On Tue, Apr 03, 2012 at 02:08:11PM +0100, Daniel P. Berrange wrote:
On Tue, Apr 03, 2012 at 05:41:57PM +0800, Daniel Veillard wrote:
> On Tue, Apr 03, 2012 at 10:06:02AM +0100, Daniel P. Berrange wrote:
> > On Tue, Apr 03, 2012 at 04:22:59PM +0800, Daniel Veillard wrote:
> > - libvirt-daemon - Just the libvirtd daemon, no drivers, no configs
> > - libvirt-daemon-config-network - Just the virbr0 configs
>
> let's rename to libvirt-daemon-bridge then !
No, I explicitly chose to have -config' in the name, because this
really is just the config files, and we might want to use the
name 'libvirt-daemon-bridge' at some point in the future.
> > - libvirt-daemon-config-nwfilter - Just the firewall configs
>
> and libvirt-daemon-nwfilter
Again, I want config there to ensure that the libvirt-ademon-nwfilter
RPM name is availble for future use.
> the fact that the rpm contains config files, or .so, or binaries the
> user should care only of what functionality is provided.
The actual nwfilter functionality is provided in the libvirt-daemon
RPM, so to suggest they need to use libvirt-daemon-nwfilter is
misleading - this is only some default config files which apps
do not need to use if they don't want them - oVirt uses its own
configs for example.
okay ...
> If we split it that much then we must make sure that
> rpm -i libvirt-daemon-nwfilter
> will actually restart the server in the %post (and true for most of the
> others package, may turn into a challenge when installing 4 extension
> package and we want to avoid restarting the server 4 times !)
In the re-post I did, I use %posttrans script to ensure it is restarted
once, at the end of the transaction.
good !
> > - libvirt-daemon-driver-XXX - Contains the dlopen'd
modules per driver
> > or sub-driver. One for each XXX in: (uml,
> > qemu, xen, lxc, storage, network, nwfilter,
> > interface, nodedev, secrets)
>
> this is what I dislike, I would prefer to see
> libvirt-kvm and libvirt-daemon-driver-kvm to be merged (and hence
> my question about qemu and kvm because the .so would be shared I assume)
> unless there is a good story for having only the .so in the rpm
> that's the part I really reacted to when I made my tentative build for
> the release and that looks way to intricate to my taste.
It is not possible to merge these to. For backwards compatibility,
we need 'libvirt' to depend on 'libvirt-daemon-driver-qemu'. We
do *not*, however, want 'libvirt' to pull in 'qemu' itself.
It becomes clear when you consider what the entry points for
application dependancies are:
/---> libvirt --------+---------> libvirt-daemon
| | ^
| | |
| \---------> libvirt-daemon-driver-qemu
| | |
+---> libvirt-daemon-qemu ------------+--|------> qemu
| |
+---> libvirt-daemon-kvm ----------------+------> qemu-kvm
|
App Choice
So 'yum install libvirt', does *not* pull in 'qemu', 'qemu-kvm'
or 'xen'
If we merged the packages as you describe, we'd end up with
/---> libvirt --------+---------> libvirt-daemon
| | ^
| \-------------\ |
| | | |
| | V |
|----------------------------|--> libvirt-daemon-driver-qemu ---> qemu
| |
| \------\
| |
| V
|-------------------------------> libvirt-daemon-driver-qemu ---> qemu-kvm
|
App Choice
So, 'yum install libvirt' would end up pulling in every single hypervisor
we support (qemu, qemu-kvm, xen), which is not at all what we want.
Separating the libvirt-daemon-XXX packages from the libvirt-daemon-driver-XXX
packages is key to achieving the goal of minimising install footprint, while
maintaining backwards compatibility with existing RPM deps.
I still wonder if it is worth it then. Adding an extra empty rpm just
for the sake or avoiding a explicit hypervisor dependency at the
application level. The whole scheme adds N + 1 empty rpms just for
avoiding that dep that the application need to explicitely state right
now anyway.
Daniel
--
Daniel Veillard | libxml Gnome XML XSLT toolkit
http://xmlsoft.org/
daniel(a)veillard.com | Rpmfind RPM search engine
http://rpmfind.net/
http://veillard.com/ | virtualization library
http://libvirt.org/