node-installer(k3d|k0s): job does not terminate properly

[vdice]

#389 landed to ensure the node-installer job terminates properly for many K8s distros.

However, at least two known distros still see this behavior: k3d and k0s.

Revisit these to see if we can fix the behavior (ref #140 (comment) for original description).

[AntonOfTheWoods]

@vdice Quite likely just hit this... I got a refusal to launch the sample app so instead of using the helm instructions, stopped my k3d cluster and went through the quickstart (k3d-based, but with non-vanilla images) - everything worked. So I went back to try and fix my original cluster and voila, it was already working.

So is the workaround to just restart the cluster?

[vdice]

@AntonOfTheWoods it's been awhile since I looked at this behavior, but I recall that the first node-installer job would finish with 'Unknown' status, causing a subsequent job to run, which would finish with 'Completed' (rather than the expected behavior of just the one job running and finishing with 'Completed'). Otherwise, the shim binary itself should be successfully installed, even with this behavior. Thus, you may have run into a slightly different issue -- or my understanding could use a refresher 😅.

A full cluster restart shouldn't be necessary. Let me know if you still encounter this behavior and/or Spin App custom resources fail to run (presumably b/c the shim binary isn't installed or is misconfigured) and I'll revisit.