Ganglia的安装配置

1. 在主节点上安装ganglia-webfrontend和ganglia-monitor

1. sudo apt-get install ganglia-webfrontend ganglia-monitor

复制代码

在主节点上安装ganglia-webfrontend和ganglia-monitor。在其他监视节点上，只需要安装ganglia-monitor即可
将ganglia的文件链接到apache的默认目录下

1. sudo ln -s /usr/share/ganglia-webfrontend /var/www/ganglia

复制代码

2. 安装ganglia-monitor
在其他监视节点上，只需要安装ganglia-monitor

1. sudo apt-get install ganglia-monitor

复制代码

3. Ganglia配置
gmond.conf
在每个节点上都需要配置/etc/ganglia/gmond.conf,配置相同如下所示

1. sudo vim /etc/ganglia/gmond.conf

复制代码

修改后的/etc/ganglia/gmond.conf

1. globals {                    
   
2.   daemonize = yes  ##以后台的方式运行            
   
3.   setuid = yes             
   
4.   user = ganglia     #运行Ganglia的用户              
   
5.   debug_level = 0               
   
6.   max_udp_msg_len = 1472        
   
7.   mute = no             
   
8.   deaf = no             
   
9.   host_dmax = 0 /*secs */ 
   
10.   cleanup_threshold = 300 /*secs */ 
    
11.   gexec = no             
    
12.   send_metadata_interval = 10     #发送数据的时间间隔
    
13. } 
    
14. 
    
15. /* If a cluster attribute is specified, then all gmond hosts are wrapped inside 
    
16. * of a <CLUSTER> tag.  If you do not specify a cluster tag, then all <HOSTS> will 
    
17. * NOT be wrapped inside of a <CLUSTER> tag. */ 
    
18. cluster { 
    
19.   name = "hadoop-cluster"         #集群名称
    
20.   owner = "ganglia"               #运行Ganglia的用户
    
21.   latlong = "unspecified" 
    
22.   url = "unspecified" 
    
23. } 
    
24. 
    
25. /* The host section describes attributes of the host, like the location */ 
    
26. host { 
    
27.   location = "unspecified" 
    
28. } 
    
29. 
    
30. /* Feel free to specify as many udp_send_channels as you like.  Gmond 
    
31.    used to only support having a single channel */ 
    
32. udp_send_channel { 
    
33.   #mcast_join = 239.2.11.71     #注释掉组播
    
34.   host = master                 #发送给安装gmetad的机器
    
35.   port = 8649                   #监听端口
    
36.   ttl = 1 
    
37. } 
    
38. 
    
39. /* You can specify as many udp_recv_channels as you like as well. */ 
    
40. udp_recv_channel { 
    
41.   #mcast_join = 239.2.11.71     #注释掉组播
    
42.   port = 8649 
    
43.   #bind = 239.2.11.71 
    
44. } 
    
45. 
    
46. /* You can specify as many tcp_accept_channels as you like to share 
    
47.    an xml description of the state of the cluster */ 
    
48. tcp_accept_channel { 
    
49.   port = 8649 
    
50. }

复制代码

gmetad.conf
在主节点上还需要配置/etc/ganglia/gmetad.conf,这里面的名字hadoop-cluster和上面gmond.conf中name应该一致。　
/etc/ganglia/gmetad.conf

1. sudo vim /etc/ganglia/gmetad.conf

复制代码

修改为以下内容

1. data_source "hadoop-cluster" 10 master:8649 slave:8649
   
2. setuid_username "nobody"
   
3. rrd_rootdir "/var/lib/ganglia/rrds"
   
4. gridname "hadoop-cluster"
   
5. 注:master:8649 slave:8649为要监听的主机和端口,data_source中hadoop-cluster与gmond.conf中name一致

复制代码

4. Hadoop配置
在所有hadoop所在的节点，均需要配置hadoop-metrics2.properties，配置如下：

1. #   Licensed to the Apache Software Foundation (ASF) under one or more
   
2. #   contributor license agreements.  See the NOTICE file distributed with
   
3. #   this work for additional information regarding copyright ownership.
   
4. #   The ASF licenses this file to You under the Apache License, Version 2.0
   
5. #   (the "License"); you may not use this file except in compliance with
   
6. #   the License.  You may obtain a copy of the License at
   
7. #
   
8. #       http://www.apache.org/licenses/LICENSE-2.0
   
9. #
   
10. #   Unless required by applicable law or agreed to in writing, software
    
11. #   distributed under the License is distributed on an "AS IS" BASIS,
    
12. #   WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
    
13. #   See the License for the specific language governing permissions and
    
14. #   limitations under the License.
    
15. #
    
16. 
    
17. # syntax: [prefix].[source|sink].[instance].[options]
    
18. # See javadoc of package-info.java for org.apache.hadoop.metrics2 for details
    
19. 
    
20. #注释掉以前原有配置
    
21. 
    
22. #*.sink.file.class=org.apache.hadoop.metrics2.sink.FileSink
    
23. # default sampling period, in seconds
    
24. #*.period=10
    
25. 
    
26. # The namenode-metrics.out will contain metrics from all context
    
27. #namenode.sink.file.filename=namenode-metrics.out
    
28. # Specifying a special sampling period for namenode:
    
29. #namenode.sink.*.period=8
    
30. 
    
31. #datanode.sink.file.filename=datanode-metrics.out
    
32. 
    
33. # the following example split metrics of different
    
34. # context to different sinks (in this case files)
    
35. #jobtracker.sink.file_jvm.context=jvm
    
36. #jobtracker.sink.file_jvm.filename=jobtracker-jvm-metrics.out
    
37. #jobtracker.sink.file_mapred.context=mapred
    
38. #jobtracker.sink.file_mapred.filename=jobtracker-mapred-metrics.out
    
39. 
    
40. #tasktracker.sink.file.filename=tasktracker-metrics.out
    
41. 
    
42. #maptask.sink.file.filename=maptask-metrics.out
    
43. 
    
44. #reducetask.sink.file.filename=reducetask-metrics.out
    
45. 
    
46. *.sink.ganglia.class=org.apache.hadoop.metrics2.sink.ganglia.GangliaSink31  
    
47. *.sink.ganglia.period=10
    
48. 
    
49. *.sink.ganglia.slope=jvm.metrics.gcCount=zero,jvm.metrics.memHeapUsedM=both  
    
50. *.sink.ganglia.dmax=jvm.metrics.threadsBlocked=70,jvm.metrics.memHeapUsedM=40  
    
51. 
    
52. namenode.sink.ganglia.servers=master:8649  
    
53. resourcemanager.sink.ganglia.servers=master:8649  
    
54. 
    
55. datanode.sink.ganglia.servers=master:8649    
    
56. nodemanager.sink.ganglia.servers=master:8649    
    
57. 
    
58. 
    
59. maptask.sink.ganglia.servers=master:8649    
    
60. reducetask.sink.ganglia.servers=master:8649

复制代码

5. Hbase配置
在所有的hbase节点中均配置hadoop-metrics2-hbase.properties,配置如下:

1. # syntax: [prefix].[source|sink].[instance].[options]
   
2. # See javadoc of package-info.java for org.apache.hadoop.metrics2 for details
   
3. 
   
4. #*.sink.file*.class=org.apache.hadoop.metrics2.sink.FileSink
   
5. # default sampling period
   
6. #*.period=10
   
7. 
   
8. # Below are some examples of sinks that could be used
   
9. # to monitor different hbase daemons.
   
10. 
    
11. # hbase.sink.file-all.class=org.apache.hadoop.metrics2.sink.FileSink
    
12. # hbase.sink.file-all.filename=all.metrics
    
13. 
    
14. # hbase.sink.file0.class=org.apache.hadoop.metrics2.sink.FileSink
    
15. # hbase.sink.file0.context=hmaster
    
16. # hbase.sink.file0.filename=master.metrics
    
17. 
    
18. # hbase.sink.file1.class=org.apache.hadoop.metrics2.sink.FileSink
    
19. # hbase.sink.file1.context=thrift-one
    
20. # hbase.sink.file1.filename=thrift-one.metrics
    
21. 
    
22. # hbase.sink.file2.class=org.apache.hadoop.metrics2.sink.FileSink
    
23. # hbase.sink.file2.context=thrift-two
    
24. # hbase.sink.file2.filename=thrift-one.metrics
    
25. 
    
26. # hbase.sink.file3.class=org.apache.hadoop.metrics2.sink.FileSink
    
27. # hbase.sink.file3.context=rest
    
28. # hbase.sink.file3.filename=rest.metrics
    
29. 
    
30. 
    
31. *.sink.ganglia.class=org.apache.hadoop.metrics2.sink.ganglia.GangliaSink31  
    
32. *.sink.ganglia.period=10  
    
33. 
    
34. hbase.sink.ganglia.period=10  
    
35. hbase.sink.ganglia.servers=master:8649

复制代码

6. 启动hadoop、hbase集群

1. start-dfs.sh
   
2. start-yarn.sh
   
3. start-hbase.sh

复制代码

7. 启动Ganglia
先需要重启hadoop和hbase 。在各个节点上启动gmond服务，主节点还需要启动gmetad服务。
使用apt-get方式安装的Ganglia，可以直接用service方式启动。

1. sudo service ganglia-monitor start（每台机器都需要启动）
   
2. 
   
3. sudo service gmetad start（在安装了ganglia-webfrontend的机器上启动）

复制代码

8. 检验

登录浏览器查看：http://master/ganglia，可能会出现网络页面找不到的问题，只需要将ganglia的子域添加到监控的机子上即可。

如果Hosts up为9即表示安装成功。

若安装不成功，有几个很有用的调试命令：
以调试模式启动gmetad：gmetad -d 9
查看gmetad收集到的XML文件：telnet master 8649


9. 截图