1、动手实战和调试Spark文件操作

 

　　这里，我以指定executor-memory参数的方式，启动spark-shell。

 

启动hadoop集群

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ jps
8457 Jps
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ sbin/start-dfs.sh

 

 

启动spark集群

spark@SparkSingleNode:/usr/local/spark/spark-1.5.2-bin-hadoop2.6$ sbin/start-all.sh

 

spark@SparkSingleNode:/usr/local/spark/spark-1.5.2-bin-hadoop2.6/bin$ ./spark-shell --master spark://SparkSingleNode:7077 --executor-memory 1g

　　在命令行中，我指定了spark-shell运行时暂时用的每个机器上executor的内存大小为1GB。

 

从HDFS上读取该文件

scala> val rdd1 = sc.textFile("/README.md")

或

scala> val rdd1 = sc.textFile("hdfs:SparkSingleNode:9000/README.md")

 

返回，MapPartitionsRDD

 

使用，toDebugString，可以查看其lineage的关系。

rdd1: org.apache.spark.rdd.RDD[String] = MapPartitionsRDD[1] at textFile at <console>:21

scala> rdd1.toDebugString
16/09/26 22:47:01 INFO mapred.FileInputFormat: Total input paths to process : 1
res0: String = 
(2) MapPartitionsRDD[1] at textFile at <console>:21 []
| /README.md HadoopRDD[0] at textFile at <console>:21 []

scala>

 

可以看出，MapPartitionsRDD是HadoopRDD转换而来的。

 

hadoopFile，这个方法，产生HadoopRDD

map，这个方法，产生MapPartitionsRDD

 

从源码分析过程

 

 

 

 

scala> val result = rdd1.flatMap(_.split(" ")).map((_,1)).reduceByKey(_+_).collect

 

le>:23, took 15.095588 s
result: Array[(String, Int)] = Array((package,1), (this,1), (Version"](http://spark.apache.org/docs/latest/building-spark.html#specifying-the-hadoop-version),1), (Because,1), (Python,2), (cluster.,1), (its,1), ([run,1), (general,2), (have,1), (pre-built,1), (locally.,1), (locally,2), (changed,1), (sc.parallelize(1,1), (only,1), (several,1), (This,2), (basic,1), (Configuration,1), (learning,,1), (documentation,3), (YARN,,1), (graph,1), (Hive,2), (first,1), (["Specifying,1), ("yarn-client",1), (page](http://spark.apache.org/documentation.html),1), ([params]`.,1), (application,1), ([project,2), (prefer,1), (SparkPi,2), (<http://spark.apache.org/>,1), (engine,1), (version,1), (file,1), (documentation,,1), (MASTER,1), (example,3), (distribution.,1), (are,1), (params,1), (scala>,1), (DataFram...
scala>

 

不可这样使用toDebugString

scala> result.toDebugString
<console>:26: error: value toDebugString is not a member of Array[(String, Int)]
result.toDebugString

 

 

scala> val wordcount = rdd1.flatMap(_.split(" ")).map((_,1)).reduceByKey(_+_)
wordcount: org.apache.spark.rdd.RDD[(String, Int)] = ShuffledRDD[10] at reduceByKey at <console>:23

scala> wordcount.toDebugString
res3: String = 
(2) ShuffledRDD[10] at reduceByKey at <console>:23 []
+-(2) MapPartitionsRDD[9] at map at <console>:23 []
| MapPartitionsRDD[8] at flatMap at <console>:23 []
| MapPartitionsRDD[1] at textFile at <console>:21 []
| /README.md HadoopRDD[0] at textFile at <console>:21 []

scala>

 

或者

 

 疑问：为什么没有MappedRDD？难道是版本问题？？

 

 

2、动手实战操作搜狗日志文件

本节中所用到的内容是来自搜狗实验室，网址为：http://www.sogou.com/labs/dl/q.html

我们使用的是迷你版本的tar.gz格式的文件，其大小为87K，下载后如下所示：

因为，考虑我的机器内存的自身情况。

 

或者

spark@SparkSingleNode:~$ wget http://download.labs.sogou.com/dl/sogoulabdown/SogouQ/SogouQ2012.mini.tar.gz

spark@SparkSingleNode:~$ tar -zxvf SogouQ2012.mini.tar.gz

 

查看它的部分内容

spark@SparkSingleNode:~$ head SogouQ.mini 

 

该文件的格式如下所示：

访问时间 \t 用户ID \t 查询词 \t 该URL在返回结果中的排名 \ t用户点击的顺序号 \t 用户点击的URL

 

开启hdfs和spark集群

 

把解压后的文件上传到hdfs的/目录下

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ bin/hadoop fs -copyFromLocal ~/SogouQ.mini /

 

 

开启spark-shell

spark@SparkSingleNode:/usr/local/spark/spark-1.5.2-bin-hadoop2.6/bin$ ./spark-shell --master spark://SparkSingleNode:7077

 

接下来 我们使用Spark获得搜索结果排名第一同时点击结果排名也是第一的数据量，也就是第四列值为1同时第五列的值也为1的总共的记录的个数。

 读取SogouQ.mini文件

scala> val soGouQRdd = sc.textFile("hdfs://SparkSingleNode:9000/SogouQ.mini")

scala> soGouQRdd.count

took 10.753423 s
res0: Long = 2000

可以看出，count之后有2000条记录

 

首先过滤出有效的数据：

scala> val mapSoGouQRdd = soGouQRdd.map((_.split("\t"))).filter(_.length == 6)
mapSoGouQRdd: org.apache.spark.rdd.RDD[Array[String]] = MapPartitionsRDD[3] at filter at <console>:23

scala> mapSoGouQRdd.count

 

took 2.175379 s
res1: Long = 2000

可以发现该文件中的数据都是有效数据。

 

该文件的格式如下所示：
访问时间 \t 用户ID \t 查询词 \t 该URL在返回结果中的排名 \ t用户点击的顺序号 \t 用户点击的URL

 

下面使用spark获得搜索结果排名第一同时点击结果排名也是第一的数据量：

scala> val filterSoGouQRdd = mapSoGouQRdd.filter(_(3).toInt == 1).filter(_(4).toInt == 1)
filterSoGouQRdd: org.apache.spark.rdd.RDD[Array[String]] = MapPartitionsRDD[5] at filter at <console>:25

scala> filterSoGouQRdd.count

 

 

可以发现搜索结果排名第一同时点击结果排名也是第一的数据量为794条；

 

使用toDebugString查看一下其lineage：

scala> filterSoGouQRdd.toDebugString

 

res3: String = 
(2) MapPartitionsRDD[5] at filter at <console>:25 []
| MapPartitionsRDD[4] at filter at <console>:25 []
| MapPartitionsRDD[3] at filter at <console>:23 []
| MapPartitionsRDD[2] at map at <console>:23 []
| MapPartitionsRDD[1] at textFile at <console>:21 []
| hdfs://SparkSingleNode:9000/SogouQ.mini HadoopRDD[0] at textFile at <console>:21 []

scala>

 

为什么没有？

 

 HadoopRDD->MappedRDD->MappedRDD->FilteredRDD->FilteredRDD->FilteredRDD

 

 3、搜狗日志文件深入实战

 下面看，用户ID查询次数排行榜：

该文件的格式如下所示：
访问时间 \t 用户ID \t 查询词 \t 该URL在返回结果中的排名 \ t用户点击的顺序号 \t 用户点击的URL

 

scala> val sortedSoGouQRdd = mapSoGouQRdd.map(x => (x(1),1)).reduceByKey(_+_).map(x => (x._2,x._1)).sortByKey(false).map(x => (x._2,x._1))

 

 

 对sortedSogouQRdd进行collect操作：（不要乱collect 会出现OOM的）

scala> sortedSoGouQRdd.collect

 

res4: Array[(String, Int)] = Array((f6492a1da9875f20e01ff8b5804dcc35,14), (e7579c6b6b9c0ea40ecfa0f425fc765a,11), (d3034ac9911c30d7cf9312591ecf990e,11), (5c853e91940c5eade7455e4a289722d6,10), (ec0363079f36254b12a5e30bdc070125,10), (828f91e6717213a65c97b694e6279201,9), (2a36742c996300d664652d9092e8a554,9), (439fa809ba818cee624cc8b6e883913a,9), (45c304b5f2dd99182451a02685252312,8), (5ea391fd07dbb616e9857a7d95f460e0,8), (596444b8c02b7b30c11273d5bbb88741,8), (a06830724b809c0db56263124b2bd142,8), (6056710d9eafa569ddc800fe24643051,7), (bc8cc0577bb80fafd6fad1ed67d3698e,7), (8897bbb7bdff69e80f7fb2041d83b17d,7), (41389fb54f9b3bec766c5006d7bce6a2,7), (b89952902d7821db37e8999776b32427,6), (29ede0f2544d28b714810965400ab912,6), (74033165c877f4082e14c1e94d1efff4,6), (833f242ff430c83d293980ec10a42484,6...
scala>

 

把结果保存在hdfs上：

scala> sortedSoGouQRdd.saveAsTextFile("hdfs://SparkSingleNode:9000/soGouQSortedResult.txt")

 

把这些，输出信息，看懂，深入，是大牛必经之路。

scala> sortedSoGouQRdd.saveAsTextFile("hdfs://SparkSingleNode:9000/soGouQSortedResult.txt")
16/09/27 10:08:34 INFO Configuration.deprecation: mapred.tip.id is deprecated. Instead, use mapreduce.task.id
16/09/27 10:08:34 INFO Configuration.deprecation: mapred.task.id is deprecated. Instead, use mapreduce.task.attempt.id
16/09/27 10:08:34 INFO Configuration.deprecation: mapred.task.is.map is deprecated. Instead, use mapreduce.task.ismap
16/09/27 10:08:34 INFO Configuration.deprecation: mapred.task.partition is deprecated. Instead, use mapreduce.task.partition
16/09/27 10:08:34 INFO Configuration.deprecation: mapred.job.id is deprecated. Instead, use mapreduce.job.id
16/09/27 10:08:35 INFO spark.SparkContext: Starting job: saveAsTextFile at <console>:28
16/09/27 10:08:35 INFO spark.MapOutputTrackerMaster: Size of output statuses for shuffle 0 is 155 bytes
16/09/27 10:08:35 INFO scheduler.DAGScheduler: Got job 5 (saveAsTextFile at <console>:28) with 2 output partitions
16/09/27 10:08:35 INFO scheduler.DAGScheduler: Final stage: ResultStage 10(saveAsTextFile at <console>:28)
16/09/27 10:08:35 INFO scheduler.DAGScheduler: Parents of final stage: List(ShuffleMapStage 9)
16/09/27 10:08:35 INFO scheduler.DAGScheduler: Missing parents: List()
16/09/27 10:08:35 INFO scheduler.DAGScheduler: Submitting ResultStage 10 (MapPartitionsRDD[13] at saveAsTextFile at <console>:28), which has no missing parents
16/09/27 10:08:35 INFO storage.MemoryStore: ensureFreeSpace(128736) called with curMem=105283, maxMem=560497950
16/09/27 10:08:35 INFO storage.MemoryStore: Block broadcast_8 stored as values in memory (estimated size 125.7 KB, free 534.3 MB)
16/09/27 10:08:36 INFO storage.MemoryStore: ensureFreeSpace(43435) called with curMem=234019, maxMem=560497950
16/09/27 10:08:36 INFO storage.MemoryStore: Block broadcast_8_piece0 stored as bytes in memory (estimated size 42.4 KB, free 534.3 MB)
16/09/27 10:08:36 INFO storage.BlockManagerInfo: Added broadcast_8_piece0 in memory on 192.168.80.128:33999 (size: 42.4 KB, free: 534.5 MB)
16/09/27 10:08:36 INFO spark.SparkContext: Created broadcast 8 from broadcast at DAGScheduler.scala:861
16/09/27 10:08:36 INFO scheduler.DAGScheduler: Submitting 2 missing tasks from ResultStage 10 (MapPartitionsRDD[13] at saveAsTextFile at <console>:28)
16/09/27 10:08:36 INFO scheduler.TaskSchedulerImpl: Adding task set 10.0 with 2 tasks
16/09/27 10:08:36 INFO scheduler.TaskSetManager: Starting task 0.0 in stage 10.0 (TID 14, 192.168.80.128, PROCESS_LOCAL, 1901 bytes)
16/09/27 10:08:36 INFO storage.BlockManagerInfo: Added broadcast_8_piece0 in memory on 192.168.80.128:59936 (size: 42.4 KB, free: 534.5 MB)
16/09/27 10:08:41 INFO scheduler.TaskSetManager: Starting task 1.0 in stage 10.0 (TID 15, 192.168.80.128, PROCESS_LOCAL, 1901 bytes)
16/09/27 10:08:41 INFO scheduler.TaskSetManager: Finished task 0.0 in stage 10.0 (TID 14) in 5813 ms on 192.168.80.128 (1/2)
16/09/27 10:08:43 INFO scheduler.DAGScheduler: ResultStage 10 (saveAsTextFile at <console>:28) finished in 7.719 s
16/09/27 10:08:43 INFO scheduler.DAGScheduler: Job 5 finished: saveAsTextFile at <console>:28, took 8.348232 s
16/09/27 10:08:43 INFO scheduler.TaskSetManager: Finished task 1.0 in stage 10.0 (TID 15) in 2045 ms on 192.168.80.128 (2/2)
16/09/27 10:08:43 INFO scheduler.TaskSchedulerImpl: Removed TaskSet 10.0, whose tasks have all completed, from pool

scala>

 

 

 

 

 

hdfs命令行查询：

part-0000：

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ bin/hadoop fs -text /soGouQSortedResult.txt/part-00000

 

 

hdfs命令行查询：

part-0000：

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ bin/hadoop fs -text /soGouQSortedResult.txt/part-00001

 

 我们通过hadoop命令把上述两个文件的内容合并起来：

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ bin/hadoop fs -getmerge hdfs://SparkSingleNode:9000/soGouQSortedResult.txt combinedSortedResult.txt      //注意，第二个参数，是本地文件的目录

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ bin/hadoop fs -ls /
Found 6 items
-rw-r--r-- 1 spark supergroup 3593 2016-09-18 10:15 /README.md
-rw-r--r-- 1 spark supergroup 216118 2016-09-27 09:17 /SogouQ.mini
drwxr-xr-x - spark supergroup 0 2016-09-26 21:17 /result
drwxr-xr-x - spark supergroup 0 2016-09-26 21:49 /resultDescSorted
drwxr-xr-x - spark supergroup 0 2016-09-27 10:08 /soGouQSortedResult.txt
drwx-wx-wx - spark supergroup 0 2016-09-09 16:28 /tmp
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ ls
bin etc libexec NOTICE.txt share
combinedSortedResult.txt include LICENSE.txt README.txt tmp
dfs lib logs sbin
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$

 

 

或者

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ bin/hdfs dfs -getmerge hdfs://SparkSingleNode:9000/soGouQSortedResult.txt combinedSortedResult.txt       //两者是等价的

spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ ls
bin etc lib LICENSE.txt NOTICE.txt sbin tmp
dfs include libexec logs README.txt share
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ cd bin
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0/bin$ ls
container-executor hdfs mapred.cmd yarn
hadoop hdfs.cmd rcc yarn.cmd
hadoop.cmd mapred test-container-executor
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0/bin$ cd hdfs
bash: cd: hdfs: Not a directory
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0/bin$ cd ..
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ bin/hdfs dfs -getmerge hdfs://SparkSingleNode:9000/soGouQSortedResult.txt combinedSortedResult.txt
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$ ls
bin etc libexec NOTICE.txt share
combinedSortedResult.txt include LICENSE.txt README.txt tmp
dfs lib logs sbin
spark@SparkSingleNode:/usr/local/hadoop/hadoop-2.6.0$

 

 

 

本文转自大数据躺过的坑博客园博客，原文链接：http://www.cnblogs.com/zlslch/p/5911131.html，如需转载请自行联系原作者