[libvirt-users] A RFE for libvirt

[Osier Yang]

[ CC to libvir-list for further discussing ]

On 2012年06月07日 20:20, Eric Blake wrote:
> On 06/07/2012 03:04 AM, Osier Yang wrote:
>
>> I once thought the RFE is not a valid request from a command p.o.v
>> (if we provide too much options for the various config tags, it's
>> just duplicate work of tools like virt-install, and somehow it's
>> even not better than XML), but now IMHO it'd be wonderful if we
>> have the best default configuration template for each driver. And
>> if so, using the templates to implement the command will make sense.
>
> The idea of templates has merit.  I could see it being similar to how we
> have<nwfilter>  and even<network>  as templates, where a<domain>  can
> refer to one of these other entities by name instead of calling out the
> details itself.  That is, we would add a new virDomainTemplatePtr, whose
> XML is a subset of<domain>  XML, then you could write a domain that uses
> <domain template='xyz'>...</domain>  which picks up defaults from
> template xyz instead of the current default of omitting anything not
> specified.  A template could specify typical memory usage, things such
> as default disk policy (cache='writethrough' vs. cache='none', use of
> virtio, and so forth), a default graphics device (spice vs. vnc setup,
> or serial console only), existence of the plumbing for a guest agent,
> and so forth.

It's quite good thought from my p.o.v. That means we don't have to
focus too much on what are the best defaults for each driver, as
we could have multiple templates, each of them could use the
best choices for different purposes.

I could also see libosinfo providing a set of templates
> for use in conjunction with various guest OS, even having such things as
> a template tuned for a guest acting as a mail hub vs. a different
> template tuned for a guest acting as a database server, or a template
> that knows Windows and Linux support a guest agent vs. a guest that
> lacks an agent.

Sounds excited. Yeah, libosinfo could help.

>
> But it sounds like quite a bit of work,

Yes. Work will be big.

  and it's not at the top of my
> personal to-do list (since I have a lot on my plate for improving
> snapshot and storage volume management).  I'm more throwing the idea out
> there in the hopes that someone else will be able to take and run with
> it (I will gladly review patches, though!)

It needs to figure out lot of stuffs before the real work, I'm willing
to do the work if we have a clear plan after the discussion and if
I'm not buried in the work on my plate too much.

Regards,
Osier