kubernetes: Restarting containerd (and kubelet) does not impact "runtime" startTime in summary API

What happened: Containerd (and optionally the kubelet) was restarted.

What you expected to happen: runtime’s startTime should be the restart time.

$ curl -s localhost:10255/stats/summary | grep -i runtime -A 2
    "name": "runtime",
    "startTime": "2021-01-22T01:29:09Z",
    "cpu": {

How to reproduce it (as minimally and precisely as possible):

# Note the current startTime
$ curl -s localhost:10255/stats/summary | grep -i runtime -A 2

# Restart containerd
$ sudo systemctl restart containerd

# Note the current startTime
$ curl -s localhost:10255/stats/summary | grep -i runtime -A 2

I would expect the time to change. The reason why it does not is because cAdvisor is looking in the cgroup fs and finding the oldest modified time of cgroup.clone_children in all subsystems. In other words, for each of the supported subsystems, it’s looking at the oldest mtime of /sys/fs/cgroup/$subsystem/system.slice/containerd.service/cgroup.clone_children. When containerd is restarted, that file is unchanged.

https://github.com/google/cadvisor/blob/730e7df6dbddf323b4cdd54cc91156cfdd9cf127/container/common/helpers.go#L74-L91

In this code, for the runtime:

• cgroupPaths is a map (eg { cpu: /sys/fs/cgroup/cpu/system.slice/containerd.service/, memory: /sys/fs/cgroup/memory/system.slice/containerd.service/ })
• spec.CreationTime is set as the startTime in summary API

Anything else we need to know?:

Environment:

• Kubernetes version (use kubectl version): At least 1.16+ just from reading it and experiments.
• Cloud provider or hardware configuration: GKE
• OS (e.g: cat /etc/os-release): Container-Optimized OS however this seems OS-agnostic
• Kernel (e.g. uname -a): 5.4.49+ however kernel shouldn’t matter
• Install tools:
• Network plugin and version (if this is a network-related bug):
• Others: