This message was deleted.

What are you actually trying to accomplish here? Kubernetes won’t reason about nodes that are down, because it doesn’t know if it is actually down and the pod is not running, or if the node is just not checking in due to a network outage or something. So you’ll see the pod go into Terminating but it won’t progress from there until either you delete the node, or the node comes back online and can confirm that the pod is not running.

We want to ensure that if a pod is Terminating (and the machine is actually off/offline) it doesn't block other pods from updating. And when that machine is back online it will then update.

yeah kubernetes won’t do that out of the box. You could add your own controller to force-delete pods from nodes that are down for a certain period of time, but Kubernetes won’t assume that just because a node is down the pod is actually terminated

What is the workload? Maybe something like Balena Cloud would be more suitable than k3s?

Thank you all for the answers! We looked at Balena, and I have used them previously. There were a few reasons why we would rather not use them, including driver support, vendor lock in, costs, etc... I think the controller to evict pods is probably what we will look at next.