On Wed, Oct 24, 2012 at 5:19 PM, Musayev, Ilya <[email protected]> wrote:
> In single datacenter/zone i have many pods and many clusters to belong to 
> specific business units that are segregated on multiple levels ranging from 
> network to storage restrictions.
>
> For example I have 3 independent QA envs, 3  perf envs, 2 DEV, 2 LAB and 2 
> prod env Each env has it's own network, set of hypervisors and storage.
>
> If I follow simplified AWS like logic of creating 1 zone per datacenter, 
> multiple pods and clusters within - I loose the ability of being granular as 
> to where my instance should live.
>
> My env is an example of typical Corp environment with many segregated envs 
> for various business units.
>
> Creating 11+ zones and matching zone to cluster is also doable but - but i 
> think it's not how CS was designed to be used. Instead, we would like to have 
> 1 Zone per major business unit or even datacenter and have some other 
> allocators do VM allocation.
>
> Hence my question on best way of achieving this.
>
> Thanks
> Ilya
>
> On Oct 24, 2012, at 4:12 PM, "David Nalley" <[email protected]> wrote:
>
>> On Wed, Oct 24, 2012 at 4:04 PM, Musayev, Ilya <[email protected]> wrote:
>>> I need more granular ability to deploy on cluster level, as deploying on 
>>> the ZONE level is a bit too general in a complex environment like mine.
>>>
>>> My logic would be as follows for now:
>>>
>>> listClusters - find the cluster id I need
>>> listHost - used with clusterid to get the list of the hosts and get hostid
>>> deployVirtualMachine - deploy a VM and define hostid to make sure VM is 
>>> deployed on the desired cluster.
>>>
>>>
>>> This is certainly doable - but as you can see - there are multiple steps I 
>>> need to do in order to deploy on cluster level.
>>>
>>> Is there a simpler approach to this?
>>>
>>> Has anyone used allocators and what is their purpose - a usage example of 
>>> allocators would be great.
>>>
>>> As always, any feedback is appreciated.
>>
>>
>> I understand what you are wanting to do, but not why?
>> What advantage are you trying to achieve?
>>
>> --David
>>
>

If it were me I'd use a combination of domains (zones allocated to
domains) and tags. (tags the hosts - and have corresponding service
offerings for those tags - and then control which accounts/domains see
what service offerings.) I.e. QA folks would never see a service
offering for production, and thus couldn't get a machine deployed on
production hardware. In this scenario you would define all of the
hosts in your production cluster with a tag that matches a production
service offering.

Of course if you can define your allocation logic - an allocator
would like be better long term solution.

Reply via email to