[Users] Memory usage

[Doron Fediuck]

----- Original Message -----
> From: "René Koch" <rkoch at linuxland.at>
> To: "Doron Fediuck" <dfediuck at redhat.com>
> Cc: "Koen Vanoppen" <vanoppen.koen at gmail.com>, users at ovirt.org, "Martin Sivak" <msivak at redhat.com>
> Sent: Wednesday, February 12, 2014 7:31:47 PM
> Subject: Re: [Users] Memory usage
> 
> On Wed, 2014-02-12 at 11:22 -0500, Doron Fediuck wrote:
> > 
> > ----- Original Message -----
> > > From: "René Koch" <rkoch at linuxland.at>
> > > To: "Koen Vanoppen" <vanoppen.koen at gmail.com>
> > > Cc: users at ovirt.org
> > > Sent: Wednesday, February 12, 2014 4:18:37 PM
> > > Subject: Re: [Users] Memory usage
> > > 
> > > On Wed, 2014-02-12 at 15:14 +0100, Koen Vanoppen wrote:
> > > > In The GUI, it says it's using 25% of the memory.
> > > 
> > > 
> > > I guess it's the real value, right?
> > > The same happened for the memcached vm, someone reported to me -
> > > negative value in REST-API, but correct graph in oVirt webadmin GUI.
> > > 
> > > I fear I have no idea how this can happen - so maybe someone else can
> > > help you troubleshoot this issue.
> > > > 
> > > > 
> > > > 
> > > > 2014-02-12 15:10 GMT+01:00 Koen Vanoppen <vanoppen.koen at gmail.com>:
> > > >         Thanks for the quick respons, but there is no memcached
> > > >         running on that VM.
> > > >         
> > > >         
> > > >         Kind regards
> > > >         
> > > >         
> > > >         
> > > >         2014-02-12 15:06 GMT+01:00 René Koch <rkoch at linuxland.at>:
> > > >         
> > > >                 On Wed, 2014-02-12 at 14:55 +0100, Koen Vanoppen
> > > >                 wrote:
> > > >                 > Dear all,
> > > >                 >
> > > >                 >
> > > >                 > When we monitor one of our machines, we noticed that
> > > >                 there was one vm
> > > >                 > that was constantly giving a error of memory usage.
> > > >                 But when we took a
> > > >                 > look at it, there is actually nothing wrong with it.
> > > >                 Now we looked
> > > >                 > furhter then that. We looked at the API of the
> > > >                 machine and noticed
> > > >                 > something very strange:
> > > >                 >
> > > >                 > <statistic
> > > >                 >
> > > >                 href="/api/vms/3b9aa245-75ff-42e8-b921-1c9ce61826bf/statistics/b7499508-c1c3-32f0-8174-c1783e57bb08"
> > > >                 id="b7499508-c1c3-32f0-8174-c1783e57bb08"><name>memory.used</name><description>Memory
> > > >                 used (agent)</description><values
> > > >                 type="INTEGER"><value><datum>-944892806</datum></value></values><type>GAUGE</type><unit>BYTES</unit>
> > > >                 >
> > > >                 >
> > > >                 > It's a negative...
> > > >                 
> > > >                 
> > > >                 
> > > >                 Do you have memcached running in this vm?
> > > >                 
> > > >                 I heard about this issue with memcached, but never
> > > >                 tested memcached in
> > > >                 my oVirt environment. You get the real usage value
> > > >                 with
> > > >                 memory.used = memory.installed + memory.used
> > > >                 
> > > >                 
> > > >                 Regards,
> > > >                 René
> > > >                 
> > > >                 >
> > > >                 >
> > > >                 > What could be the problem?
> > > >                 >
> > > >                 >
> > > >                 > Kind regards,
> > > >                 >
> > > >                 > koen
> > > >                 >
> > > >                 
> > 
> > Guys,
> > these values are usually a result of overcommitment mechanism usage.
> > For example, if KSM is effective, it will free a lot of memory pages,
> > and total-free-committed becomes negative.
> 
> 
> Thanks a lot for the information.
> 
> 
> > 
> > This was reported in https://bugzilla.redhat.com/show_bug.cgi?id=977758
> > and the engine is using memFree reported by vdsm, which is more accurate.
> > 
> > The API reports the old version due to backwards compatibility.
> 
> 
> I just had a look at the bugzilla report and the RHEV documentation
> which says "Current memory in bytes used by the virtual machine.".
> So this means, the reported values are totally useless for monitoring
> memory usage of a virtual machine if KSM is active. I would expect to
> get the memory usage of a virtual machine and not how much memory is
> consumed on the hypervisor (this is pretty useless information for me).
> 
> Is it planned to report the memory usage in a virtual machine in the API
> as well?
> 
> 
> Regards,
> René
> 
> 

René,
started digging into this, but it may take some time.

Note that there are 2 memory usage reports; one for the host,
and another for each VM. 
My response is related to the host, while the original question was
for the VM. I still think the root cause is the same (ie- related to
overcommitment; for example the VM may have a balloon inflated), but
would like to properly check and fix if needed.

In order to make sure we keep track of it, please open a bug with all the
relevant info. 

Thanks and keep reporting!
Doron