[PVE-User] New Proxmox cluster

[Lutz Willek]

Hey Edgardo,

Am 06.06.2016 um 11:32 schrieb Edgardo Ghibaudo:
> Il 06/06/2016 11:00, Lutz Willek ha scritto:
>> Relative to the original post: I wonder if it is possible simply to
>> use NFS instead of iSCSI? This would be easy-peasy to implement, has
>> cluster capability (in proxmox terms) and qcow2 (i.e snapshots are
>> possible) could be used in addition to raw image-files.

>>>> Storing images on local (or any other not shared) storage will break
>>>> live-migration and HA, so it's better to store all VM images on shared
>>>> storage, if you can.

>>> Following the document http://pve.proxmox.com/wiki/Storage_Model, in the
>>> _LVM Groups with Network Backing section_:
>>> after the definition of the Volume Group Name on the LUN, Proxmox see in
>>> the Storage of the Datacenter the Volume Group.

>>> How can create different Logical Volumes using EXT3 or EXT4 file systems
>>> in the Proxmox Volume Group ?

>> You may be able to do this, but doing so is useless, because ext3/4 is
>> not cluster-aware. (i.e. can not be used in a "clustered proxmox
>> setup" on different nodes at the same time)

> Following your experience, what kind of file system do you suggest for
> the iSCSI LUN of my cluster ?
> When I prepare the node of the cluster installing Debian Jessie, what
> kind of file system do you suggest for each node ?

As written before,

If you want to use isci: setup a LVM vg (without any filesystem!) ontop 
of iscsi. Then use lv's inside the the vg for your virtual machines in a 
clustered proxmox setup.

Or, if you want also (easy) use snapshots/backups, do not use iscsi/lvm, 
instead setup a nfs-server. Then use nfs for your virtual machines in a 
clustered proxmox setup.

Or follow Dmitry's suggests and setup a clustered filesystem ontop of 
your iscsi LUN. Maybe this is also possible with zfs, dunno.

Remember: It is NOT possible to use a regular filesystem like ext{2,3,4} 
or xfs ontop of a *shared* LUN, because these kinds of regular 
filesystems are not cluster-aware. (Bad things will be happen if you try 
to access files from more than one node at the same time)


Freundliche Grüße / Best Regards

          Lutz Willek

-- 
________________________________creating IT solutions
Lutz Willek                     science + computing ag
Senior Systems Engineer         Geschäftsstelle Berlin
IT Services Berlin              Friedrichstraße 187
phone +49(0)30 2007697-21       10117 Berlin, Germany
fax   +49(0)30 2007697-11       www.science-computing.de