监控开发之用munin来自定义插件监控redis和mongodb

求监控组的大哥大妹子们干点事，真不容易 ！ 要问他们是谁？  他们是神 。轻易别找他们，因为找了也是白找。  


上次因为python和redis长时间brpop的时候，会有线程休眠挂起的情况，所有通知报警平台被下线了。这次算是完美解决了。再把他给上线。这两公司的告警已经开始往我这边的接口开始仍了。


这边正在改zabbix cmdb的控制，所以暂时不能登录。等搞好了后，让他们搞下redis和mogodb的监控，居然还让我发邮件和提供脚本及思路啥的。。。   一寻思，又要去zabbix，又要写脚本，还不如把监控都集合在自己的平台上的了。


这次没用选用钟爱的ganglia，麻烦。 也没用另一个graphite，而是用的是munin 。   一个直接yum后就可以访问的性能监控页面。



官方的redis监控和mongodb看起来很麻烦的样子，算了。直接看他们是怎么写的。源码是perl写的，插件好多是shell写的。



写法是相当的简单，只需要指明下图片的显示Y X 轴 ,然后echo就可以了！  


下面是redis 的token使用热点数据，队列的数据，及mongodb count的数据。



脚本的位置：   /etc/munin/plugins


监控mongodb的脚本：
    
#xiaorui.cc
if [ "$1" = "autoconf" ]; then
        echo yes
        exit 0
fi
 
 
if [ "$1" = "config" ]; then
 
        echo 'graph_title mongodb count mail'
        echo 'graph_args --base 1000 -l 0'
        echo 'graph_vlabel mail queue'
        echo 'graph_scale no'
        echo 'graph_category system'
        echo 'load.label load'
        echo 'graph_info The load average of the machine describes how many processes are in the run-queue (scheduled to run "immediately").'
        echo 'load.info 5 minute load average'
        exit 0
fi
echo -n "load.value "
mongo reportops --eval "db.reportops_log_mail.count()"|tail -n1

监控redis队列的脚本：   
#xiaorui.cc
if [ "$1" = "config" ]; then
 
        # The host name this plugin is for. (Can be overridden to have
        # one machine answer for several)
 
        # The title of the graph
        echo 'graph_title redis mail'
        # Arguments to "rrdtool graph". In this case, tell it that the
        # lower limit of the graph is '0', and that 1k=1000 (not 1024)
        echo 'graph_args --base 1000 -l 0'
        # The Y-axis label
        echo 'graph_vlabel load'
        # We want Cur/Min/Avg/Max unscaled (i.e. 0.42 load instead of
        # 420 milliload)
        echo 'graph_scale no'
        # Graph category. Defaults to 'other'
        echo 'graph_category system'
        # The fields. "label" is used in the legend. "label" is the only
        # required subfield.
        echo 'load.label load'
        # These two read the environment for warning values for the field
        # "load".  If "load_warning" or "warning" aren't set in the
        # environment, no warning levels are set.  Likewise for "load_critical"
        # and "critical".
        print_warning load
        print_critical load
       
        # This one is purely to add an explanation to the web page. The first
        # one is for the graph itself, while the second one is for the field
        # "load".
        echo 'graph_info The load average of the machine describes how many processes are in the run-queue (scheduled to run "immediately").'
        echo 'load.info 5 minute load average'
 
        # Last, if run with the "config"-parameter, quit here (don't
        # display any data)
        exit 0
fi
 
# If not run with any parameters at all (or only unknown ones), do the
# real work - i.e. display the data. Almost always this will be
# "value" subfield for every data field.
 
echo -n "load.value "
 
redis-cli LLEN sendmaillist|cut -d '' -f2


写完了后，/etc/init.d/munin-node restart 就可以了。等一会刷新下页面就出来了。


关键就是最后那两行。。
   
echo -n "load.value "
 
redis-cli LLEN sendmaillist|cut -d '' -f2


网上有人做了python的munin操作模块，有兴趣的朋友可以试试。


 
#!/usr/bin/env python
 
import os
from munin import MuninPlugin
 
class LoadAVGPlugin(MuninPlugin):
    title = "Load average"
    args = "--base 1000 -l 0"
    vlabel = "load"
    scale = False
    category = "system"
 
    @property
    def fields(self):
        warning = os.environ.get('load_warn', 10)
        critical = os.environ.get('load_crit', 120)
        return [("load", dict(
                label = "load",
                info = 'The load average of the machine describes how many processes are in the run-queue (scheduled to run "immediately").',
                type = "GAUGE",
                min = "0",
                warning = str(warning),
                critical = str(critical)))]
 
    def execute(self):
        if os.path.exists("/proc/loadavg"):
            loadavg = open("/proc/loadavg", "r").read().strip().split(' ')
        else:
            from subprocess import Popen, PIPE
            output = Popen(["uptime"], stdout=PIPE).communicate()[0]
            loadavg = output.rsplit(':', 1)[1].strip().split(' ')[:3]
        return dict(load=loadavg[1])
 
if __name__ == "__main__":
    LoadAVGPlugin().run()


总结下，munin真的够简单的了，他的简单也意味着，他也就 适合我这样的运维研发人员临时做些统计的场景。  记得以前使用munin，当时做zeromq的统计，超过几十台是没啥问题，当然这话是（feihua），要是几十台都有问题，那这监控的水准确实够烂。  这东西的局限确实够大。也就临时画画图还行。