libvirt users - Aug 2016 - Re: Mounting VM filesystem on host while VM running

hello Michal,

That was my worries as well, and I think that your advice is probably 
the right one. Or even, I'm might request an enhancement to the collect 
libvirt plugin to report also FS information.

Kind regards
patrick

On , Michal Privoznik wrote:
> On 08.08.2016 09:33, Patrick PICHON wrote:
>> Hello,
>> 
>> 
>> All my VMs are using LVs created on the host side.
>> 
>> I'm using collectd to monitor some of the ressources of my host as 
>> well
>> as the libvirt pluging to monitor my VMs.
>> 
>> Collectd has an interesting plugin ( df ) which can monitor the
>> filesystem usage.
>> 
>> I would like to use it to monitor the VMs filesytsem usage.
>> 
>> As it is obvious that if I mount the VM filesystem on the host while 
>> the
>> VM is running, I'm wondering if I could mount it in READONLY
without
>> expecting any panic/corruption problem.
>> 
>> 
>> Any advices ?
> 
> I don't think it is a good idea. Mostly because of caching - you
> probably won't get consistent results. Moreover, some filesystems (e.g.
> ext4) replay journal even if mounted as RO (frankly, I haven't found a
> way how to mount ext really RO). I don't know of any filesystem that
> would support being mounted multiple times from two different hosts
> (except network FS obviously).
> 
> What you could do is to write small guest agent that will report disk
> usage to host (e.g. via virito serial console).
> 
> Michal

On 08.08.2016 10:23, Patrick PICHON wrote:
> hello Michal,
> 
> That was my worries as well, and I think that your advice is probably
> the right one. Or even, I'm might request an enhancement to the collect
> libvirt plugin to report also FS information.
What I'd like to see is libvirt's ability to spawn commands inside
guest. But it is not that simple. I mean, the hardest part is to design
public libvirt API. For instance, we'd have to deal with both blocking &
non-blocking (i.e. long-running) commands; then we would need to give
users possibility to read command's output (possibly asynchronously).

With this, you'd just: virDomainRunCommand(dom, 'df', '-h',
NULL); or
something and you'd be good.

Michal

On Mon, 2016-08-08 at 10:37 +0200, Michal Privoznik
wrote:
> On 08.08.2016 10:23, Patrick PICHON wrote:
> > 
> > hello Michal,
> > 
> > That was my worries as well, and I think that your advice is probably
> > the right one. Or even, I'm might request an enhancement to the
collect
> > libvirt plugin to report also FS information.
> 
> What I'd like to see is libvirt's ability to spawn commands inside
> guest. But it is not that simple. I mean, the hardest part is to design
> public libvirt API. For instance, we'd have to deal with both blocking
&
> non-blocking (i.e. long-running) commands; then we would need to give
> users possibility to read command's output (possibly asynchronously).
> 
> With this, you'd just: virDomainRunCommand(dom, 'df',
'-h', NULL); or
> something and you'd be good.
Not as straightforward as you'd like, but if you use QEMU you
can already achieve something similar via qemu-guest-agent.

https://www.redhat.com/archives/libvir-list/2016-August/msg00339.html

-- 
Andrea Bolognani / Red Hat / Virtualization