spark implementation hadoop setup,cleanup

 

写道

 

def main(args: Array[String]) {
val sc = new SparkContext("local", "xxx")
val inputData = sc.textFile("hdfs://master:8020/data/spark/user-history-data")
val lines = inputData.map(line => (line, line.length))

val result = lines.mapPartitions { valueIterator =>
if (valueIterator.isEmpty) {
Iterator[String]()
} else {
val transformedItem = new ListBuffer[String]() //setup ListBuffer
val fs: FileSystem = FileSystem.get(new Configuration()) //setup FileSystem

valueIterator.map { item =>
transformedItem += item._1 +":"+item._2
val outputFile = fs.create(new Path("/home/xxx/opt/data/spark/" + item._1.substring(0,item._1.indexOf("\t")) + ".txt"))
outputFile.write((item._1 +":"+item._2).getBytes())
if (!valueIterator.hasNext) {
transformedItem.clear() //cleanup transformedItem
outputFile.close() //cleanup outputFile
fs.close() //cleanup fs
}
transformedItem
}
}
}

result.foreach(println(_))
sc.stop()

 

将hdfs数据：
zhangsan 1 2015-07-30 20:01:01 127.0.0.1
zhangsan 2 2015-07-30 20:01:01 127.0.0.1
zhangsan 3 2015-07-30 20:01:01 127.0.0.1
zhangsan 4 2015-07-31 20:01:01 127.0.0.1
zhangsan 5 2015-07-31 20:21:01 127.0.0.1
lisi 1 2015-07-30 21:01:01 127.0.0.1
lisi 2 2015-07-30 22:01:01 127.0.0.1
lisi 3 2015-07-31 23:31:01 127.0.0.1
lisi 4 2015-07-31 22:21:01 127.0.0.1
lisi 5 2015-07-31 23:11:01 127.0.0.1
wangwu 1 2015-07-30 21:01:01 127.0.0.1
wangwu 2 2015-07-30 22:01:01 127.0.0.1
wangwu 3 2015-07-31 23:31:01 127.0.0.1
wangwu 4 2015-07-31 22:21:01 127.0.0.1
wangwu 5 2015-07-31 23:11:01 127.0.0.1
读取到spark中，并统计每行长度，再将数据写到本地的文件中（文件名称以每行第一个单词）
最终实现hadoop中setup, cleanup
强烈阅读如下链接：
http://mail-archives.apache.org/mod_mbox/spark-user/201407.mbox/%3CCAPH-c_O9kQO6yJ4khXUVdO=+D4vj=JfG2tP9eqn5RPko=dRNAg@mail.gmail.com%3E
http://blog.cloudera.com/blog/2014/09/how-to-translate-from-mapreduce-to-apache-spark/
http://apache-spark-user-list.1001560.n3.nabble.com/how-to-split-RDD-by-key-and-save-to-different-path-td11887.html#a11983
http://stackoverflow.com/questions/24520225/writing-to-hadoop-distributed-file-system-multiple-times-with-spark