Reboot on takeover install

[zimme]

The current behaviour to poweroff during takeover installs is unexpected. This PR changes that behaviour to reboot instead.

[dweomer]

This might be better addressed by something like #143 which I've been meaning to validate and include in a release.

[zimme]

From what I can tell #143 fixes another problem entirely, it's fixing some reboot/poweroff problem when running the install.sh script. This fix is not for when you run install.sh but after that first reboot and k3os is actually installed, i.e. after the first reboot it installs k3os and powers down instead of reboots another time to start the new system.

[dweomer]

As discussed with @zimme in the rancher-users #k3os slack channel my current hold up with this is an as of yet not validated hunch that the current behavior is depended upon by the AWS/GCP packer builds. Working on validating this Soon ™️

[zimme]

I've tested the AWS packer thing now and it seems to still work. Running packer build ./template.json from form the aws directory with a changed region ´eu-north-1` creates a packer build instance and terminates it after an AMI has been created and I can launch an instance from that.

I don't know if it's a problem that the installed k3os system probably boots during the packer build stage so the AMI is generateed from a running k3os instance?

In the case before this change I guess the installed k3os instance during the build stage is never started and the AMI is generated from a k3os system that has never been started.

[zimme]

So I just checked this again and it seems my suspicions were correct.

Looking at the output of kubectl get nodes -o wide on a freshly deployed instance from the AMI build via packer with this change I'm seeing the following

NAME                                           STATUS     ROLES    AGE   VERSION         INTERNAL-IP     EXTERNAL-IP   OS-IMAGE       KERNEL-VERSION      CONTAINER-RUNTIME
ip-172-31-2-104.eu-north-1.compute.internal    NotReady   master   10h   v1.16.2-k3s.1   172.31.2.104    <none>        k3OS adfea69   4.15.0-66-generic   containerd://1.3.0+unknown
ip-172-31-13-221.eu-north-1.compute.internal   Ready      master   11m   v1.16.2-k3s.1   172.31.13.221   <none>        k3OS adfea69   4.15.0-66-generic   containerd://1.3.0+unknown
ip-172-31-13-221 [~]$

As you can see there are 2 nodes present, the one that's not ready is the master node that was configured during the first boot while building the AMI so we need to find a way to make the reboot on takeover install configurable in all stages of the takeover install or find another way of building an AMI.

[zimme]

Updated this to respect --poweroff instead if just blindly rebooting.

I used the updated check from #143 instead of the current check that's used so this is best merged in tandem with #143.

[zimme]

Fixed the check for k3os/system/poweroff and verified that it works on AWS.

I don't know if any other packer config needs updating too?

[zimme]

Is there any reason not to merge this and #143?

[thomasmhofmann]

I would also like to see this merged.

[zimme]

Just a note on this. There might be updates to packer configs to add --poweroff needed as I haven't tested all packer configs.

[thomasmhofmann]

Thank you. Working now.