Open Stack

Le 14/10/2014 01:46, Adam Lawson a écrit :
>
>     /I think Adam is talking about this bp:
>     https://blueprints.launchpad.net/nova/+spec/evacuate-instance-automatically/
>
>
> Correct - yes. Sorry about that. ; )
>
> So it would seem the question is not whether to support auto-evac but 
> how it should be handled. If not handled by Nova, it gets complicated. 
> Asking a user to configure a custom Nagios trigger/action... not sure 
> if we'd recommend that as our definition of ideal.
>
>   * I can foresee Congress being used to control whether auto-evac is
>     required and what other policies come into play by virtue of an
>     unplanned host removal from service. But that seems like a bit
>     overkill.
>   * i can foresee Nova/scheduler being used to perform the evac
>     itself. Are they still pushing back?
>   * I can foresee Ceilometer being used to capture service state and
>     define how long a node should be inaccessible before it's
>     considered offline. But seems a bit out of scope for what
>     ceilometer was meant to do.
>

Well, IMHO Gantt should just enforce policies (possibly defined by 
Congress or whatever else) so if a condition is not met (here, HA on a 
VM), it should issue a reschedule. That said, Gantt is not responsible 
for polling all events and updating its internal view, that's another 
project which should send those metrics to it.

I'm not having a preference in between Heat, Ceilometer or whatever else 
for notifying Gantt. I even think that whatever the solution would be 
(even a Nagios handler), that's Gantt at the end which would trigger the 
evacuation by calling Nova to fence that compute node and move the VM to 
another host (like rescheduling already does, but in a manual way).


-Sylvain


> I'm all about making this super easy to do a simple task though, at 
> least so the settings are all defined in one place. Nova seems logical 
> but I'm wondering if there is still resistance.
>
> So curious; how are these higher-level discussions 
> initiated/facilitated? TC?
>
I proposed a cross-project session at the Paris Summit about scheduling 
and Gantt (yet to be accepted), that usecase could be discussed there.

-Sylvain

>
> */
> Adam Lawson/*
>
> AQORN, Inc.
> 427 North Tatnall Street
> Ste. 58461
> Wilmington, Delaware 19801-2230
> Toll-free: (844) 4-AQORN-NOW ext. 101
> International: +1 302-387-4660
> Direct: +1 916-246-2072
>
>
> On Mon, Oct 13, 2014 at 3:21 PM, Russell Bryant <rbryant at redhat.com 
> <mailto:rbryant at redhat.com>> wrote:
>
>     On 10/13/2014 06:18 PM, Jay Lau wrote:
>     > This is also a use case for Congress, please check use case 3 in the
>     > following link.
>     >
>     >
>     https://docs.google.com/document/d/1ExDmT06vDZjzOPePYBqojMRfXodvsk0R8nRkX-zrkSw/edit#
>
>     Wow, really?  That honestly makes me very worried about the scope of
>     Congress being far too big (so early, and maybe period).
>
>     --
>     Russell Bryant
>
>     _______________________________________________
>     OpenStack-dev mailing list
>     OpenStack-dev at lists.openstack.org
>     <mailto:OpenStack-dev at lists.openstack.org>
>     http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev
>
>
>
>
> _______________________________________________
> OpenStack-dev mailing list
> OpenStack-dev at lists.openstack.org
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack-dev

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack-dev/attachments/20141014/21306b33/attachment.html>