lvm snapshots on shared storage

[hk@]

Hi
I'd like to know if someone already tested this with PVE 2.0 RC1:
assuming two nodes share one iSCSI target or drbd-blockdevice where several lvm volumes reside and vzdump processes backups in parallel using lvm snapshots from both nodes - is this (I remember in 1.9 it is) still a problem or have there been some new features introduced to solve this?

If it is still a potential issue, please advise on how to set something like this up (eg in a two node HA cluster or similar).

thank you in advance
hk

 

[e100]

I use DRBD, it is my opinion that making snapshot backups from two nodes on the same LVM volume at the same time is a bad idea.

All of my DRBD nodes have two DRBD volumes all the VMs on node A use DRBD0 and VMS on node B use DRBD1.
This has worked well on 1.X and 2.0 for me.

I have had issues on 1.X with occasional split-brain when making backups on nodeA DRBD0 and nodeB DRBD1 at the same time.
This has been discussed in 1.X forum in great detail in the past.

On 1.9 we have observed that increasing CPU usage during the backup process (by using more cores in pigz) has resulted in more frequent split-brain during backups.
So my theory today is lack of CPU resources may play a role in triggering a split-brain during backups.
With lzo in 2.0, this will likely be much less of a problem.

I have not had a split-brain in 2.0 during a backup.
I also have not backed up the same volumes of data we backup on 1.9 and my 2.0 servers have faster CPUs.
So I can not conclude that the problems on 1.9 do not exist in 2.0 all I can say is that I have not seen them so far.

LVM is handled better in 2.0.
For one, volumes are not activated on a node unless they are needed.