flume-ng负载均衡load-balance、failover集群搭建

ftlk

　　转自：http://blog.csdn.net/morning_pig/article/details/9093149
　　
集群采用3台机器：
[html] viewplaincopy

• host1 load-balance设置  
  
• host2 机器1  
  
• host3 机器2  
  

其中，host1 机器配置：
[html] viewplaincopy

• #Define a memory channel called c1 on a1  
  
• a1.channels = c1  
  
• a1.sources = r1  
  
• a1.sinks = k1 k2  
  
• a1.sinkgroups = g1  
  
• a1.sinkgroups.g1.sinks = k1 k2  
  
• a1.sinkgroups.g1.processor.type = load_balance  
  
• a1.sinkgroups.g1.processor.selector = round_robin  
  
• a1.sinkgroups.g1.processor.backoff = true  
  
•   
  
• a1.channels.c1.type = file   
  
• a1.channels.c1.checkpointDir = /tmp/flume/loadcheckpoint  
  
• a1.channels.c1.dataDirs = /tmp/flume/loaddata  
  
•   
  
• a1.sources.r1.channels = c1  
  
• a1.sources.r1.type = avro  
  
• a1.sources.r1.bind = 0.0.0.0  
  
• a1.sources.r1.port = 41415  
  
•   
  
• a1.sinks.k1.channel = c1  
  
• a1.sinks.k1.type = avro  
  
• a1.sinks.k1.hostname = host2  
  
• a1.sinks.k1.port = 41414  
  
• a1.sinks.k2.channel = c1  
  
• a1.sinks.k2.type = AVRO  
  
• a1.sinks.k2.hostname = host3  
  
• a1.sinks.k2.port = 41414  
  

host2 机器配置：
[html] viewplaincopy

• a2.channels = c1  
  
• a2.sources = r1  
  
• a2.sinks = k1  
  
•   
  
• a2.channels.c1.type = FILE  
  
• a1.channels.c1.checkpointDir = /tmp/flume/checkpoint  
  
• a1.channels.c1.dataDirs = /tmp/flume/data  
  
•   
  
• a2.sources.r1.channels = c1  
  
• a2.sources.r1.type = AVRO  
  
• a2.sources.r1.bind = 0.0.0.0  
  
• a2.sources.r1.port = 41414  
  
•   
  
• a2.sinks.k1.channel = c1  
  
• a2.sinks.k1.type = file_roll  
  
• a2.sinks.k1.sink.directory = /tmp/load/  
  
• a2.sinks.k1.sink.rollInterval = 0  
  

host3 机器配置：
[html] viewplaincopy

• a2.channels = c1  
  
• a2.sources = r1  
  
• a2.sinks = k1  
  
•   
  
• a2.channels.c1.type = FILE  
  
• a1.channels.c1.checkpointDir = /tmp/flume/checkpoint  
  
• a1.channels.c1.dataDirs = /tmp/flume/data  
  
•   
  
• a2.sources.r1.channels = c1  
  
• a2.sources.r1.type = AVRO  
  
• a2.sources.r1.bind = 0.0.0.0  
  
• a2.sources.r1.port = 41414  
  
•   
  
• a2.sinks.k1.channel = c1  
  
• a2.sinks.k1.type = file_roll  
  
• a2.sinks.k1.sink.directory = /tmp/load/  
  
• a2.sinks.k1.sink.rollInterval = 0  
  

客户端机器暂时使用flume-ng的agent发送，配置如下：
[html] viewplaincopy

• # Define a memory channel called c1 on a1  
  
• a1.channels.c1.type = file   
  
• a1.channels.c1.checkpointDir = /tmp/flume/checkpoint  
  
• a1.channels.c1.dataDirs = /tmp/flume/data  
  
•   
  
•   
  
• a1.sources.r1.channels = c1  
  
• a1.sources.r1.type = exec  
  
• a1.sources.r1.command = cat /tmp/linux.log  
  
•   
  
• a1.sinks.k1.type = avro   
  
• a1.sinks.k1.channel = c1  
  
• a1.sinks.k1.hostname = host1  
  
• a1.sinks.k1.port = 41415  
  
•   
  
• a1.channels = c1  
  
• a1.sources = r1  
  
• a1.sinks = k1  
  

客户端/tmp/linux.log文件3G左右，发送给host1。

启动：
host2：
[html] viewplaincopy

• bin/flume-ng agent -c conf -f conf/load-sink1.conf -n a2  
  

host3：
[html] viewplaincopy

• bin/flume-ng agent -c conf -f conf/load-sink2.conf -n a2  
  

host1：
[html] viewplaincopy

• bin/flume-ng agent -c conf -f conf/load-balance.conf -n a1  
  

客户端：
[html] viewplaincopy

• bin/flume-ng agent -c conf -f conf/client.conf -n a1  
  

注意：启动方式最好是从下往上启动。即：先启动host2和host3，然后启动host1，最后启动client。

测试过程中，可以随时将host2或host3停止，过一段时间再启动。
这样，就测试了flume-ng的load-balance和failover功能。