On Thu, Oct 06, 2016 at 10:59:06AM +0100, Ian Jackson wrote:
>Martin Kletzander writes ("Re: [OSSTEST PATCH 2/2] libvirt: Do not attempt save/restore when migration not advertised"):
>> Since offline migration (as in migrating a domain between hosts without
>> being running) is not that used in the code and talked about, I'm
>> guessing offline means save restore.  Looking at the history it was
>> added before the "offline" migration, so it probably means
>> save/restore.  To avoid confusion, I would suggest we add either
>> <offline/> or rather <save/> (the naming is not important) and document
>> what it means.  And then you can use it exactly how you'd like.  And
>> you'll be also sure it means what you need it to mean ;)  The patches
>> will be straigh-forward, let me know if I can help anyhow.
>
>Except that the point of the exercise is to detect which features are
>supported in which versions.  Whatever I do in osstest needs to work
>with older libvirt versions, which do not report
>  /capabilities/host/migration_features/save
>even on x86, where it is supported.  I suppose I could detect
>  /capabilities/host/migration_features/live
>and assume that save/restore was supported (since it's unlikely that
>live migration would be supported but not save/restore).
>
>So for now I think I need to use
>  /capabilities/host/migration_features
>as a proxy for save/restore ?
>

Well then, unfortunately you do.

Also, looking at how the code is structured, if you have live migration
but don't have save/restore, you won't have <migration_features/> there
at all.

>Ian.