[ovirt-users] VM "hot" RAM memory decrease

[Martin Sivak]

Hi Nicolas,

you are right, the log shows exactly the right values. Those numbers
are exactly what is sent to libvirt using currentMemory field in
https://libvirt.org/formatdomain.html#elementsMemoryAllocation

So if the VM does not have enough memory, we will have to start looking there.

Best regards

Martin Sivak

On Wed, Mar 29, 2017 at 11:04 AM,  <nicolas at devels.es> wrote:
> Hi Martin,
>
> I do, I'm attaching it. I merged the full day's log into one. Significant
> hours start at 11:24:57 (log time). The interesting machine is called
> jbaspre01, and as per what I can deduce, the values in log are correct (it
> never goes under the threshold of 1GB, which is its minimum). However, my
> colleagues and I are 100% sure we saw this machine having about 600K KiB
> when running top.
>
> This machine has:
> Memory size: 8196 MB
> Minimal guaranteed memory: 1024 MB
>
> If you need additional information don't hesitate to ask.
>
> Regards.
>
>
> El 2017-03-29 09:44, Martin Sivak escribió:
>>
>> Hi Nicolas,
>>
>> KiB Mem tells you how much memory the host has. I am actually more
>> interested in the VM. Do you still have the mom.log from the time this
>> happened? We keep the logs for some time (they are rotated).
>>
>> Best regards
>>
>> Martin
>>
>> On Wed, Mar 29, 2017 at 10:28 AM,  <nicolas at devels.es> wrote:
>>>
>>> El 2017-03-28 15:33, Martin Sivak escribió:
>>>>>
>>>>>
>>>>> min guaranteed should be respected, adding mom maintainer Martin
>>>>
>>>>
>>>>
>>>> I can't really help without the mom.log and the virsh dumpxml output
>>>> of that VM in the final state (all memory ballooned).
>>>>
>>>> Min guaranteed should be respected, but the question is how the
>>>> physical memory was measured (which number from top was used).
>>>>
>>>> --
>>>> Martin Sivak
>>>> SLA / oVirt
>>>>
>>>
>>> Hi Martin,
>>>
>>> I tried to reproduce the issue again and I was unable to. I had a host
>>> with
>>> ~90% of RAM utilization and in mom.log I could see that ballooning was
>>> indeed happening, but I couldn't see a VM going under its minimum
>>> guaranteed
>>> threshold (at least not significantly).
>>>
>>> On top, I was using the 'KiB Mem' value to compare. I know this is in KiB
>>> and that oVirt has its values as MB in the webadmin, but when I sent this
>>> message the difference was quite more significant (1024MB of minimum
>>> guaranteed and about 640K KiB on the VM).
>>>
>>> It's also true that meanwhile we upgraded from 4.0.2 to 4.1.1 and this
>>> time
>>> I couldn't reproduce it anymore.
>>>
>>> If at some point I am able to reproduce it again, I'll send detailed logs
>>> of
>>> the issue.
>>>
>>> Thanks.
>>>
>>>
>>>> On Tue, Mar 28, 2017 at 4:29 PM, Michal Skrivanek
>>>> <michal.skrivanek at redhat.com> wrote:
>>>>>
>>>>>
>>>>>
>>>>>> On 27 Mar 2017, at 23:15, Nicolás <nicolas at devels.es> wrote:
>>>>>>
>>>>>> Hi,
>>>>>>
>>>>>> I don't really know if this question is suitable on this list, as I
>>>>>> doubt it's an oVirt bug, neither I know if it shall be considered a
>>>>>> bug.
>>>>>>
>>>>>> We recently run a VM on a host that was memory over-used (around 80%
>>>>>> of
>>>>>> usage). The VM booted correctly, then we run "top" and saw how
>>>>>> physical RAM
>>>>>> started decreasing every two seconds. At first it was 4GB, then less
>>>>>> and
>>>>>> less until it stabilized at around 600MB.
>>>>>>
>>>>>> Based on this (correct me if I'm wrong), we believe this is an effect
>>>>>> of
>>>>>> having this VM with ballooning enabled, as it does exactly this: It
>>>>>> adds/removes RAM depending on host decision. Thing is that this VM had
>>>>>> a
>>>>>> minimal guaranteed RAM of 1GB, so even if this happened due to
>>>>>> ballooning,
>>>>>> I'm not sure if it should have honored the minimum guaranteed RAM.
>>>>>>
>>>>>> When this happened, we run a 'ps' to see with what options the qemu
>>>>>> process was invoked, and parameters seemed correct (that's why I don't
>>>>>> know
>>>>>> if it should be posted here or even if it's a bug).
>>>>>
>>>>>
>>>>>
>>>>> It does belong here, it’s a different component doing it “externally”
>>>>> to
>>>>> the qemu process- mom.
>>>>>
>>>>>>
>>>>>> Is this the expected behavior?
>>>>>
>>>>>
>>>>>
>>>>> min guaranteed should be respected, adding mom maintainer Martin
>>>>>
>>>>> Thanks,
>>>>> michal
>>>>>
>>>>>>
>>>>>> Thanks.
>>>>>> _______________________________________________
>>>>>> Users mailing list
>>>>>> Users at ovirt.org
>>>>>> http://lists.ovirt.org/mailman/listinfo/users
>>>>>>
>>>>>>
>>>>>
>