一、压测目的

验证thanos-sidecar采集数据对比云原生prometheus性能影响。

**

二、压测准备

1、采集源机器配置

CPU：24核
MEM：125G

2、export数量

4000个节点

3、单条metrics大小

总大小为2100字节

  1. # HELP requests_total HTTP requests total
  2. # TYPE requests_total counter
  3. requests_total{clientip="127.0.0.1",dc="dx",ent="xfime1",method="get",sub="iat",type="percent"} 3.0
  4. requests_total{clientip="127.0.0.2",dc="dx",ent="xfime2",method="get",sub="iat",type="percent"} 3.0
  5. requests_total{clientip="127.0.0.3",dc="gz",ent="xfime3",method="get",sub="iat",type="percent"} 3.0
  6. requests_total{clientip="127.0.0.4",dc="hu",ent="xfime4",method="get",sub="iat",type="percent"} 3.0
  7. requests_total{clientip="127.0.0.5",dc="da",ent="xfime5",method="get",sub="iat",type="percent"} 3.0
  8. requests_total{clientip="127.0.0.17",dc="azx",ent="xfi6me",method="get",sub="iat",type="percent"} 3.0
  9. requests_total{clientip="127.0.0.14",dc="dax",ent="xfime6",method="get",sub="iat",type="percent"} 3.0
  10. requests_total{clientip="127.0.0.121",dc="dfxx",ent="xfime7",method="get",sub="iat",type="percent"} 3.0
  11. requests_total{clientip="127.0.0.11",dc="dxa",ent="xfime8",method="get",sub="iat",type="percent"} 3.0
  12. # HELP requests_created HTTP requests total
  13. # TYPE requests_created gauge
  14. requests_created{clientip="127.0.0.1",dc="dx",ent="xfime1",method="get",sub="iat",type="percent"} 1.621652127910199e+09
  15. requests_created{clientip="127.0.0.2",dc="dx",ent="xfime2",method="get",sub="iat",type="percent"} 1.6216521279102333e+09
  16. requests_created{clientip="127.0.0.3",dc="gz",ent="xfime3",method="get",sub="iat",type="percent"} 1.6216521279102564e+09
  17. requests_created{clientip="127.0.0.4",dc="hu",ent="xfime4",method="get",sub="iat",type="percent"} 1.6216521279102776e+09
  18. requests_created{clientip="127.0.0.5",dc="da",ent="xfime5",method="get",sub="iat",type="percent"} 1.621652127910298e+09
  19. requests_created{clientip="127.0.0.17",dc="azx",ent="xfi6me",method="get",sub="iat",type="percent"} 1.6216521279103184e+09
  20. requests_created{clientip="127.0.0.14",dc="dax",ent="xfime6",method="get",sub="iat",type="percent"} 1.6216521279103408e+09
  21. requests_created{clientip="127.0.0.121",dc="dfxx",ent="xfime7",method="get",sub="iat",type="percent"} 1.6216521279103608e+09
  22. requests_created{clientip="127.0.0.11",dc="dxa",ent="xfime8",method="get",sub="iat",type="percent"} 1.6216521279103804e+09

三、压测过程

1、压测思路

  1. 原生prometheu获取4000个节点数据源。
  2. prometheus+sidecar获取4000个节点数据源。
  3. 对比云原生prometheus和prometheus+sidecar的性能数据。

2、压测准备

采集4000个节点的metrics数据，每秒采集一次，单条metrics数据大小在2100个字节。

3、原生prometheus架构图

image.png

4、prometheus+sidecar+S3架构图

image.png

四、压测结果

1）原生prometheus的性能延迟数据

计算方法：采集4000个节点采集性能指标的平均值，采集半个小时+数据

4000云原生性能图.png
结论：平均4000个节点性能采集延迟在3ms

2）原生prometheus+thanos-sidecar性能延迟数据

计算方法：采集4000个节点采集性能指标的平均值，采集一个小时+数据

4000 sidecar抓取性能图.png
结论：平均4000个节点性能采集延迟在3.5ms

五、压测结论

根据性能数据分析和结论

  • 采集4000个节点metrics数据，thanos-sidecar比云原生prometheus数据采集影响在0.5ms左右。