[Ovirt-devel] Enumerating node information to the WUI

[Chris Lalancette]

Darryl L. Pierce wrote:
> CPU: the family, model, cpuid level, cache size, number of cores and the 
> set of CPU flags. This would give us a nice idea of the individual CPU's 
> capabilities.
> 
> As a result, we'd want to refactor the hosts table a bit. Rather than 
> capturing as a column the number of CPUs, we could get that implicitly 
> by having a record-per-cpu for the host. Then each row would capture the 
> important details from above.

Well, this is kind of interesting.  After thinking about it some, having the
additional CPU information would be worthwhile.  It won't necessarily help
taskomatic make better decisions at this point, but if we add SLA's for "start
on a Barcelona processor" or whatever in the future, this stuff could be useful.
 A record-per-cpu might be overkill, though; one CPU record per host might be
enough.

> 
> MEM: We currently return the total system memory reported but should 
> probably also grab the free memory after the managed node image is loaded.

Unfortunately, the free memory number is basically useless, since Linux is going
to be using lots of memory for page cache pages in the future.  Not something I
think we need to track.  We *do* need to track (in the database) how much memory
we have given to VMs over on a particular machine so we don't load it up too
much, but free memory on the machine itself is not this number.

> 
> DISK: Returning the physical and logical partitions with free space 
> available.

Hard to say.  We've been going with the idea of totally diskless machines until
now, with totally remote storage.  That being said, if we are going to support
other things in the future (like Xen), we will need some sort of stuff for
disks.  This needs more thought; I would punt on this one for now until we
better understand our requirements for local disks.

> 
> NETWORK: Iterating over the physical interface adaptors. Also returning 
> any IP addresses already assigned and routing information.

Yes, definitely.  Along with this is a total DB schema redesign of the network
parts, since it is pretty bad right now.  This also goes hand-in-hand with
apevec's "boot-time" changes; basically, you start up, DHCP on one interface,
fetch your network configuration information from the WUI, and then re-configure
your interfaces based on the config info.  More thought needs to go into how
this initial "auto-discovery" phase fits in with that plan.

Chris Lalancette