Flume

Flume安装和使用说明

1.flume简介

Flume是Cloudera提供的日志收集系统，Flume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（可定制）的能力。
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。

2.安装和使用说明:
2.1 安装
a. 下载: http://archive.cloudera.com/cdh/3/ flume-0.9.0+1.tar.gz
接着解压.暂时用$flume代表解压路径.
b. 用户文档：http://archive.cloudera.com/cdh/3/flume/UserGuide.html
c. 下载: http://archive.cloudera.com/cdh/3/ zookeeper-3.3.1.tar.gz
d. 安装zookeeper
yum install hadoop-zookeeper –y
yum install hadoop-zookeeper-server –y
修改/zookeeper-3.3.1/conf/ zoo_sample.cfg重命名为zoo.cfg
执行如下命令：

export  ZOOKEEPER_HOME=/home/hadoop/zookeeper-3.3.1
export  FLUME_HOME=/home/hadoop/flume-0.9.0+1
export  PATH=.:$FLUME_HOME/bin:$ZOOKEEPER_HOME/bin:$PATH

2.2 使用

执行>flume
输出如下：
usage: flume command [args...]
commands include:
  dump            Takes a specified source and dumps to console
  node            Start a Flume node/agent (with watchdog)
  master          Start a Flume Master server (with watchdog)
  version         Dump flume build version information
  node_nowatch    Start a flume node/agent (no watchdog)
  master_nowatch  Start a Flume Master server (no watchdog)
      class <class>   Run specified fully qualified class using Flume environment (no watchdog)
                   ex: flume com.cloudera.flume.agent.FlumeNode
  classpath       Dump the classpath used by the java executables
  shell           Start the flume shell
启动flume的master节点执行:bin/flume master
通过flume打开文件
输入命令
$ flume dump 'tail("/home/hadoop/log/bb.txt")'
输出：


通过flume导入文件到hdfs
可打开http://10.1.27.30:35871/flumemaster.jsp 即可看到整理节点的情况
从上面URL打开的选项卡config,输入节点配置，然后点提交查询内容
如下：

Source为数据源，可有多种输入源,sink为接收器,当启动master节点时，会把文件写入到hdsf里
启动配置好的节点：bin/flume node –n master
通过flume读取syslog-ng
 
分别启动节点host和collector节点






3.附录:
Flume Event
Sources console
Stdin console
text("filename")
One shot text file source. One line is one event
tail("filename")
Similar to Unix’s tail -F. One line is one event. Stays open for more data and follows filename if file rotated.
multitail("file1"[, "file2"[, …]])
Similar to tail source but follows multiple files.
asciisynth(msg_count,msg_size)
A source that synthetically generates msg_count random messages of size msg_size. This converts all characters into printable ASCII characters.
syslogUdp(port)
Syslog over UDP port, port. This is syslog compatible.
syslogTcp(port)
Syslog over TCP port, port. This is syslog-ng compatible.
Flume Event Sinks
null Null sink. Events are dropped.
console[("format")] Console sink. Display to console’s stdout. The "format" argument is optional and defaults to the "debug" output format.
text("txtfile"[,"format"]) Textfile sink. Write the events to text file txtfile using output format "format". The default format is "raw" event bodies with no metadata.
dfs("dfsfile") DFS seqfile sink. Write serialized Flume events to a dfs path such as hdfs://namenode/file or file:///file in Hadoop’s seqfile format. Note that because of the hdfs write semantics, no data for this sink write until the sink is closed.
syslogTcp("host",port) Syslog TCP sink. Forward to events to host on TCP port port in syslog wire format (syslog-ng compatible), or to other Flume nodes setup to listen for syslogTcp.

默认端口如下：
TCP ports are used in all situations.
node collector port flume.collector.port 35853+
node status web server flume.node.http.port 35862+
master status web server flume.master.http.port 35871
master heartbeat port flume.master.heartbeat.port 35872
master admin/shell port flume.master.admin.port 35873
master gossip port flume.master.gossip.port 35890
master → zk port flume.master.zk.client.port 3181
zk → zk quorum port flume.master.zk.server.quorum.port 3182
zk → zk election port
flume.master.zk.server.election.port 3183