Proxmox VE 5 iscsi + LVM

J

juniper

Renowned Member
Oct 21, 2013
84
0
71
Hi,

i use on my 3 node cluster iscsci storage + LVM (shared)

it works only if i change lvm.conf with

locking_type = 3

is correct?
 
I'm having the same problem. Shared LVM nodes fall out of sync with each other using the standard LVM settings. Going to turn this on and see what the results are like.

I can't commit to continue testing Proxmox unless the cluster is synchronized. This seems like a pretty bad error, since it definitely has resulted in data loss for me.

On a side note, make sure you disable lvmetad if you do this. Set use_lvmetad = 0 in the config, per the LVM documentation.
 
Last edited:
Yes, i change also use_lvmetad = 0 and i update initramfs with update-initramfs -u

in this way all seems working but i think that isn't a solution...
 
Agreed. I've read back through some of the historical messages about this now, and it appears the Proxmox guys wrote their own locking mechanism that's compatible with locking_type = 1.

Unfortunately, something else appears to be a problem. I can definitively state, at this point, that my very simple LVM structure gets out of sync in as little as 6 hours with no iSCSI errors on the single iSCSI lun. My storage:

Code: 
dir: local
        path /var/lib/vz
        content backup,iso,vztmpl

lvmthin: local-lvm
        thinpool data
        vgname pve
        content images,rootdir

nfs: GrimlockISO1
        export /xfer
        path /mnt/pve/GrimlockISO1
        server 192.168.210.90
        content vztmpl,backup,iso
        maxfiles 1
        options vers=3

iscsi: ProxmoxVM
        portal 192.168.210.250
        target iqn.2001-05.com.equallogic:0-1cb196-59e8a545b-87600527581596bb-proxmoxvm
        content none

lvm: lvm001
        vgname vg001
        base ProxmoxVM:0.0.0.scsi-36019cbb145a5e859bb96155827056087
        content images,rootdir
        shared 1

I have 6 hosts connected to my cluster. lvm001 falls out of sync across all of the hosts. This leads to inconsistencies in the file naming, etc. This *seems* to happen more often when I try to clone a VM. I haven't tested straight up regular installations to see if I can stabilize it, but honestly if you're running a virtual infrastructure, templating is a critical component of that.

Maybe it's a 5.0 stability problem?
 
You must log in or register to reply here.
Top Bottom