Open Stack

Hi guys,

Is there anyone working on this yet? If not, I'd like to help.

-Lianhao 


JC Martin wrote on 2012-11-21:
> I believe that it's critical for openstack to manage physical hardware,
> even more so than VM. The VM management is the responsibility of the
> tenant, and can be achieved from the VM, through some external services.
> Openstack can provide VM performance management but I believe that it's
> not core. On the other end, it should be the primary responsibility of
> openstack to make sure that its infrastructure is healthy. This includes :
>    - Monitoring Hypervisor helath, capacity, performance
>    - Predicting and repairing Hardware failures (e.g. Disk SMART)
>    - Recovering VMs from failed hardware
> This, in the order on importance. This would also include the monitoring
> and management of the associated components (nova scheduler, APIs, ...)
> 
> Thanks,
> 
> JC
> 
> 
> On 11/20/12 3:50 AM, "Jiang, Yunhong" <yunhong.jiang at intel.com> wrote:
> 
>> Firstly, openstack is mostly a system manage the VM instance (create,
>> schedule, desotry etc), instead of the host system, thus host management
>> itself does not belongs to openstack scope IMHO.
>> 
>> Secondly, usually there are already some monitor/management system in
>> data center that manages the host, checking IPMI event etc, and those
>> system is not part of openstack.
>> 
>> However, the host system state IS really important to openstack, for
>> example, if some system is over-hot, it's important to notify nova
>> scheduler to stop load instance to that system etc.
>> 
>> So I think the best solution is, host management system will input data
>> to ceilometer, and some system (either part of ceilometer, or 3rd party
>> software) will monitor these data, will provide policy/action to handle
>> host situation. As these monitor system exist already, I think a REST API
>> will be much easier to wrap them, rather than ceilometer message bus. And
>> I don't think we need re-invent the wheel in openstack to manage the host
>> system, which is not openstack's scope.
>> 
>> Any idea?
>> 
>> Thanks
>> --jyh
>> 
>> Zehnder Toni (zehndton) wrote on 2012-11-20:
>>> Hi,
>>> 
>>> I can't see why an internal resource should send data via REST API post.
>>> Could you explain why a REST API post is required?
>>> I think the new agent shouldn't send data in a different way than the
>>> existing
>>> agents.
>>> 
>>> Toni
>>> 
>>> Jiang, Yunhong wrote on 2012-11-15:
>>>> As I'm beginning work on
>>>> https://blueprints.launchpad.net/ceilometer/+spec/meter-post-api, so
>>>> my question to the architect is, would it be possible for the
>>>> hardware monitor to send information through REST API, instead of the
>>>> event bus? That will make the hardware agent simpler IMHO.
>>>> 
>>>> Thanks
>>>> --jyh
>>>> 
>>>> Zehnder Toni (zehndton) wrote on 2012-11-15:
>>>>> Hi there,
>>>>> 
>>>>> 
>>>>> 
>>>>> As we discussed last week there is a need for hardware monitoring in
>>>>> the OpenStack environment. I worked out a blueprint (see
>>>>> 
>>>>> https://blueprints.launchpad.net/ceilometer/+spec/monitoring-physical
>>>>> - device s
>>>>> 
>>>>> <https://blueprints.launchpad.net/ceilometer/+spec/monitoring-physica
>>>>> l -devic es> ) for this case. There is also a diagram of the
>>>>> architecture on es> the wiki (see
>>>>> http://wiki.openstack.org/Ceilometer/MonitoringPhysicalDevices) so
>>>>> we can discuss about it more specifically.
>>>>> 
>>>>> 
>>>>> 
>>>>> Toni