The logic introduced in PR #85 that is meant to restart a running kubelet if containerd's cgroup driver gets updated to make kubelet pick up that change as well is using systemctl is-active kubelet.service to determine if kubelet is running or not. If the kubelets systemd unit is either active or reloading, systemctl will return with 0 and thus, kubelet gets restarted. However, there is another status that may require a kubelet restart: activating. Unfortunately this status produces return code 3 and leads to the kubelet not getting restarted even though that would be absolutely necessary. This situation can lead to a race which leads to the problems described in issue #98.
Which issue(s) this PR fixes:
Fixes #98
Release note:
A condition that may lead to containerd and kubelet use different cgroup drivers in rare cases was fixed.
danielfoehrKn
commented
11 months ago
How to categorize this PR?
/area os /kind bug /os garden-linux
What this PR does / why we need it:
The logic introduced in PR #85 that is meant to restart a running kubelet if containerd's cgroup driver gets updated to make kubelet pick up that change as well is using
systemctl is-active kubelet.serviceto determine if kubelet is running or not. If the kubelets systemd unit is eitheractiveorreloading,systemctlwill return with0and thus, kubelet gets restarted. However, there is another status that may require a kubelet restart:activating. Unfortunately this status produces return code3and leads to the kubelet not getting restarted even though that would be absolutely necessary. This situation can lead to a race which leads to the problems described in issue #98.Which issue(s) this PR fixes: Fixes #98
Release note: