On 04/14/2013 05:59 PM, Doron Fediuck wrote:
----- Original Message -----
> From: "Itamar Heim" <iheim(a)redhat.com>
> To: "Laszlo Hornyak" <lhornyak(a)redhat.com>
> Cc: "engine-devel" ovirt.org>
> Sent: Sunday, April 14, 2013 5:04:55 PM
> Subject: Re: [Engine-devel] watchdog
>
> On 04/11/2013 03:54 PM, Laszlo Hornyak wrote:
>> Hi,
>>
>> It is not really an optimization based only on console connection. In my
>> opinion this should be a decision factor when determining which VM to
>> migrate from an overloaded host. If it is a desktop and no console
>> connection, then it is better candidate than a desktop with console
>> connection or a server with or without console connection. Basically this
>> logic would just want to minimize the pain caused by the short lag when a
>> VM is live migrated.
>>
>> Can I assume that a Desktop is only used when a console is connected?
>
> connected by spice? by vnc to host or to guest? by rdp? by remote
> debugging to it from a remote machine, by someone developing on it with
> ssh, etc...
>
> i agree the general use case you are describing make sense, but i rather
> we model scheduling decisions directly if/where possible
Some parts of this thread are looking for perfection and on the way
creating additional tasks and reviewing cycles that are only remotely
related to the watchdog device;
Yes, we should try and reduce the existing differentiation between
desktop and a server.
No, this feature is not about closing that gap. Watchdog device is
about handling guest crash.
I strongly suggest that merging desktop and server dialogues will be
done as a different feature. The watchdog feature will take one step in
the 'right' direction- adding the (currently missing) HA tab into the
desktop dialog based on current design.
1. it is a different task.
2. i'm suggesting you can drop the code doing that distinction for the
watchdog in this patch.
Doron