Kubernetes的生态中,cAdvisor是作为容器监控数据采集的Agent,其部署在每个节点上,内部代码结构大致如下:代码结构很良好,collector和storage部分基本可做到增量扩展开发。
关于cAdvisor支持自定义指标方式能力,其自身是通过容器部署的时候设置lable标签项:io.cadvisor.metric.开头的lable,而value则为自定义指标的配置文件,形如下:
{
"endpoint" : {
"protocol": "https",
"port": 8000,
"path": "/nginx_status"
},
"metrics_config" : [
{ "name" : "activeConnections",
"metric_type" : "gauge",
"units" : "number of active connections",
"data_type" : "int",
"polling_frequency" : 10,
"regex" : "Active connections: ([0-9]+)"
},
{ "name" : "reading",
"metric_type" : "gauge",
"units" : "number of reading connections",
"data_type" : "int",
"polling_frequency" : 10,
"regex" : "Reading: ([0-9]+) .*"
},
{ "name" : "writing",
"metric_type" : "gauge",
"data_type" : "int",
"units" : "number of writing connections",
"polling_frequency" : 10,
"regex" : ".*Writing: ([0-9]+).*"
},
{ "name" : "waiting",
"metric_type" : "gauge",
"units" : "number of waiting connections",
"data_type" : "int",
"polling_frequency" : 10,
"regex" : ".*Waiting: ([0-9]+)"
}
] }
当前cAdvisor只支持http接口方式,也就是被监控容器应用必须提供http接口,所以能力较弱,如果我们在collector这一层做扩展增强,提供数据库,mq等等标准应用的监控模式是很有价值的。在此之前的另一种方案就是如上图所示搭配promethuese(其内置有非常丰富的标准应用的插件涵盖了APM所需的采集大部分插件),但是这往往会导致系统更复杂(如果应用层并非想使用promethuse)
在Kubernetes监控生态中,一般是如下的搭配使用: