MySQL master nodes on Kubernetes (AKS) dies for no apparent reason
You must check "kubectl get events" when this happens. Your pods use Burstable QoS type, so it can be Evicted in response to resource pressure. Evicted is not the same as OOMKill. Generally Evicted pods are shown in "kubectl get pod" output but only starting with some version of Kubernetes.
What you should do is to make Mysql pods Guaranteed, that means make requests and limits equal. Such important services like Mysql simply must not run as Burstable or Best Effort type.
Hope you already fixed this issue. By chance, I studied on this error, and it turned out to be a bug of containerd and was fixed since containerd 1.2.6 (docker 19.03.1).
Key information from your log:
level=info msg="shim reaped"level=warning msg="cleaning up after killed shim"References: