[libvirt] managedsave vs. transient domains

[Eric Blake]

How in the world is managedsave supposed to work on transient domains? 
The moment you save a domain, the qemu process is ended; but ending the 
qemu process means that a transient domain no longer has any state 
tracked by libvirt.

Should libvirt be temporarily defining a domain when creating a 
managedsave of a transient domain?  Seeing as how persistent domains 
cannot be marked autostart, does this mean that you have to use 'virsh 
start dom' rather than autostart to restart a transient domain with 
managedsave state?

Or should managedsave be prohibited on transient domains, and only exist 
as a possibility for persistent ones?  After all, 
qemuDomainHasManagedSaveImage() will always return false for a running 
domain - a managed save state file only exists if you have shut down the 
domain into managed state file with plans to later start from that file.

Personally, I'm leaning towards the latter - just like 
virDomainSetAutostart fails on transient domains, I think 
virDomainManagedSave should likewise fail on transient domains, and that 
applications that favor transient domains should also be managing state 
files themselves, via virDomainSave[Flags], rather than relying on 
virDomainManagedSave.

-- 
Eric Blake   eblake at redhat.com    +1-801-349-2682
Libvirt virtualization library http://libvirt.org