pserv failing to serve on TFTP

[Julian Edwards]

If I run maas-test like this:

sudo maas-test --bmc-ip=10.0.0.101 --bmc-user=maas --bmc-password=foo --archive=ppa:maas-maintainers/dailybuilds --series=saucy --architecture=i386 eth1

(Notice the architecture)

When maas-test runs it gets as far as powering up the node which stalls when trying to TFTP the pxeconfig. In the pserv log I see:

        provisioningserver.kernel_opts.EphemeralImagesDirectoryNotFound: The directory containing the ephemeral images/info is missing (u'/var/lib/maas/ephemeral/saucy/ephemeral/amd64'). Make sure to run the script 'maas-import-pxe-files'.

Why does it need amd64 files when I explicitly asked for i386?

[Jeroen T. Vermeulen (jtv)]

The problem here looks to me as if the --architecture you pass has to match the actual architecture which the node will later report for itself. What seems to happen is: you select i386; maas imports images for i386; the node powers up; the node asks for an amd64 boot image, which was never imported.

Once the node is enlisted, we could use the API to check that its architecture matches the one you specified, and/or forcibly set it to the architecture you specified.

[Julian Edwards (julian-edwards)]

On 09/01/14 13:10, Jeroen T. Vermeulen wrote:
> The problem here looks to me as if the --architecture you pass has to
> match the actual architecture which the node will later report for
> itself. What seems to happen is: you select i386; maas imports images
> for i386; the node powers up; the node asks for an amd64 boot image,
> which was never imported.
>
> Once the node is enlisted, we could use the API to check that its
> architecture matches the one you specified, and/or forcibly set it to
> the architecture you specified.

Indeed, this is what is going on.

It's a nasty error as you have no idea what's up unless you debug it on
the VM like I did. Do we capture the maas logs anywhere in the test output?

[Jeroen T. Vermeulen (jtv)]

I think we attach the host-side logs, but apparently not the VM-side logs. I suppose we'd want to rsync those over when tearing down the tests.

[Jeroen T. Vermeulen (jtv)]

Actually, we should already retrieve them (though using "ssh cat", not rsync) and so they should be visible in the failure output.

[Julian Edwards (julian-edwards)]

On 09/01/14 17:07, Jeroen T. Vermeulen wrote:
> I think we attach the host-side logs, but apparently not the VM-side
> logs. I suppose we'd want to rsync those over when tearing down the
> tests.
>

It's surprising that it doesn't already, but yes, it should.

[Raphaël Badin (rvb)]

> It's surprising that it doesn't already, but yes, it should.

Logs definitely should be included. I'm working on bug 1267497 right now.