fallback from specific to generic subarch broken

[Michael Hudson-Doyle]

I think this is basically bug 1294131 but trying to commission a midway node fails, unless you undertake massive hacking (edit database, file system paths, tgt config) to pretend that the armhf-generic images you imported were armhf-highbank images.

[Michael Hudson-Doyle (mwhudson)]

A better workaround is to hack pxeconfig in api.py to ignore the passed in subarch and always use generic.

[Julian Edwards (julian-edwards)]

I'm trying to get my head around this but failing miserably. Can you describe a scenario to recreate it please?

[Michael Hudson-Doyle (mwhudson)]

Did you see https://bugs.launchpad.net/maas/+bug/1294131/comments/1 ? basically instead of returning file not found, probing for default-arm-highbank returns a syntactically valid config file that refers to "no-such-image".

[Julian Edwards (julian-edwards)]

Right, thanks, I just re-created this:

If I mark a node's subarch as, for example, hwe-s, then I can ask to boot "precise" and it works fine. If, however, I ask to boot trusty which does not have hwe-s kernels (it's newer), then you get a request for this boot file in pserv.log:

amd64/hwe-s/trusty/no-such-image/boot-kernel

What should have happened is that the subarch is overridden to generic because trusty is newer than hwe-s. Remember that the subarch lists a *minimum* kernel version.

[Julian Edwards (julian-edwards)]

I think this is related to but not quite the same as bug 1294131, because that one is talking about genuinely missing images. This is more about downgrading the required subarch.

[Raphaël Badin (rvb)]

@mwhudson: note that you can work around that bug by editing the cluster's /etc/maas/bootresources.yaml file and forcing it to "import" the highbank images (add a section like http://paste.ubuntu.com/7466068/ in /etc/maas/bootresources.yaml and trigger the import again). This will create a symlink from "highbank" to "generic" and let MAAS know that the highbank images are available.