[openstack-dev] [nova][libvirt] Should file injection work for boot from volume images?

Michael Davies michael at the-davies.net
Thu Sep 26 22:12:17 UTC 2013


On Mon, Sep 23, 2013 at 6:20 PM, Thierry Carrez <thierry at openstack.org> wrote:
> Monty Taylor wrote:
> > On 09/20/2013 02:47 PM, Michael Still wrote:
> >> Before https://review.openstack.org/#/c/46867/ if file injection of a
> >> mandatory file fails, nova just silently ignores the failure, which is
> >> clearly wrong. However, that review now can't land because its
> >> revealed another failure in the file injection code via tempest, which
> >> is...
> >>
> >> Should file injection work for instances which are boot from volume?
> >> Now that we actually notice injection failures we're now failing to
> >> boot such instances as file injection for them doesn't work.
> >>
> >> I'm undecided though -- should file injection work for boot from
> >> volume at all? Or should we just skip file injection for instances
> >> like this? I'd prefer to see us just support config drive and metadata
> >> server for these instances, but perhaps I am missing something really
> >> important.
> >
> > Well, first of all, I think file injection should DIAF everywhere.
>
> +1
>
> > That said, it may be no surprise that I think boot-from-volume should
> > just do config drive and metadata.
>
> That sounds like the simplest way to preserve behavior. From what you
> said the current behavior is "try, fail and ignore failure" -- having
> noop instead is probably the right thing to do for havana.

This behaviour is what is causing https://bugs.launchpad.net/nova/+bug/1188543

I've submitted a patch (https://review.openstack.org/#/c/48533/) that
addresses the issue.

It appears that:
1) File injection for instances which are boot from volume doesn't
appear to have ever worked.
2) Attempting file injection just fails quietlyish and causes instance
spawning slowdown
3) The code needed to do this properly isn't trivial and probably
wouldn't land in Havana so late in the cycle.

Instead of attempting file injection on a boot volume, my patch simply
LOG.warns the user. I think that's the best solution for now. However
I think we should address file injection in Icehouse as discussed in
this thread.

Thanks in advance,

Michael...
-- 
Michael Davies   michael at the-davies.net
Rackspace Australia



More information about the OpenStack-dev mailing list