用python编写exporter监控接口访问数量+错误率+响应时间

1.编写exporter

import prometheus_client
from prometheus_client import Gauge,start_http_server,Counter
import pycurl
import time
import threading
from io import BytesIO

#创建client_python里提供的prometheus Counter数据类型
url_http_code = Counter("url_http_code", "request http_code of the host",['code','url'])
url_http_request_time = Counter("url_http_request_time", "request http_request_time of the host",['le','url'])
http_request_total = Counter("http_request_total", "request request total of the host",['url'])

#curl url，返回状态码和总共耗时 -- 返回状态和响应时间
def test_website(url):
    buffer_curl = BytesIO()
    c = pycurl.Curl()
    c.setopt(pycurl.URL, url)
#    c.setopt(pycurl.WRITEDATA, buffer_curl)
    c.setopt(pycurl.CONNECTTIMEOUT, 3)
    c.setopt(pycurl.TIMEOUT, 3)
    try:
        c.perform()
    except pycurl.error:
        http_code = 500
        http_total_time = 999
    else:
        http_code = c.getinfo(pycurl.HTTP_CODE)
        http_total_time = c.getinfo(pycurl.TOTAL_TIME)
    return http_code, http_total_time
    
#根据curl返回值，统计放到exporter显示的数据 -- 统计各个状态的总数
def count_metric(url):
    http_code, http_total_time = test_website(url)
    if http_code >= 100 and http_code < 200 :
        url_http_code.labels('1xx',url).inc()
    elif http_code >= 200 and http_code < 300 :
        url_http_code.labels('2xx',url).inc()
    elif http_code >= 300 and http_code < 400 :
        url_http_code.labels('3xx',url).inc()
    elif http_code >= 400 and http_code < 500 :
        url_http_code.labels('4xx',url).inc()
    else:
        url_http_code.labels('5xx',url).inc()
    if http_total_time < 1 :
        url_http_request_time.labels('1',url).inc()
    elif http_total_time < 2 :
        url_http_request_time.labels('2',url).inc()
    elif http_total_time < 3 :
        url_http_request_time.labels('3',url).inc()
    else :
        url_http_request_time.labels('+Inf',url).inc()
    http_request_total.labels(url).inc()
    
#线程控制，每隔5s执行curl url
def count_threads(url):
    while True:
        t = threading.Thread(target=count_metric,args=(url,))
        t.setDaemon(True)
        t.start()
        time.sleep(5)
#将每个需要监控的域名起一个进程
if __name__ == '__main__':
    start_http_server(9091)
    server_list = [
            'www.baidu.com',
            'www.qq.com',
            'blog.csdn.net',
            'github.com',
            'google.com'
            ]
    threads = []
    for url in server_list:
        t = threading.Thread(target=count_threads,args=(url,))
        threads.append(t)
    for thread in threads:
        thread.setDaemon(True)
        thread.start()
    thread.join()

 坑：prometheus不会提醒metrics的名字，要主动复制进去：curl http://10.0.0.111:19091/metrics

复制：https://blog.csdn.net/specter11235/article/details/87927202

--------------------------------------------------------------------------------------

计算exporter的metrics的比率

#自定义exporter-counter
##状态码是500的个数
url_http_code_total{code="5xx",url="10.0.0.111:55555/a.txt"} 
#访问接口的个数
http_request_total{url="10.0.0.111:55555/a.txt"}
------------------------------------------------------------------
#错误率
delta(url_http_code_total{code="5xx",url="10.0.0.111:55555/a.txt"}[1m]) / on(url) group_left delta(http_request_total{url="10.0.0.111:55555/a.txt"}[1m])

#http code的每分钟增长率，如果出现5xx，就说明有问题了
irate(http_request_total[1m])

#显示期望时间的比例，比如只显示小于1秒，占总次数的比例
delta(url_http_request_time_total{le='1'}[1m]) / on(url) group_left delta(http_request_total[1m])
#复制
https://blog.csdn.net/specter11235/article/details/87927202