This message was deleted.

I guess what I'm asking is: Should I always drain before restarting rke2-agent.service ?

agents are pretty safe, servers are another story

"pretty safe" ? are there any official guidelines on this? Like for example: A couple of versions back of containerd, containerd had the habbit of leaving behind orphan containers when it got restarted

agents don't have "state" or "etcd" so even if not draining it will boot up processes on another available node

I know that. 🙂 But still it might not always be completely safe to restart daemons.

it might be, depending on the environment normally even if you're stopping it signals the master it is being unavailable

"it might" .. That's exactly the thing 🙂 So, my original question was really: Does RKE2 aim to make restart of the agent service graceful? Or is there an official recommendation to always drain?

I haven't found it, but depends on your expectations and experience

my experience with kubernetes in general since 2017 is not good with regards to just restarting components 😄 But I have less experience with RKE ... I will just go with draining I thing, at least for production.

I only have experience with rke2

One thing I'd warn you about if you're doing reboots, is if you're also doing patches that RKE2 I installed on CentOS via the install script (allowing default of RPM install) put yum repos to auto update RKE2 when I did my yum upgrade, so I did run into some problems with that happening unexpectedly (note the RKE2 deployed from Rancher uses the TAR method and will NOT update with a yum update). I assume the SuSE & Ubuntu do the same for repos.