What happened:
Node became NotReady, most likely because kubelet/docker was OOM killed (in our case this killed the Gardener API server just because a logging curator job/pod consumed too much memory).
What you expected to happen:
More buffer before this happens, so that the kubelet can take action in time..
Anything else we need to know?:
While trying to understand why this happens, we (@adracus and I) saw that the reserved memory is configured to only 100Ki, which may lead to the observed instability.
What happened: Node became
NotReady, most likely becausekubelet/dockerwas OOM killed (in our case this killed the Gardener API server just because a logging curator job/pod consumed too much memory).What you expected to happen: More buffer before this happens, so that the
kubeletcan take action in time..Anything else we need to know?: While trying to understand why this happens, we (@adracus and I) saw that the reserved memory is configured to only 100Ki, which may lead to the observed instability.