[Ovirt-devel] ovirt dependencies

[Chris Lalancette]

Richard W.M. Jones wrote:
> I'm trying to build a definitive list of 'external' dependencies for
> oVirt.  By 'external' I mean dependencies on non-Fedora packages,
> network services, anything which needs a difficult or unusual
> configuration.
> 
> The underlying question here is what would it take to be able to
> simply 'yum install ovirt-wui' to create a WUI?
> 
> Please follow-up if I've missed any.
> 
> (1a) FreeIPA server
> (1b) Kerberos support in the browser
> 
> Does someone have Scott Seago's patches for "null" authentication?

I never saw them, so I can't help you there.

> 
> (2) DHCP, PXE, TFTP
> 
> At the moment we provide some very complex instructions for setting up
> dhcpd & TFTP.
> 
> DHCP is used for two things: (a) to pass the name of the PXE server to
> a booting node and (b) to pass a single configuration option to the
> managed node.  As Dan suggested, (b) could be done with zeroconf.
> (a) seems like it will always require configuration.

Right.  That being said, it is just the "next-server" directive, which I think
is a required part of the DHCP spec (that is, it is not some exotic DHCP option
record anymore).  However, for people who don't have control of their DHCP
server, it could be an issue.

> 
> PXE, TFTP is used to boot the managed nodes.
> 
> Can we run a self-contained dnsmasq (similar to the dnsmasq
> configuration used by libvirt) to do all of this work?  Yes, in as
> much as dnsmasq works for me to PXEboot the managed host.
> 
> Maybe we should mandate that in the "minimal" configuration people
> should always boot managed hosts using a USB key?

Eh.  I'm guessing that for the initial target, if people are using real,
physical machines as managed nodes, they'll either have a DHCP setup with PXE
that works, have the ability to do so, or can use a USB key.  I don't think we
really need to mandate this.

> 
> (3) Apache
> 
> The ovirt-wui RPM already drops the right files into /etc/httpd/conf.d
> to make an ovirt virtual host.
> 
> (4) PostgreSQL
> 
> Setting up databases is always hard: Should we create the database?
> What happens if the database already exists?  (Upgrades are hard to do
> and error-prone).  But leaving a SQL file around and asking the user
> to load it by hand seems reasonable enough.
> 
> I notice that the current ovirt-wui RPM leaves a script around to
> create the database but my ruby isn't good enough to tell how the full
> database schema is created.

There are essentially 4 parts for setting up the database (all of which the
appliance do at first-boot time):

1)  service postgresql initdb
2)  Run a series of postgresql commands to create the ovirt database user and
create the ovirt database (this is in /usr/share/ovirt-wui/psql.cmds on the
appliance)
3)  cd /usr/share/ovirt-wui ; rake db:migrate (this actually creates the needed
database tables)
4)  /usr/share/ovirt-wui/scripts/grant_admin_privileges admin (this grants
top-level admin privileges to the kerberos user "admin", so that he can then
manage the other users/resources)

> 
> (5) iSCSI server
> 
> This is a bit of an unknown quantity.  Can oVirt run without an iSCSI
> server?

Yes, the WUI can start without iSCSI.  What iSCSI is really required for right
now is for starting VMs on the managed nodes.  Given the storage API's, it
should be possible to use other things (like NFS, etc.); we just haven't coded
it up yet.

> 
> (6) collectd
> 
> Should soon be added to Fedora, so not a problem.
> 
> (7) Packages from Fedora: ruby, rails, kvm, libvirt, ntp, etc.
> 
> These aren't really an issue because RPM can deal with installing
> them.

And don't forget the custom RPMs we currently have in the ovirt repository,
notably updated libvirt with the storage patches + my custom iSCSI patch,
updated KVM (from rawhide), updated kernel (from rawhide), updated livecd-tools
(with rjones live-cd-to-pxe-script), custom ruby-libvirt bindings, and the
custom ruby-kerberos bindings.

Chris Lalancette