[Pacemaker] how to use dlm and clvm with pacemaker ?

[denis bonnenfant]

Hello,

I'm currently running in production a 3-nodes 
corosync/pacemaker/drbd/clvm xen virtualisation cluster, based on debian 
wheezy.

Two nodes are running drbd inactive/active setup, with clustered VGs 
(/dev/xen-data) on top of drbd disk for local domUs , and ISCSI target  
LV ( /dev/xen-data/iscsi-target)  for the third node with a virtual ip 
for failover.

On both nodes another clustered VG is running on the iscsi initiator 
(/dev/xen-iscsi).

LV-based DomUs can live migrate between the 2 drbd nodes if they are on 
drbd VG,  or between all the nodes if they are on the iscsi VG.

This setup works quite well, except for one point :  sometimes, when a 
node falls down, dlm is locking down all the lvm layer, causing all 
nodes to fence each other. The problem is not related with the stacked 
lvm setup, as it occurs even when no iscsi resources are started.

This problem is well known, and new versions of cman totally remove 
dlm_controld.pcmk. But then documentations are quite confusing :

- 
http://www.clusterlabs.org/doc/en-US/Pacemaker/1.1/html/Clusters_from_Scratch/apcs02.html 
advises to run cman from init scripts with a configured 
/etc/cluster.conf, and then start pacemaker  from init,
- 
http://www.clusterlabs.org/doc/en-US/Pacemaker/1.1/html/Clusters_from_Scratch/apcs02s02.html 
advises to run corosync with a cman plugin, but don't say how to start 
pacemaker?
- some other advise to modify controld RA to use directly new 
dlm_controld (how ? symlink dlm_controld.pcmk ?)

Before bringing my cluster down, I want to be sure about the best path 
for upgrading.


As I'm running redundant rings in corosync, adding cman stuff in my 
corosync.conf  seems for me the simplest, but in this case, do I still 
need to configure controld and clvm in crm, or are they completely 
managed by the cman plugin ? do I need to start pacemaker from corosync 
config too ?

http://theclusterguy.clusterlabs.org/post/907043024/introducing-the-pacemaker-master-control-process-for 
seems to suggest that in my case, with dlm and clvm involved, option 3 
with cluster.conf is the best approach. But how pacemaker is started in 
this case,  from init  after cman ?
Do I need up-to date version of cman, or is debian wheezy  version 
(3.0.12) ok ?
Do I need up-to date version of pacemaker, or is debian wheezy version 
(1.1.7) ?, recompiled with --with-cman ?

Regards,

Denis