tachyon与hdfs,以及spark整合

最新推荐文章于 2019-10-26 15:02:13 发布

原创最新推荐文章于 2019-10-26 15:02:13 发布 · 2.7w 阅读

2 ·

CC 4.0 BY-SA版权

文章标签：

#spark #tachyon #hdfs #分布式 #测试

spark 同时被 2 个专栏收录

62 篇文章

订阅专栏

tachyon

4 篇文章

订阅专栏

本文介绍如何在Tachyon0.7.1中整合HDFS并进行伪分布式集群安装与测试。通过配置Tachyon环境变量指向HDFS地址，实现从HDFS读取文件并加载到Tachyon内存中。尽管遇到读取失败的问题，但通过检查发现文件已成功加载。

Tachyon 0.7.1伪分布式集群安装与测试:
http://blog.youkuaiyun.com/stark_summer/article/details/48321605
从官方文档得知，Spark 1.4.x和Tachyon 0.6.4版本兼容，而最新版的Tachyon 0.7.1和Spark 1.5.x兼容，目前所用的Spark为1.4.1，tachyon为 0.7.1

tachyon 与 hdfs整合

修改tachyon-env.sh

export TACHYON_UNDERFS_ADDRESS=hdfs://master:8020
Dtachyon.data.folder=$TACHYON_UNDERFS_ADDRESS/tmp/tachyon/data

上传文件到hdfs

 hadoop fs -put /home/cluster/data/test/bank/ /data/spark/

 hadoop fs -ls /data/spark/bank/
Found 3 items
-rw-r--r--   3 wangyue supergroup    4610348 2015-09-11 20:02 /data/spark/bank/bank-full.csv
-rw-r--r--   3 wangyue supergroup       3864 2015-09-11 20:02 /data/spark/bank/bank-names.txt
-rw-r--r--   3 wangyue supergroup     461474 2015-09-11 20:02 /data/spark/bank/bank.csv

通过tachyon 读取/data/spark/bank/bank-full.csv文件

val bankFullFile = sc.textFile("tachyon://master:19998/data/spark/bank/bank-full.csv/bank-full.csv")
2015-09-11 20:08:20,136 INFO  [main] storage.MemoryStore (Logging.scala:logInfo(59)) - ensureFreeSpace(177384) called with curMem=630803, maxMem=257918238
2015-09-11 20:08:20,137 INFO  [main] storage.MemoryStore (Logging.scala:logInfo(59)) - Block broadcast_3 stored as values in memory (estimated size 173.2 KB, free 245.2 MB)
2015-09-11 20:08:20,154 INFO  [main] storage.MemoryStore (Logging.scala:logInfo(59)) - ensureFreeSpace(17665) called with curMem=808187, maxMem=257918238
2015-09-11 20:08:20,155 INFO  [main] storage.MemoryStore (Logging.scala:logInfo(59)) - Block broadcast_3_piece0 stored as bytes in memory (estimated size 17.3 KB, free 245.2 MB)
2015-09-11 20:08:20,156 INFO  [sparkDriver-akka.actor.default-dispatcher-2] storage.BlockManagerInfo (Logging.scala:logInfo(59)) - Added broadcast_3_piece0 in memory on localhost:41040 (size: 17.3 KB, free: 245.9 MB)
2015-09-11 20:08:20,157 INFO  [main] spark.SparkContext (Logging.scala:logInfo(59)) - Created broadcast 3 from textFile at <console>:21
bankFullFile: org.apache.spark.rdd.RDD[String] = MapPartitionsRDD[7] at textFile at <console>:21

count

bankFullFile.count()
但是发现报错如下:
2015-09-11 21:34:31,494 WARN  [Executor task launch worker-6]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,495 WARN  [Executor task launch worker-6]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,489 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,495 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,495 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,495 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,495 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,495 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,496 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,496 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,496 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,496 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,496 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,496 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing
2015-09-11 21:34:31,496 WARN  [Executor task launch worker-7]  (RemoteBlockInStream.java:retrieveByteBufferFromRemoteMachine(320)) - Read nothing

感觉错误很诡异,有人知道这是什么原因？tell me why?

但是我在tachyon 文件系统中可以看到如下内容:

./bin/tachyon tfs ls /data/spark/bank/bank-full.csv/
4502.29 KB09-11-2015 20:09:02:078  Not In Memory  /data/spark/bank/bank-full.csv/bank-full.csv

而bank-full.csv在hdfs文件是

hadoop fs -ls /data/spark/bank/
Found 3 items
-rw-r--r--   3 wangyue supergroup    4610348 2015-09-11 20:02 /data/spark/bank/bank-full.csv
-rw-r--r--   3 wangyue supergroup       3864 2015-09-11 20:02 /data/spark/bank/bank-names.txt
-rw-r--r--   3 wangyue supergroup     461474 2015-09-11 20:02 /data/spark/bank/bank.csv

其实Tachyon本身将bank-full.csv文件加载到了内存，并存放到自身的文件系统里面：tachyon://master:19998/data/spark/bank/bank-full.csv/bank-full.csv”
Tachyon的conf/tachyon-env.sh文件里面配置的，通过export TACHYON_UNDERFS_ADDRESS=hdfs://master:8020配置,这样tachyon://localhost:19998就可以获取hdfs文件指定路径文件

好吧，那我就先通过hdfs方式读取文件然后保存到tachyon

scala> val bankfullfile =  sc.textFile("/data/spark/bank/bank-full.csv")
scala> bankfullfile.count
res0: Long = 45212

scala> bankfullfile.saveAsTextFile("tachyon://master:19998/data/spark/bank/newbankfullfile")

未完成,待续~