如何在 Spark 中设置 FTP 被动模式？...从 FTP 服务器读取文件

Question

我正在像这样将FTP server 的文件读入spark rdd

val rdd = spark.sparkContext.textFile("ftp://anonymous:pwd@<hostname>/data.gz")
rdd.count
...

当我从本地机器 (Mac) 运行 spark 应用程序时，这实际上有效，但是当我尝试从 docker 容器 运行相同的应用程序时（在Mac），我收到以下异常，

Exception in thread "main" org.apache.commons.net.ftp.FTPConnectionClosedException: Connection closed without indication.
    at org.apache.commons.net.ftp.FTP.__getReply(FTP.java:313)
    at org.apache.commons.net.ftp.FTP.__getReply(FTP.java:290)
    at org.apache.commons.net.ftp.FTP.sendCommand(FTP.java:479)
    at org.apache.commons.net.ftp.FTP.sendCommand(FTP.java:552)
    at org.apache.commons.net.ftp.FTP.sendCommand(FTP.java:601)
    at org.apache.commons.net.ftp.FTP.quit(FTP.java:809)
    at org.apache.commons.net.ftp.FTPClient.logout(FTPClient.java:979)
    at org.apache.hadoop.fs.ftp.FTPFileSystem.disconnect(FTPFileSystem.java:168)
    at org.apache.hadoop.fs.ftp.FTPFileSystem.getFileStatus(FTPFileSystem.java:415)
    at org.apache.hadoop.fs.Globber.getFileStatus(Globber.java:57)
    at org.apache.hadoop.fs.Globber.glob(Globber.java:252)
    at org.apache.hadoop.fs.FileSystem.globStatus(FileSystem.java:1676)
    at org.apache.hadoop.mapred.FileInputFormat.singleThreadedListStatus(FileInputFormat.java:259)
    at org.apache.hadoop.mapred.FileInputFormat.listStatus(FileInputFormat.java:229)
    at org.apache.hadoop.mapred.FileInputFormat.getSplits(FileInputFormat.java:315)
    at org.apache.spark.rdd.HadoopRDD.getPartitions(HadoopRDD.scala:205)
    at org.apache.spark.rdd.RDD.$anonfun$partitions$2(RDD.scala:276)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.partitions(RDD.scala:272)
    at org.apache.spark.rdd.MapPartitionsRDD.getPartitions(MapPartitionsRDD.scala:49)
    at org.apache.spark.rdd.RDD.$anonfun$partitions$2(RDD.scala:276)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.partitions(RDD.scala:272)
    at org.apache.spark.rdd.MapPartitionsRDD.getPartitions(MapPartitionsRDD.scala:49)
    at org.apache.spark.rdd.RDD.$anonfun$partitions$2(RDD.scala:276)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.partitions(RDD.scala:272)
    at org.apache.spark.MapOutputTrackerMaster.getPreferredLocationsForShuffle(MapOutputTracker.scala:626)
    at org.apache.spark.rdd.ShuffledRDD.getPreferredLocations(ShuffledRDD.scala:99)
    at org.apache.spark.rdd.RDD.$anonfun$preferredLocations$2(RDD.scala:300)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.preferredLocations(RDD.scala:300)
    at org.apache.spark.scheduler.DAGScheduler.getPreferredLocsInternal(DAGScheduler.scala:2098)
    at org.apache.spark.scheduler.DAGScheduler.getPreferredLocs(DAGScheduler.scala:2072)
    at org.apache.spark.SparkContext.getPreferredLocs(SparkContext.scala:1794)
    at org.apache.spark.rdd.DefaultPartitionCoalescer.currPrefLocs(CoalescedRDD.scala:180)
    at org.apache.spark.rdd.DefaultPartitionCoalescer$PartitionLocations.$anonfun$getAllPrefLocs$1(CoalescedRDD.scala:198)
    at scala.collection.IndexedSeqOptimized.foreach(IndexedSeqOptimized.scala:36)
    at scala.collection.IndexedSeqOptimized.foreach$(IndexedSeqOptimized.scala:33)
    at scala.collection.mutable.ArrayOps$ofRef.foreach(ArrayOps.scala:198)
    at org.apache.spark.rdd.DefaultPartitionCoalescer$PartitionLocations.getAllPrefLocs(CoalescedRDD.scala:197)
    at org.apache.spark.rdd.DefaultPartitionCoalescer$PartitionLocations.<init>(CoalescedRDD.scala:190)
    at org.apache.spark.rdd.DefaultPartitionCoalescer.coalesce(CoalescedRDD.scala:391)
    at org.apache.spark.rdd.CoalescedRDD.getPartitions(CoalescedRDD.scala:90)
    at org.apache.spark.rdd.RDD.$anonfun$partitions$2(RDD.scala:276)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.partitions(RDD.scala:272)
    at org.apache.spark.rdd.MapPartitionsRDD.getPartitions(MapPartitionsRDD.scala:49)
    at org.apache.spark.rdd.RDD.$anonfun$partitions$2(RDD.scala:276)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.partitions(RDD.scala:272)
    at org.apache.spark.rdd.MapPartitionsRDD.getPartitions(MapPartitionsRDD.scala:49)
    at org.apache.spark.rdd.RDD.$anonfun$partitions$2(RDD.scala:276)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.partitions(RDD.scala:272)
    at org.apache.spark.rdd.MapPartitionsRDD.getPartitions(MapPartitionsRDD.scala:49)
    at org.apache.spark.rdd.RDD.$anonfun$partitions$2(RDD.scala:276)
    at scala.Option.getOrElse(Option.scala:189)
    at org.apache.spark.rdd.RDD.partitions(RDD.scala:272)
    at org.apache.spark.SparkContext.runJob(SparkContext.scala:2158)
    at org.apache.spark.rdd.RDD.count(RDD.scala:1227)
    at com.mypackage.Myapp$.parseData(Myapp.scala:76)

在容器中，即使是ftp命令行实用程序也有同样的问题，但通过在ftp CLI中设置passive模式发现，我能够成功地将文件从FTP服务器传输到容器,

ftp <host>
...
ftp> passive
Passive mode on.
ftp> get data.gz
227 Entering Passive Mode ...
226 Transfer complete
20676672 bytes received in 25.53 secs (790.9552 kB/s)

所以我的问题是...如何设置passive mode 属性？...在​​使用param.spark.sparkContext.textFile("ftp://anonymous:pwd@<hostname>/data.gz") 在Spark 中读取文件时

Answer 1

我没有使用 Spark 的经验，所以我不知道它如何与 Hadoop 结合。但是在Hadoop中，你可以通过设置fs.ftp.data.connection.mode configuration option来设置FTP被动模式：

fs.ftp.data.connection.mode=PASSIVE_LOCAL_DATA_CONNECTION_MODE

您至少需要 Hadoop 2.9：https://issues.apache.org/jira/browse/HADOOP-13953