On 21/01/2021 21:31, n...@li.nux.ro wrote:
Well, there you go..
On 2021-01-21 18:50, Simon Weller wrote:
We used to use CLVM a while ago before we shifted to Ceph. Cluster
suite/corosync was a bit of a nightmare, and fencing events caused all
sorts of locking (DLM) problems.
I helped a CloudStack user out a couple of month ago, after they
upgraded and CLVM broke, so I know it's still out there in limited
places.
I wouldn't recommend using it today unless you're very brave and have
the capability of troubleshooting the code yourself.
In addition:
I assume you used CLVM with Corosync?
Yes
My concern with LVM is:
- No thin provisioning (when used with CloudStack)
Indeed and machine deployment meant a qemu-img convert qcow2 -> lv .. so
lengthy.
Yes, that's a valid argument against using LVM as you loose a lot of
features.
- No snapshots (Right?)
Don't remember honestly. If there were, they must have been slow.
- Not very much used
Yep..
OCFS starts to become more appealing. :)
At the time - and probably now as well - OCFS was best supported by
Oracle Unbreakable Linux (RHEL rebuild), might be worth looking at
running that instead of Ubuntu or CentOS, hopefully a smoother, bug-free
experience.
I would need to test that. RHEL6 has been a long time ago and we are now
at CentOS 8.
In our case we use Ubuntu 20.04 for hypervisors and that means a lot of
development went into OCFS2 in recent years.
There's also a reason that VMWare uses VMFS with VMDK images on top as
that provides a lot of flexibility.
This time the use-case is iSCSI, but thinking ahead we'll get new things
like NVMeOF which provides even lower latency.
Probably best to set up a PoC and see how it works out.
Wido
Lucian
