LVM thin over iSCSI and live migration

[TomTomGo]

Hi,

I've just configured a new thin LVM-Thin datastore over iSCSI like this :

- Datacenter > Storage > Add > iSCSI



Then from CLI :

pvcreate /dev/disk/by-id/scsi-36e843b6e6494e96d215cd4e37d9747d0
vgcreate lun1 /dev/disk/by-id/scsi-36e843b6e6494e96d215cd4e37d9747d0
lvcreate -L 1024g -T lun1/thinlun1

And finally adding a new LVM-Thin from GUI :



Then i've created a VM on this new storage : OK
Snapshots works fine (which was the main goal) but live migration fails :



Any ideas ?

Regards,

 

[TomTomGo]

Cold migration doesn't work no more :



Seems that lvmthin storage type is not supported ... Any reasons ?

Regards,

 

[spirit]

lvmthin cannot be mount on 2 servers at the same time (because of metadatas), so it's local only

you need to use classic lvm.

 

[TomTomGo]

lvmthin cannot be mount on 2 servers at the same time (because of metadatas), so it's local only

you need to use classic lvm.

Thanks for your answer !
What a pity ...
It would be a must when using a three nodes cluster with an enterprise class iSCSI array :

[ENTERPRISE CLASS STORAGE ARRAY] <=> [ISCSI LUNs] <=> [THINLVM] <=> [PROXMOX NODES]

This way we could have a "strong" LVM based vms disks allowing snapshots, which IMHO is a lack on Proxmox today.
In "cluster mode", Snapshots are only available with QCOW2 on NFS shared storage (bad performances), CEPH (not very easy to setup and maintain) or specific hardware / software (Omnios, Nexenta, ...)

Regards,

 

[spirit]

The only way to manage snapshot through issi, is to do it on the iscsi array. If your enterprise array have apis to do snapshot,it could be possible to create a plugin in proxmox. (I'm doing it for netapp for example)

 

[TomTomGo]

The only way to manage snapshot through issi, is to do it on the iscsi array. If your enterprise array have apis to do snapshot,it could be possible to create a plugin in proxmox. (I'm doing it for netapp for example)

Yes, sure ....
Or maybe, to have something more generic, implementing LVM snaphots on "classic" LVM iSCSI volumes like this :

Create snapshot : lvcreate -L SNAP_SIZE -s -n SNAP-LV1 /dev/VG1/LV1
Remove snapshot : lvremove /dev/VG1/SNAP-LV1 -f
Revert snapshot : lvconvert --merge /dev/VG1/LV1

Regards,

 

[spirit]

Yes, sure ....
Or maybe, to have something more generic, implementing LVM snaphots on "classic" LVM iSCSI volumes like this :

Create snapshot : lvcreate -L SNAP_SIZE -s -n SNAP-LV1 /dev/VG1/LV1
Remove snapshot : lvremove /dev/VG1/SNAP-LV1 -f
Revert snapshot : lvconvert --merge /dev/VG1/LV1

Regards,

classic lvm snapshot are too slow. it's really slowdown things. (It's just ok to take a snapshot, do a backup, delete snap for example)

 

[TomTomGo]

classic lvm snapshot are too slow. it's really slowdown things. (It's just ok to take a snapshot, do a backup, delete snap for example)

You mean that "lvconvert --merge" will be too slow that's it ?
It's a pity ...

BTW, does your Netapp storage plugin will be merged in the main Proxmox branch ?
It should be interesting for our future SAN choices ....

Regards

 

[spirit]

You mean that "lvconvert --merge" will be too slow that's it ?
It's a pity ...

I mean, If you have created a snasphot on classic lvm, the write/read performance will be 2 or 4x slower.
(and for each new snapshot, performance decrease).
This is because they are no metadata, so lvm need to read each snasphot to find the good version of the block.

BTW, does your Netapp storage plugin will be merged in the main Proxmox branch ?
It should be interesting for our future SAN choices ....

Not in the main branch (because proxmox team don't have netapp hardware to maintain it),
but I maintain a github
https://github.com/odiso/proxmox-pve-storage-netapp

(I'll update it soon to last proxmox 4.2)