Flume-0.9.4分布式安装与配置手册

From:  http://www.iteblog.com/archives/911

个人在运行过程中遇到的问题解决方案是：

As people in cloudera mail list suggest, there are probable reasons of this error:

1. The HDFS safemode is turned on. Try to run hadoop fs -safemode leave and see if the error goes away.
2. Flume and Hadoop versions are mismatched. To check this replace the hadoop-core.jar in flume/lib directory with the one found in hadoop's installation folder.

　　 Flume是一个分布式、可靠、和高可用的海量日志采集、聚合和传输的系统。支持在日志系统中定制各类数据发送方，用于收集数据;同时， Flume提供对数据进行简单处理，并写到各种数据接受方(比如文本、HDFS、Hbase等)的能力 。
　　Flume主要有以下几类组件：
　　（1）、Master: 负责配置及通信管理，是集群的控制器，并支持多master节点；
　　（2）、Agent: 采集数据，Agent是flume中产生数据流的地方，同时Agent会将产生的数据流传输到Collector；
　　（3）、Collector: 用于对数据进行聚合（数据收集器），往往会产生一个更大的数据流，然后加载到storage（存储）中 。
　　简单地来说，就是Agent把采集到的数据定时发送给Collector，Collector接收Agent发送的数据并把数据写到指定的位置(比如文本、HDFS、Hbase等)。
　　这篇文章主要简单介绍如何部署Flume-0.9.4分布式环境，涉及到三台机器。它们的hostname分别为master,agent,collector。
　　 1、到官网下载Flume-0.9.4，并解压：

1 [wyp@master~]$ wget https://repository.cloudera.com/content/                  \
2       repositories/releases/com/cloudera/flume-distribution/0.9.4-cdh4.0.0/    \
3        flume-distribution-0.9.4-cdh4.0.0-bin.tar.gz                            \
4 [wyp@master~]$  tar -zxvf flume-distribution-0.9.4-cdh4.0.0-bin.tar.gz
5 [wyp@master~]$ cd flume-0.9.4-cdh3u3
6 [wyp@masterflume-0.9.4-cdh3u3]$

　　2、进入$FLUME_HOMNE/bin目录，将flume-env.sh.template重命名为flume-env.sh，并在flume-env.sh文件里面设置如下变量：

1 [wyp@masterflume-0.9.4-cdh3u3]$ cd bin
2 [wyp@masterbin]$ cp flume-env.sh.template flume-env.sh
3 [wyp@masterbin]$ vim flume-env.sh
4
5 export FLUME_HOME=/home/q/flume-0.9.4-cdh3u3
6 export FLUME_CONF_DIR=$FLUME_HOME/conf
7 export PATH=$PATH:$FLUME_HOME/bin
8 export JAVA_HOME=/usr/lib/jvm/java-6-sun

　　　　3、进入$FLUME_HOMNE/conf目录，将flume-site.xml.template重命名为flume-site.xml，并修改flume-site.xml配置文件

01 [wyp@masterbin]$ cd ../conf
02 [wyp@masterconf]$ cp flume-site.xml.template flume-site.xml
03 [wyp@masterconf]$ vim flume-site.xml
04 <property>
05     <name>flume.master.servers</name>
06     <value>master</value>
07     <description>This is the addressforthe config servers status
08     server (http)
09     </description>
10 </property>
11
12 <property>
13     <name>flume.collector.output.format</name>
14     <value>raw</value>
15     <description>The output formatforthe data written by a Flume
16     collector node.  There are several formats available:
17       syslog - outputs events in a syslog-like format
18       log4j - outputs events in a pattern similar to Hadoop's log4j pattern
19       raw - Event body only.  This is most similar to copying a file but
20         does not preserve any uniqifying metadata like host/timestamp/nanos.
21       avro - Avro Native file format.  Default currently is uncompressed.
22       avrojson -thisoutputs data as json encoded by avro
23       avrodata -thisoutputs data as a avro binary encoded data
24       debug - used onlyfordebugging
25     </description>
26   </property>
27
28  <property>
29     <name>flume.collector.roll.millis</name>
30     <value>300000</value>
31     <description>The time (in milliseconds)
32     between when hdfs files are closed and anewfile is opened
33     (rolled).
34     </description>
35   </property>
36
37 <property>
38     <name>flume.agent.logdir.maxage</name>
39     <value>10000</value>
40     <description> number of milliseconds before a local log file is
41     considered closed and ready to forward.
42     </description>
43   </property>
44
45   <property>
46     <name>flume.agent.logdir.retransmit</name>
47     <value>60000</value>
48     <description>The time (in milliseconds) before a sent event is
49     assumed lost and needs to be retried in end-to-end reliability
50     mode again.  This should be at least 2x the
51     flume.collector.roll.millis.
52     </description>
53   </property>

　　4、将配置好的Flume整个文件夹打包，并发送到agent和collector的机器上：

1 [wyp@master~]$ tar -zcvf flume-0.9.4-cdh3u3.tar.gz flume-0.9.4-cdh3u3
2 [wyp@master~]$ scp flume-0.9.4-cdh3u3.tar.gz agent:/home/wyp
3 [wyp@master~]$ scp flume-0.9.4-cdh3u3.tar.gz collector:/home/wyp

　　5、分别在agent和collector机器上解压上述包，并在master，agent和collector机器上分别启动以下进程：

1 [wyp@master~]$ $FLUME_HOME/bin/flume master
2
3 [wyp@agent~]$ $FLUME_HOME/bin/flume node_nowatch –n agent
4
5 [wyp@collector~]$ $FLUME_HOME/bin/flume node_nowatch –n collector

这样master机器就充当master角色；agent 机器充当agent角色；collector机器充当collector角色。
　　 6、打开http://master:35871，看看能否进去，并看到agent和collector进程成功启动，则说明Flume安装完成！