collectors/cgroups: fix resolving container names in k8s

[ilyam8]

Summary

Fixes: #10006

Correctly resolve all kubepods/* cgroups names when running inside the k8s cluster (http req to the API server) or installed on a k8 node (via kubectl).

New naming format (breaking change):

• pod level cgroup name format: k8s_pod_<NAMESPACE>_<POD_NAME>
• container level cgroup name format: k8s_cntr_<NAMESPACE>_<POD_NAME>_<CONTAINER_NAME>

I don't add filtering to the cgroup-name.sh script, so all level cgroup are collected. I think it is better to have filtering in the cgroups collector configuration options (enable by default cgroups matching).

This PR doesn't contain enable by default cgroups matching option updates, i plan to add them in a separet PR (added in #10095).

Cgroup tree structure on the GKE and minikube nodes

  # GKE /sys/fs/cgroup/*/ tree:
  # |-- kubepods
  # |   |-- burstable
  # |   |   |-- pod98cee708-023b-11eb-933d-42010a800193
  # |   |   |   |-- 922161c98e6ea450bf665226cdc64ca2aa3e889934c2cff0aec4325f8f78ac03
  # |   |       `-- a5d223eec35e00f5a1c6fa3e3a5faac6148cdc1f03a2e762e873b7efede012d7
  # |   `-- pode314bbac-d577-11ea-a171-42010a80013b
  # |       |-- 7d505356b04507de7b710016d540b2759483ed5f9136bb01a80872b08f771930
  # |       `-- 88ab4683b99cfa7cc8c5f503adf7987dd93a3faa7c4ce0d17d419962b3220d50
  #
  # Minikube (v1.8.2) /sys/fs/cgroup/*/ tree:
  # |-- kubepods.slice
  # |   |-- kubepods-besteffort.slice
  # |   |   |-- kubepods-besteffort-pod10fb5647_c724_400c_b9cc_0e6eae3110e7.slice
  # |   |   |   |-- docker-36e5eb5056dfdf6dbb75c0c44a1ecf23217fe2c50d606209d8130fcbb19fb5a7.scope
  # |   |   |   `-- docker-87e18c2323621cf0f635c53c798b926e33e9665c348c60d489eef31ee1bd38d7.scope

This PR handles both cases, if there are other dir names formats it is likely it fails to extract podUID or containerID.

It is kind of difficult because cgroups collector passes a chart.id as an argument => all / changed to _ (_ is field delimiter...).

Component Name

collectors/cgroups

Test Plan

• test in GKE
  • copy modified cgroup-name.sh script to the children nodes
  • restart children netdata containers (docker restart <CONTAINER_ID>)
  • kubectl exec ... into the parent node, rm -rf * db cache
  • restart parent netdata container

I see all the kubepods/* cgroups on the dashboard.

• test in minikube (one agent setup)
  • copy modified cgroup-name.sh script to the agent
  • restart the agent container

I see all containers level cgroups on the dashboard. Pod and higher level cgroups filtered because of .slice suffix which doesn't match default match cgroup name simple pattern.

Additional Information