Re: [meta-virtualization][PATCH] k3s: Bump to v1.19.3+k3s3

["Bruce Ashfield" <bruce.ashfield@gmail.com>]

On Fri, Dec 18, 2020 at 9:24 AM Joakim Roubert <joakim.roubert@axis.com> wrote:
>
> On 2020-12-15 19:58, Bruce Ashfield wrote:
> >
> > For now, I'd suggest waiting a couple more days, since I may have it
> > solved shortly.
>
> If there continues to be strange problems, perhaps the Rancher guys
> would be interested in helping out? They are both skilled and friendly,
> and given Rancher's focused agenda on being _the_ Kubernetes provider
> for edge with k3s, I guess having k3s in meta-virtualization would be
> attractive to them.

I have reached out to them, but had to put down my debug on this for a
few days as I prepared 5.10 as the new reference kernel for oe-core.

I'm almost through that, and will cycle back to this shortly.

The bottom line is that only the rancher build k3s binary works with
our constructed rootfs (in my tests).

If I use our built binary, or one that I build outside of yocto
(natively, on my build server), we end up in crashloops on all the
core containers. The binaries are built from the identical hash (for
yocto) and using the official build container images / process on my
workstation .. same result. So that rules out something in the kernel
(and its config), the go version or something in our rootfs being
missing. But as soon as you fetch the rancher binary, restart the
services .. the containers come up just fine.

And that unfortunately means that it is difficult to get help from the
project, since they don't have the right understanding of the runtime
environment (but I'm trying!).

I'm diving deeper to try and sort it out.

Bruce

>
> BR,
>
> /Joakim
>


-- 
- Thou shalt not follow the NULL pointer, for chaos and madness await
thee at its end
- "Use the force Harry" - Gandalf, Star Trek II