You are viewing a plain text version of this content. The canonical link for it is here.
Posted to dev@cloudstack.apache.org by Clement Mondon_navazo <cl...@inria.fr> on 2013/06/24 10:16:32 UTC

Libvirt storage pool problem with a closed NFS Server

Hello, 


(First, sorry if I am posting on bad list) 


We are trying to use a Dell DR400 NFS server as secondary storage in our Cloudstack plateform. All hosts run on CentOS using KVM. 


Our problem is CS wants to mount directly the subdirectory of the given template to obtain /mnt/«template_id». 


For instance, CS does the equivalent of : 


mount 10.0.0.1:/containers/testosx/template/tmpl/1/3/ /mnt/UID 


But our NFS server doesn't allow us to mount a subdirectory of an export. We tried to understand in which files the problem is but without success (probably in the libvirtd storage process). 


We have the same problem with CS 3.0.2 and with current Master. 


So we don't understand why CS have this behaviour. 


Does one mount per template is efficient ? Furthermore, if we run our tests with VSphere on the same NFS server, they are successful. Indeed, CS uses one datastore (the export's root) and then navigate on it to find the good directory. 


Why does KVM not behave like Vsphere ? 


Does anybody have clue, explanation or documentation let us know. 


Looking forward to hearing from you, 


Clément Mondon

Re: Libvirt storage pool problem with a closed NFS Server

Posted by Daan Hoogland <da...@gmail.com>.
Clement,

Cs does not try to mount secondary storage as a whole on every host. I'd
say you need to look at making your nfs deamon allow the mount.

Regards,
 Op 24 jun. 2013 16:20 schreef "Clement Mondon_navazo" <
clement.mondon_navazo@inria.fr> het volgende:

>
> Hello,
>
>
> (First, sorry if I am posting on bad list)
>
>
> We are trying to use a Dell DR400 NFS server as secondary storage in our
> Cloudstack plateform. All hosts run on CentOS using KVM.
>
>
> Our problem is CS wants to mount directly the subdirectory of the given
> template to obtain /mnt/«template_id».
>
>
> For instance, CS does the equivalent of :
>
>
> mount 10.0.0.1:/containers/testosx/template/tmpl/1/3/ /mnt/UID
>
>
> But our NFS server doesn't allow us to mount a subdirectory of an export.
> We tried to understand in which files the problem is but without success
> (probably in the libvirtd storage process).
>
>
> We have the same problem with CS 3.0.2 and with current Master.
>
>
> So we don't understand why CS have this behaviour.
>
>
> Does one mount per template is efficient ? Furthermore, if we run our
> tests with VSphere on the same NFS server, they are successful. Indeed, CS
> uses one datastore (the export's root) and then navigate on it to find the
> good directory.
>
>
> Why does KVM not behave like Vsphere ?
>
>
> Does anybody have clue, explanation or documentation let us know.
>
>
> Looking forward to hearing from you,
>
>
> Clément Mondon