[ClusterLabs] Antw: Re: Antw: Re: Antw: Unexpected Resource movement after failover

[Ulrich Windl]

>>> Nikhil Utane <nikhil.subscribed at gmail.com> schrieb am 24.10.2016 um 13:22
in
Nachricht
<CAGNWmJXMLYnJjsJiPRmrKhcAPb3rNqX9xuWucqmWL5Qa8WwDgw at mail.gmail.com>:
> I had set resource utilization to 1. Even then it scheduled 2 resources.
> Doesn't it honor utilization resources if it doesn't find a free node?

Show us the config and the logs, please!


> 
> -Nikhil
> 
> On Mon, Oct 24, 2016 at 4:43 PM, Vladislav Bogdanov <bubble at hoster-ok.com>
> wrote:
> 
>> 24.10.2016 14:04, Nikhil Utane wrote:
>>
>>> That is what happened here :(.
>>> When 2 nodes went down, two resources got scheduled on single node.
>>> Isn't there any way to stop this from happening. Colocation constraint
>>> is not helping.
>>>
>>
>> If it is ok to have some instances not running in such outage cases, you
>> can limit them to 1-per-node with utilization attributes (as was suggested
>> earlier). Then, when nodes return, resource instances will return with
(and
>> on!) them.
>>
>>
>>
>>> -Regards
>>> Nikhil
>>>
>>> On Sat, Oct 22, 2016 at 12:57 AM, Vladislav Bogdanov
>>> <bubble at hoster-ok.com <mailto:bubble at hoster-ok.com>> wrote:
>>>
>>>     21.10.2016 19:34, Andrei Borzenkov wrote:
>>>
>>>         14.10.2016 10:39, Vladislav Bogdanov пишет:
>>>
>>>
>>>             use of utilization (balanced strategy) has one caveat:
>>>             resources are
>>>             not moved just because of utilization of one node is less,
>>>             when nodes
>>>             have the same allocation score for the resource. So, after
the
>>>             simultaneus outage of two nodes in a 5-node cluster, it may
>>>             appear
>>>             that one node runs two resources and two recovered nodes run
>>>             nothing.
>>>
>>>
>>>         I call this a feature. Every resource move potentially means
>>> service
>>>         outage, so it should not happen without explicit action.
>>>
>>>
>>>     In a case I describe that moves could be easily prevented by using
>>>     stickiness (it increases allocation score on a current node).
>>>     The issue is that it is impossible to "re-balance" resources in
>>>     time-frames when stickiness is zero (over-night maintenance window).
>>>
>>>
>>>
>>>             Original 'utilization' strategy only limits resource
>>>             placement, it is
>>>             not considered when choosing a node for a resource.
>>>
>>>
>>>
>>>         _______________________________________________
>>>         Users mailing list: Users at clusterlabs.org 
>>>         <mailto:Users at clusterlabs.org>
>>>         http://clusterlabs.org/mailman/listinfo/users 
>>>         <http://clusterlabs.org/mailman/listinfo/users>
>>>
>>>         Project Home: http://www.clusterlabs.org 
>>>         Getting started:
>>>         http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf 
>>>         <http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf>
>>>         Bugs: http://bugs.clusterlabs.org 
>>>
>>>
>>>
>>>     _______________________________________________
>>>     Users mailing list: Users at clusterlabs.org <mailto:
>>> Users at clusterlabs.org>
>>>     http://clusterlabs.org/mailman/listinfo/users 
>>>     <http://clusterlabs.org/mailman/listinfo/users>
>>>
>>>     Project Home: http://www.clusterlabs.org 
>>>     Getting started:
>>>     http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf 
>>>     <http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf>
>>>     Bugs: http://bugs.clusterlabs.org 
>>>
>>>
>>>
>>>
>>> _______________________________________________
>>> Users mailing list: Users at clusterlabs.org 
>>> http://clusterlabs.org/mailman/listinfo/users 
>>>
>>> Project Home: http://www.clusterlabs.org 
>>> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf 
>>> Bugs: http://bugs.clusterlabs.org 
>>>
>>>
>>
>> _______________________________________________
>> Users mailing list: Users at clusterlabs.org 
>> http://clusterlabs.org/mailman/listinfo/users 
>>
>> Project Home: http://www.clusterlabs.org 
>> Getting started: http://www.clusterlabs.org/doc/Cluster_from_Scratch.pdf 
>> Bugs: http://bugs.clusterlabs.org 
>>