[libvirt] [RFC][PATCH 3/7] interface: implement the public APIs

[Laine Stump]

On 05/19/2011 07:29 PM, Eric Blake wrote:
> On 05/10/2011 09:52 AM, Laine Stump wrote:
>> (I want to make sure that nobody thinks opening a transaction means that
>> all the changes are saved into a staging area and then later committed
>> to the live config all at once when the transaction is committed - doing
>> it this way  would eliminate the ability to test the new config prior to
>> committing).
> But suppose that the changes being made are such that I'm swapping
> connectivity from one interface to another during the course of the
> changes.  If I make the changes live, then there is a window where
> either both interfaces are connected, or where neither interface is
> connected, and both scenarios have implications on how I connect to
> issue the remainder of the commands needed to commit to the change.
>
> Maybe it is worth _also_ having the ability to queue up several changes,
> and apply them all at once, so that a single interface can queue up all
> the proposed changes and finally try out the batch, and the second
> interface then commits if all went well, rather than having to
> coordinate the handoff of half the changes being made by interface 1 and
> the other half by interface 2.

Not maybe. Definitely. That's a different feature though :-)