功能
采集數據,解析數據,上報數據至transfer
-
基本涵蓋了系統層面監控指標,直接將數據轉換為metricValue形式,上報至transfer
-
支持插件采集,代碼插件放可受git管理,放置在plugin目錄,從HBS中獲取執行周期,版本等
-
支持數據直接上報,提供HTTPAPI,將收到的數據轉換為metricValue形式,上報至transfer
-
支持HTTP API對自身控制
配置文件:
-
設置采集數據的腳本更新地址
-
指定Agent的上游模塊的地址,包括heartbeat和transfer模塊
-
agent開放的HTTP端口
-
采集數據的過濾條件
{
"debug": true, # 控制一些debug信息的輸出,生產環境通常設置為false
"hostname": "", # agent采集了數據發給transfer,endpoint就設置為了hostname,默認通過`hostname`獲取,如果配置中配置了hostname,就用配置中的
"ip": "", # agent與hbs心跳的時候會把自己的ip地址發給hbs,agent會自動探測本機ip,如果不想讓agent自動探測,可以手工修改該配置
"plugin": {
"enabled": false, # 默認不開啟插件機制
"dir": "./plugin", # 把放置插件腳本的git repo clone到這個目錄
"git": "https://github.com/open-falcon/plugin.git", # 放置插件腳本的git repo地址
"logs": "./logs" # 插件執行的log,如果插件執行有問題,可以去這個目錄看log
},
"heartbeat": {
"enabled": true, # 此處enabled要設置為true
"addr": "127.0.0.1:6030", # hbs的地址,端口是hbs的rpc端口
"interval": 60, # 心跳周期,單位是秒
"timeout": 1000 # 連接hbs的超時時間,單位是毫秒
},
"transfer": {
"enabled": true, # 此處enabled要設置為true
"addrs": [
"127.0.0.1:8433",
"127.0.0.1:8433"
], # transfer的地址,端口是transfer的rpc端口, 可以支持寫多個transfer的地址,agent會保證HA
"interval": 60, # 采集周期,單位是秒,即agent一分鍾采集一次數據發給transfer
"timeout": 1000 # 連接transfer的超時時間,單位是毫秒
},
"http": {
"enabled": true, # 是否要監聽http端口
"listen": ":1988" # 如果監聽的話,監聽的地址
},
"collector": {
"ifacePrefix": ["eth", "em"] # 默認配置只會采集網卡名稱前綴是eth、em的網卡流量,配置為空就會采集所有的,lo的也會采集。可以從/proc/net/dev看到各個網卡的流量信息
},
"ignore": { # 默認采集了200多個metric,可以通過ignore設置為不采集
"cpu.busy": true,
"mem.swapfree": true
}
}
agent 和HBS 通信
每隔一分鍾,Agent會與HBS進行一系列RPC通信,上報自己的狀態,並且獲取一些自己需要的數據。
-
向HBS匯報的數據:
-
HostName
-
IP (cfg中的IP,或者本機的IP)
-
Agent版本
-
Plugin版本(腳本所在git目錄的版本,git rev-parse HEAD)
-
-
從HBS獲取的數據:
-
Plugin信息,需要以什么頻率運行什么腳本
-
內置采集信息
-
URL_CHECK_HEALTH,需要定期檢查哪些URL
-
NET_PORT_LISTEN,需要檢查哪些端口的存活
-
DU_BS,需要檢查哪些文件夾的占用空間
-
PROC_NUM,需要檢查哪些進程的存活
-
-
信任的IP列表(執行遠程命令時需要檢查)
-
agent HTTP API
提供如下功能:
-
Admin命令
-
Agent進程退出
-
重新加載配置
-
返回工作目錄
-
返回信任的IP列表)
-
返回目前采集的數據CPU、磁盤、IO、內存……
-
-
Plugin(使用git同步遠端的腳本)
-
接收其它模塊Push數據
-
運行任意用戶命令
運行系統命令:date, uptime, loadavg
http://127.0.0.1:1988/exit
http://127.0.0.1:1988/config/reload
http://127.0.0.1:1988/workdir
http://127.0.0.1:1988/ips
http://127.0.0.1:1988/proc/cpu/num
http://127.0.0.1:1988/proc/cpu/mhz
http://127.0.0.1:1988/page/cpu/usage
http://127.0.0.1:1988/proc/cpu/usage
http://127.0.0.1:1988/page/df
http://127.0.0.1:1988/health
http://127.0.0.1:1988/version
http://127.0.0.1:1988/page/diskio
http://127.0.0.1:1988/proc/kernel/hostname
http://127.0.0.1:1988/proc/kernel/maxproc
http://127.0.0.1:1988/proc/kernel/maxfiles
http://127.0.0.1:1988/proc/kernel/version
http://127.0.0.1:1988/page/memory
http://127.0.0.1:1988/proc/memory
http://127.0.0.1:1988/
http://127.0.0.1:1988/plugin/update
http://127.0.0.1:1988/plugin/reset
http://127.0.0.1:1988/plugins
http://127.0.0.1:1988/v1/push
http://127.0.0.1:1988/system/date
http://127.0.0.1:1988/page/system/uptime
http://127.0.0.1:1988/proc/system/uptime
http://127.0.0.1:1988/page/system/loadavg
http://127.0.0.1:1988/proc/system/loadavg
上報的數據格式:
通過在transfer 的8433端口抓包拿到了agent上報數據的格式
{"method":"Transfer.Update","params":[[{"endpoint":"open-falcon-1","metric":"rate","value":0.016666596005299582,"step":60,"counterType":"GAUGE","tags":"name=pfc.push.cnt","timestamp":1491879
120},{"endpoint":"open-falcon-1","metric":"sum","value":76,"step":60,"counterType":"GAUGE","tags":"name=pfc.push.cnt","timestamp":1491879120},{"endpoint":"open-falcon-1","metric":"value","value":4,"st
ep":60,"counterType":"GAUGE","tags":"name=pfc.push.ms","timestamp":1491879120},{"endpoint":"open-falcon-1","metric":"value","value":4,"step":60,"counterType":"GAUGE","tags":"name=pfc.push.size","times
tamp":1491879120}]],"id":23}
其中,Step和CounterType中RRD中的概念。Step就是數據采集的間隔,單位為秒。而CounterType分為五個類型,分別如下表:
-
COUNTER: 必須是遞增的,除非是計數器溢出(overflows)。在這種情況下,RRDtool 會自動修改收到的值。例如網絡接口流量、收到的packets 數量都屬於這一類型。
-
DERIVE: 和 COUNTER 類似。但可以是遞增,也可以遞減,或者一會增加一會兒減少。
-
ABSOLUTE: ABSOLUTE 比較特殊,它每次都假定前一個interval的值是0,再計算平均值。
-
GAUGE: GAGUE 和上面三種不同,它沒有“平均”的概念,RRDtool 收到值之后字節存入 RRA 中。
-
COMPUTE: COMPUTE 比較特殊,它並不接受輸入,它的定義是一個表達式,能夠引用其他DS並自動計算出某個值。
agent 和transfer底層鏈接
-
agent和transfer 之間數據交互采用長連接方式
-
agent剛啟動之后,不會立即和transfer建立鏈接,而是在第一次上報數據的時候經過tcp 三次握手之后建立,此后保持鏈接不斷
-
agent 上報完數據之后,transfer 會返回給agent 上報的結果
2017/04/11 11:11:00 var.go:56: => <Total=4> <Endpoint:open-falcon-1, Metric:rate, Type:GAUGE, Tags:name=pfc.push.cnt, Step:60, Time:1491880260, Value:0.016666685985855727>
2017/04/11 11:11:00 var.go:63: <= <Total=4, Invalid:0, Latency=0ms, Message:ok>
2017/04/11 11:11:27 var.go:56: => <Total=7> <Endpoint:open-falcon-1, Metric:df.bytes.free.percent, Type:GAUGE, Tags:mount=/,fstype=ext4, Step:60, Time:1491880287, Value:95.25614893427964>
2017/04/11 11:11:27 var.go:56: => <Total=6> <Endpoint:open-falcon-1, Metric:ss.estab, Type:GAUGE, Tags:, Step:60, Time:1491880287, Value:57>
2017/04/11 11:11:27 var.go:63: <= <Total=7, Invalid:0, Latency=0ms, Message:ok>
2017/04/11 11:11:27 var.go:63: <= <Total=6, Invalid:0, Latency=0ms, Message:ok>
2017/04/11 11:11:27 var.go:56: => <Total=88> <Endpoint:open-falcon-1, Metric:agent.alive, Type:GAUGE, Tags:, Step:60, Time:1491880287, Value:1>
2017/04/11 11:11:27 var.go:63: <= <Total=88, Invalid:0, Latency=0ms, Message:ok>
##agent 和transfer建立tcp 鏈接
11:11:00.003435 IP localhost.53889 > localhost.8433: Flags [S], seq 860978551, win 32792, options [mss 16396,sackOK,TS val 5479392 ecr 0,nop,wscale 7], length 0
11:11:00.003457 IP localhost.8433 > localhost.53889: Flags [S.], seq 3616557726, ack 860978552, win 32768, options [mss 16396,sackOK,TS val 5479392 ecr 5479392,nop,wscale 7], length 0
11:11:00.003474 IP localhost.53889 > localhost.8433: Flags [.], ack 1, win 257, options [nop,nop,TS val 5479392 ecr 5479392], length 0
#agent 第一次上報數據
11:11:00.004004 IP localhost.53889 > localhost.8433: Flags [P.], seq 1:619, ack 1, win 257, options [nop,nop,TS val 5479393 ecr 5479392], length 618
11:11:00.004180 IP localhost.8433 > localhost.53889: Flags [.], ack 619, win 266, options [nop,nop,TS val 5479393 ecr 5479393], length 0
11:11:00.004976 IP localhost.8433 > localhost.53889: Flags [P.], seq 1:83, ack 619, win 266, options [nop,nop,TS val 5479394 ecr 5479393], length 82
11:11:00.005027 IP localhost.53889 > localhost.8433: Flags [.], ack 83, win 257, options [nop,nop,TS val 5479394 ecr 5479394], length 0
#agent 第二次上報數據,此次上報的數據tcp 交互不再是連續的,和日志相對應
11:11:27.387968 IP localhost.53889 > localhost.8433: Flags [P.], seq 619:1816, ack 83, win 257, options [nop,nop,TS val 5506777 ecr 5479394], length 1197
11:11:27.388750 IP localhost.8433 > localhost.53889: Flags [P.], seq 83:165, ack 1816, win 285, options [nop,nop,TS val 5506778 ecr 5506777], length 82
11:11:27.388758 IP localhost.53889 > localhost.8433: Flags [.], ack 165, win 257, options [nop,nop,TS val 5506778 ecr 5506778], length 0
11:11:27.397886 IP localhost.53889 > localhost.8433: Flags [P.], seq 1816:2631, ack 165, win 257, options [nop,nop,TS val 5506787 ecr 5506778], length 815
11:11:27.398225 IP localhost.8433 > localhost.53889: Flags [P.], seq 165:247, ack 2631, win 304, options [nop,nop,TS val 5506787 ecr 5506787], length 82
11:11:27.437698 IP localhost.53889 > localhost.8433: Flags [.], ack 247, win 257, options [nop,nop,TS val 5506827 ecr 5506787], length 0
11:11:27.472026 IP localhost.53889 > localhost.8433: Flags [P.], seq 2631:15063, ack 247, win 257, options [nop,nop,TS val 5506861 ecr 5506787], length 12432
11:11:27.474822 IP localhost.8433 > localhost.53889: Flags [P.], seq 247:330, ack 15063, win 384, options [nop,nop,TS val 5506863 ecr 5506861], length 83
11:11:27.474837 IP localhost.53889 > localhost.8433: Flags [.], ack 330, win 257, options [nop,nop,TS val 5506863 ecr 5506863], length 0