Re: [linux-lvm] ubuntu xenial + lvmlockd

[David Teigland <teigland@redhat.com>]

On Thu, Apr 20, 2017 at 11:26:18AM +0200, Charles Koprowski wrote:
> Would you consider the use of lvmlockd + sanlock as "production ready" ?

I believe it will work better than clvmd.  There is one main thing, unique
to using sanlock, that you should verify in your environment.
lvmlockd+sanlock is sensitive to spikes in i/o delays.  If sanlock sees
several consecutive large i/o delays (> 10 sec each), e.g. during heavy
use from applications, or during path switching, this can trigger spurious
failure detection.  (This is analogous to network delays when using
network-based solutions.)

(We can increase i/o timeouts to compensate if really necessary.)

> My goal here is to replace an existing cluster of 5 nodes using clvmd + dlm
> + corosync to access a shared VG of 6 TB containing around 300 LVs.
> 
> The current solution is working fine but I find using dlm + corosync "just"
> for locking a bit overkill.

I agree.

Dave