What filesystem to use on shared storage

[kameleon]

I finally got my interfaces config file setup to where I want it and able to access my SAN now from both (soon to be 3) of my proxmox 2.0 machines. However before I continue I am wondering what filesystem should I use on the shared iSCSI storage to store my kvm images and such? I was looking at GlusterFS but everywhere I look says is should be avoided. What about cLVM? Any others I should look at? What are you guys using? Currently I only have the two machines connected but soon I will have up to 4 machines on this setup. I plan to have full HA failover working with the VM's disks being stored on the iSCSI SAN.

 

[udo]

Hi,
go for lvm-storage (also non filesystem) for kvm-images. Works fast and stable (ok, some people have trouble with snapshots/backup).

Instead of an raw-file on an file-system you have your VM-disk in a logical volume on the volumegroup.

Udo

 

[kameleon]

I guess I just have not read up enough on this but I thought I had to have a cluster-aware filesystem on the iSCSI lun in order to have more than one machine access it at the same time. Otherwise how will the secondary/failover machine pick it up?

 

[udo]

I guess I just have not read up enough on this but I thought I had to have a cluster-aware filesystem on the iSCSI lun in order to have more than one machine access it at the same time. Otherwise how will the secondary/failover machine pick it up?

Hi,
the trick is, that one lv on the vg was accessed only from one node - if you move the vm to the other node, this node will access the lv instead of the node before.
All nodes activate the volumegroup, but the logical volumes was only used by the node with run the guest.

Udo

 

[kameleon]

Hmmm. I think I am following you now. So I could theoretically have each server access separate lv's on a shared vg and once I fail one server the other should take ownership of the other lv's and just pick up where left off right?

 

[udo]

Hmmm. I think I am following you now. So I could theoretically have each server access separate lv's on a shared vg and once I fail one server the other should take ownership of the other lv's and just pick up where left off right?

Right,
the vm must do an filesystem-check if they started on the other node (after one node fail), but this is normal (it's like to pull the power cord on a real machine) and is not different to use raw-files on an cluster-fs.

This is also the reason, why it's nessesary to use fencing with HA-machines (it's absolutly important, that not two nodes write in one lv at the same time).

Udo

 

[sv63rus]

Hello! Please help ? How use OpenVZ and online/offline migration on shared LUN on SAN via SAS ?
https://forum.proxmox.com/threads/11473-File-system-type-on-shared-storage-(SAN-IBM-DS3512)