docker - perf监视未计数的Docker容器

Question

我想使用perf来监视docker容器:

$sudo perf stat -e cycles -a -I 1000 -o out.txt --cgroup=docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95 sleep 5

但是，它无法计算out.txt中的周期:

 1 # started on Thu Jun  7 22:49:34 2018
  2 
  3 #           time             counts unit events
  4      1.000224074      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  5      2.000515835      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  6      3.000805321      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  7      4.001008437      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  8      5.001234875      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95

并监视多个事件，例如:

$sudo perf stat -e cycles,instructions -a -I 1000 -o out2.txt --cgroup=docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95,docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95 sleep 5

在out2.txt中得到了相同的结果:

  1 # started on Thu Jun  7 23:23:15 2018
  2 
  3 #           time             counts unit events
  4      1.000326390      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  5      1.000326390      <not counted>      instructions              docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  6      2.000665181      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  7      2.000665181      <not counted>      instructions              docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  8      3.000979419      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
  9      3.000979419      <not counted>      instructions              docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
 10      4.001281621      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
 11      4.001281621      <not counted>      instructions              docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
 12      5.001233401      <not counted>      cycles                    docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95
 13      5.001233401      <not counted>      instructions              docker/c04e4686d0195f9863112983e472ec7b519642bf4cfd3856f327d02e05047f95

如何计算docker容器的周期？

我看了 Monitoring Linux performance events for cgroups和 kubelet fails to get cgroup stats for docker and kubelet services。

Answer 1

不适合使用--cgroup=。我应该使用-p配置进程ID，以监视特定进程。