Open Stack

This seems apropos given something I ran into today with
/var/lib/libvirt/qemu/save not having enough capacity to store my suspended
instances. If the RAM is freed up (and state saved to disk) then sizing the
required size of the qemu partition is pretty big. Based on a rough
back-of-napkin calculation you could have up to ( ( max_instances_per_host
- 1 ) * physical_memory * memory_overcommit_ratio) disk used by saved
instances. For a 128GB machine that's 10T of data. I'd be pretty unhappy
about having to add that extra disk to my compute nodes :)

If folks really want to release the resources for a suspended machine
perhaps an approach similar to what AWS seems to do would work. By this I
mean we would save the instance disk/RAM out to the image store available
for future resumption and rescheduling on another machine.

-Aaron


On Thu, Jun 26, 2014 at 12:00 AM, Wangpan <hzwangpan at corp.netease.com>
wrote:

>   Oh, sorry, I misunderstood your idea,
> Ok, then the last problem is how to release the vcpus/mem resources of
> suspended instances on the compute nodes?
>
> 2014-06-26 11:57 (UTC+8)
> Wangpan
>
> ----- Original Message -----
> > From: Ricky Saltzer <ricky at cloudera.com>
> > To: "Wangpan"<hzwangpan at corp.netease.com>
> > Sent: 2014-06-25 03:58
> > Subject: Re: Re: [Openstack] Why doesn't suspend release vCPUs/memory?
>
> I don't see how there would be a problem if you still enforced a hard
> limit of the number of instances a user can have. The proposal here is
> suspending would release *only* vCPUs and memory, so the user could still
> potentially exceed their quota by reaching their allotted amount of
> instances. Let's say 5/10 instances for a user accumulated to 20 vCPUs (4
> vCPUs per node), suspending those 5 instances would release the 20 vCPUs
> back, but those 5 suspended instances would still count towards their total
> number of instances quota.
>
>
>
>
> On Mon, Jun 23, 2014 at 11:24 PM, Wangpan <hzwangpan at corp.netease.com>
> wrote:
>
>>   I think if we do this, a serious security risk is imported, think this
>> use case:
>> 1) an user has quotas like 10 instances, 20 vcpus, 100G ram and 200G disks
>> 2) he boots 10 instances under his quotas
>> 3) he suspends all this instances
>> 4) he repeats step 2&3 day and night
>> 5) then the cloud platform will have no resources to supply eventually
>>
>> 2014-06-24 11:17 (UTC+8)
>> Wangpan
>>
>> ----- Original Message -----
>> > From: Ricky Saltzer <ricky at cloudera.com>
>> > To: "John Griffith"<john.griffith at solidfire.com>
>> > Sent: 2014-06-24 01:05
>> > Subject: Re: [Openstack] Why doesn't suspend release vCPUs/memory?
>>
>>  That seems to be the case, and I can see where you're coming from, but
>> if the resources aren't released at the quota level, then they're
>> effectively being used from a user's point of view. It would be nice if
>> *suspend* released resources after the instance is shutdown, and a
>> *resume* would reclaim the resources (provided enough are available).
>> For instance, if I had 210/210 vCPUs used, and I suspend *instance_a*
>> with 1 vCPU, and then launch *instance_b *with 1 vCPU...*instance_b *should
>> successfully deploy, but resuming *instance_a* should fail with a quota
>> exceeded exception.
>>
>>
>>
>> On Mon, Jun 23, 2014 at 12:54 PM, John Griffith <
>> john.griffith at solidfire.com> wrote:
>>
>>>
>>>
>>>
>>> On Mon, Jun 23, 2014 at 10:49 AM, Ricky Saltzer <ricky at cloudera.com>
>>> wrote:
>>>
>>>> Right, the quotas don't seem to be released. If I have 210/210 vCPUs
>>>> used, and I suspend an instance with 4 vCPUs, I still have 210/210 vCPUs
>>>> used.
>>>>
>>>>
>>>> On Mon, Jun 23, 2014 at 11:38 AM, John Griffith <
>>>> john.griffith at solidfire.com> wrote:
>>>>
>>>>>
>>>>> On Mon, Jun 23, 2014 at 7:45 AM, Ricky Saltzer <ricky at cloudera.com>
>>>>> wrote:
>>>>>
>>>>>>
>>>>>> https://ask.openstack.org/en/question/32826/why-doesnt-suspend-release-vcpusmemory/
>>>>>
>>>>>
>>>>> ​My understanding was always that the instance is no longer consuming
>>>>> any resources via the virt layer, so in essence the resources are in fact
>>>>> freed up on the Compute Node.  Quotas and such however aren't modified
>>>>> (which seems correct to me).  Are you saying you want to see quota's
>>>>> adjusted here? ​
>>>>>
>>>>>
>>>>
>>>>
>>>> --
>>>>  Ricky Saltzer
>>>> http://www.cloudera.com
>>>>
>>>> ​Yeah, I think that makes sense and is expected, as a user you're still
>>> consuming those "items" even if they're not active.  The alternative would
>>> be (which I think is what you're getting at) to actually deduct items that
>>> are suspended from the tenants quota count.  I guess when I think of it
>>> though those resources are still "reserved" even if they're not in use.  I
>>> suppose you could do this and then if on resume the quota isn't there we
>>> don't actually resume... but I think this could be argued either way.
>>>
>>> Maybe seperate quotas for active vs suspended?  ​
>>>
>>>
>>
>>
>> --
>>  Ricky Saltzer
>> http://www.cloudera.com
>>
>>
>
>
> --
>  Ricky Saltzer
> http://www.cloudera.com
>
>
> _______________________________________________
> Mailing list:
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
> Post to     : openstack at lists.openstack.org
> Unsubscribe :
> http://lists.openstack.org/cgi-bin/mailman/listinfo/openstack
>
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.openstack.org/pipermail/openstack/attachments/20140701/fd467932/attachment.html>