Spark生态之Alluxio学习6---集群版搭建和运行（Tachyon）

更多代码请见：https://github.com/xubo245/SparkLearning

环境：

hadoop-2.6.0

spark-1.5.2

scala-1.7

1解释

完成tachyon的集群配置与运行

前提：安装好java、hdfs等

2.下载
https://github.com/Alluxio/alluxio/releases
具体的：
https://github.com/Alluxio/alluxio/releases/download/v0.7.1/tachyon-0.7.1-hadoop2.6-bin.tar.gz
如果下载tachyon-0.7.1-bin.tar.g会出现版本不兼容，会报错：

Server IPC version 9 cannot communicate with client version 4

3.配置：【3】讲的比较详细

在conf/tachyon-env.sh配置：

export TACHYON_MASTER_ADDRESS=<strong>Master</strong>
#export TACHYON_UNDERFS_ADDRESS=$TACHYON_HOME/underFSStorage
export TACHYON_UNDERFS_ADDRESS=hdfs://<span style="font-family: Arial, Helvetica, sans-serif; font-size: 12px;"><strong>Master</strong></span>:9000
export TACHYON_WORKER_MEMORY_SIZE=256MB

Master为集群master IP，或者自己在本地设置好映射

4.安装：

$ cd tachyon
$ ./bin/tachyon format
$ ./bin/tachyon-start.sh # use the right parameters here. e.g. all Mount

一般start需要sudo才能mount：

./bin/tachyon-start.sh all SudoMount

每个节点都要输入密码，好麻烦

5.查看：

在Master的19999端口可以查看，六个worker启动了三个，

解决办法：http://blog.csdn.net/xubo245/article/details/53056482

6.使用：

上传：

hadoop@Master:~/cloud/tachyon-0.7.1-hadoop2.6-bin$ ./bin/tachyon tfs copyFromLocal test/2.txt /
Copied test/2.txt to /

展示没问题

7.记录

ipc版本不同：

hadoop@Master:~/cloud/tachyon-0.7.1$ ./bin/tachyon format
Connecting to Mcnode1 as hadoop...
Formatting Tachyon Worker @ Mcnode1
Connection to mcnode1 closed.
Connecting to Mcnode2 as hadoop...
Formatting Tachyon Worker @ Mcnode2
Connection to mcnode2 closed.
Connecting to Mcnode3 as hadoop...
dirname: missing operand
Try 'dirname --help' for more information.
Formatting Tachyon Worker @ Mcnode3
/home/hadoop/cloud/tachyon-0.7.1/bin/tachyon: line 264: /../bin/java: No such file or directory
Connection to mcnode3 closed.
Connecting to Mcnode4 as hadoop...
dirname: missing operand
Try 'dirname --help' for more information.
Formatting Tachyon Worker @ Mcnode4
/home/hadoop/cloud/tachyon-0.7.1/bin/tachyon: line 264: /../bin/java: No such file or directory
Connection to mcnode4 closed.
Connecting to Mcnode5 as hadoop...
dirname: missing operand
Try 'dirname --help' for more information.
Formatting Tachyon Worker @ Mcnode5
/home/hadoop/cloud/tachyon-0.7.1/bin/tachyon: line 264: /../bin/java: No such file or directory
Connection to mcnode5 closed.
Connecting to Mcnode6 as hadoop...
Formatting Tachyon Worker @ Mcnode6
Connection to mcnode6 closed.
Formatting Tachyon Master @ localhost
Exception in thread "main" java.lang.IllegalArgumentException: All eligible Under File Systems were unable to create an instance for the given path: hdfs://Master:9000/tmp/tachyon/data
java.lang.RuntimeException: org.apache.hadoop.ipc.RemoteException:Server IPC version 9 cannot communicate with client version 4

at tachyon.underfs.UnderFileSystemRegistry.create(UnderFileSystemRegistry.java:132)
at tachyon.underfs.UnderFileSystem.get(UnderFileSystem.java:99)
at tachyon.underfs.UnderFileSystem.get(UnderFileSystem.java:83)
at tachyon.Format.formatFolder(Format.java:38)
at tachyon.Format.main(Format.java:77)

参考

【1】http://alluxio.org/documentation/v0.7.1/

【2】https://github.com/xubo245/SparkLearning

【3】 http://alluxio.org/documentation/v0.7.1/Running-Tachyon-on-a-Cluster.html