flume 是 cloudera公司研发的日志收集系统,采用3层结构:1. agent层,用于直接收集日志;2.connect 层,用于接受日志; 3. 数据存储层,用于保存日志。由一到多个master管理1和2层节点。
本文采用单节点试用flume日志处理。
flume 单节点使用示例
flume号称水道,其设计思路是水源(source)和水槽(sink)都是可以人意组合的。这样在汇集水的过程是非常灵活的。每一个逻辑节点由事件生产(source)和事件消费(sink)组成。从源拉取数据,从sink推走。单节点缺省sink是console。单节点源采用dump的方式。命令行 flume dump source
常用source源:
- console
- 标准输入控制台
- text(“filename”)
- 单文本文件源,一行一事件
- tail(“filename”)
- 和 Unix 的tail -F 类似。一行一事件。一直打开等待数据,会跟踪文件切换。
- multitail(“file1″[, "file2"[, …]])
- 同 tail 源类似,但可以跟踪多文件。
- asciisynth(msg_count,msg_size)
- 一个源,用于产生msg_count 个msg_size大小的随机消息,转成可打印 ASCII字符。
- syslogUdp(port)
- UDP 端口上的 Syslog,和syslog兼容。
- syslogTcp(port)
- TCP 端口上的 Syslog,和syslog-ng兼容。
[zhouhh@Hadoop48 ~]$ flume dump console 你好 Hadoop48 [INFO Mon Jan 21 11:37:52 CST 2013] ?? haha Hadoop48 [INFO Mon Jan 21 11:37:56 CST 2013] haha hello Hadoop48 [INFO Mon Jan 21 11:37:59 CST 2013] hello this is a test Hadoop48 [INFO Mon Jan 21 11:38:07 CST 2013] this is a test
[zhouhh@Hadoop48 ~]$ flume dump 'text("/etc/services")' ... Hadoop48 [INFO Mon Jan 21 10:57:15 CST 2013] iqobject\t48619/udp\t\t\t# iqobject Hadoop48 [INFO Mon Jan 21 10:57:15 CST 2013] # Local services 2013-01-21 10:57:15,498 [logicalNode dump-9] INFO debug.TextFileSource: File /etc/services closed 2013-01-21 10:57:15,498 [logicalNode dump-9] INFO debug.ConsoleEventSink: ConsoleEventSink( debug ) closed
[zhouhh@Hadoop48 ~]$ flume dump 'tail("testfile")'
[zhouhh@Hadoop48 ~]$ echo hello world >> testfile [zhouhh@Hadoop48 ~]$ echo 你好 >> testfile
输出
[zhouhh@Hadoop48 ~]$ flume dump 'tail("testfile")' ... Hadoop48 [INFO Mon Jan 21 10:55:39 CST 2013] { tailSrcFile : (long)8387236824819002469 (string) 'testfile' (double)4.914663849160389E252 } hello world Hadoop48 [INFO Mon Jan 21 10:56:07 CST 2013] { tailSrcFile : (long)8387236824819002469 (string) 'testfile' (double)4.914663849160389E252 } \u4F60\u597D
[zhouhh@Hadoop48 ~]$ flume dump 'multitail("test1", "test2")'
[zhouhh@Hadoop48 ~]$ echo Hello world test1! >> test1 [zhouhh@Hadoop48 ~]$ echo Hello world test2! >> test2
输出
[zhouhh@Hadoop48 ~]$ flume dump 'multitail("test1", "test2")'
... Hadoop48 [INFO Mon Jan 21 10:59:51 CST 2013] { tailSrcFile : test1 } Hello world test1! Hadoop48 [INFO Mon Jan 21 11:00:00 CST 2013] { tailSrcFile : test2 } Hello world test2!
tail 缺省分隔符是\n,并且\n不包含在事件中。tail 支持自定义分隔符,还可指定分隔符是包含在前一事件(prev),后一事件(next),还是不包含在事件中(exclude).
两个及以上\n换行 tail("file", delim="\n\n+", delimMode="exclude") 用</a>做分隔符,并且做为前一事件结尾。 tail("file", delim="</a>", delimMode="prev") 用"\n\d\d\d\d"正则表达式做分隔符,并且做为后一事件的开头。用于分析stack dump日志。如年开头四位数字。 tail("file", delim="\\n\\d\\d\\d\\d", delimMode="next")
获取2行3个随机字符日志
[zhouhh@Hadoop48 ~]$ flume dump 'asciisynth(2,3)' Hadoop48 [INFO Mon Jan 21 11:12:36 CST 2013] :,J Hadoop48 [INFO Mon Jan 21 11:12:36 CST 2013] o
用udp或tcp接受syslog格式日志
[zhouhh@Hadoop48 ~]$ flume dump 'syslogUdp(5140)'
发送消息到5140端口
[zhouhh@Hadoop48 ~]$ echo "hello " | nc -u hadoop48 5140 [zhouhh@Hadoop48 ~]$ flume dump 'syslogUdp(5140)' 2013-01-21 11:15:10,587 [logicalNode dump-9] WARN syslog.SyslogUdpSource: 1 rejected packets. packet: java.net.DatagramPacket@75037619 com.cloudera.flume.handlers.text.EventExtractException: Failed to extract syslog wire entry at com.cloudera.flume.handlers.syslog.SyslogWireExtractor.extract(SyslogWireExtractor.java:178) at com.cloudera.flume.handlers.syslog.SyslogWireExtractor.extractEvent(SyslogWireExtractor.java:89) at com.cloudera.flume.handlers.syslog.SyslogUdpSource.next(SyslogUdpSource.java:88) at com.cloudera.flume.core.connector.DirectDriver$PumperThread.run(DirectDriver.java:105) 这是因为发送的消息没有wireformat数据 [zhouhh@Hadoop48 ~]$ echo "<37>hello " | nc -u hadoop48 5140 [zhouhh@Hadoop48 ~]$ flume dump 'syslogUdp(5140)' ... Hadoop48 [INFO Mon Jan 21 11:24:02 CST 2013] { syslogfacility : 4 } { syslogseverity : 5 } hello
<37>用于指定log分类和log级别
接受tcp数据
[zhouhh@Hadoop48 ~]$ flume dump 'syslogTcp(5140)'
[zhouhh@Hadoop48 ~]$ echo "<37>hello via syslog" | nc -t localhost 5140
显示结果 [zhouhh@Hadoop48 ~]$ flume dump 'syslogTcp(5140)' ... Hadoop48 [INFO Mon Jan 21 11:33:01 CST 2013] { syslogfacility : 4 } { syslogseverity : 5 } hello via syslog
参考
http://archive.cloudera.com/cdh/3/flume/UserGuide/index.html