess(cmd)).!
      logger.info("s3_dist_cp command exit code: {} and s3 copy got " + (if (exitCode == 0) "SUCCEEDED" else "FAILED"), exitCode)
    } catch {
      case ex: Exception =>
        logger.error(
          "there was an exception while copying orc file to s3 bucket. {} {}",
          "", ex.getMessage, ex)
        throw new IngestionException("s3 dist cp command failure", null, Some(StatusEnum.S3_DIST_CP_CMD_FAILED))
    }

虽然上面的代码完全按预期工作，但也有其他观察结果，如下所示，这很奇怪：

而不是使用这个

val exitCode = (stringToProcess(cmd)).!

如果你使用这个

val exitCode = (stringToProcess(cmd)).!!

请注意单 ！ 和双!,的区别 单！ 只返回退出代码，而双 ！！ 返回进程执行的输出

所以在单 ！ 上面的代码工作正常，在双!,的情况下，它也可以工作，但它在 S3 存储桶中生成了太多文件和副本，而不是原始文件的数量。

至于 spark-submit 命令，无需担心 --driver-class-path 甚至 --jars 选项，因为我没有传递任何依赖项。

Answer 2

看起来你所做的是对的。看这里https://github.com/gorros/spark-scala-tips/blob/master/README.md

import scala.sys.process._
def s3distCp(src: String, dest: String): Unit = {
    s"s3-dist-cp --src $src --dest $dest".!
}

请检查此说明...我想知道你是否是这种情况。

关于您的--jars /usr/lib/hadoop/client/*.jar

您可以使用这样的命令附加与s3-dist-cp相关的 JAR tr。 查看我的答案

--jars $(echo /dir_of_jars/*.jar | tr ' ' ',')

---

注意：为了能够使用此方法，您需要添加Hadoop应用程序，并且需要在客户端或本地模式下运行Spark，因为s3-dist-cp在从属节点上不可用。如果要在集群模式下运行，请在引导期间将s3-dist-cp命令复制到从属服务器。