[Pacemaker] Now HOWTO: Dual primary DRBD + OCFS2

Diego Remolina diego.remolina at physics.gatech.edu
Mon Sep 14 12:11:41 UTC 2009


One thing that I believe would really complement the howto is if you 
could explain the configuration for a more complex setup, say two 
separate drbd Primary devices on the same cluster.

Going from one to two drbd devices is not always obvious and there are 
some location, collocation and order constrains that are not always easy 
to figure out.

There are also gotchas depending on how you configure your constrains as 
if you wanted to say bring down one of the filesystems on drbd or only 
one of the resources, but not the other, then depending on how your 
constrains were configured, bringing down only one of the two, can 
actually bring down other services in the cluster.

If you would like any clarification, please let me know.

Diego


Michael Schwartzkopff wrote:
> Am Montag, 14. September 2009 09:15:36 schrieb Florian Haas:
>> On 2009-09-12 21:10, Michael Schwartzkopff wrote:
>>> Hi,
>>>
>>> I tried to write a new HOWTO about a setup of a dual primary DRBD and the
>>> Cluster filesystem OCFS2 on top. You find it at:
>>> http://www.clusterlabs.org/wiki/Dual_Primary_DRBD_%2B_OCFS2
>>>
>>> Please note that it is only a first try. So I would be glad about any
>>> improvements. Please mail me your suggenstions. Thanks!
>> Here we go:
>>> Please note that the monitor-Operation of the primitive resource is not
>>> only nice to have but essential to make the setup work. Otherwise the
>>> second secondary resource will not get promoted after a failure.
>> That is not true, monitor is not essential (although highly
>> recommended). Enabling notify however *is* essential, and the RA will
>> bail out with $OCF_ERR_CONFIGURED if notifications are not enabled.
> 
> See my thread: "Strange behaviour of dual master DRBD" for the behaviour of a 
> dual master DRBD resouce WITHOUT monitoring. I still do not have a satifying 
> answer, why pacmeaker does not promote the resource on a node after it got 
> online after a standy. With monitoring it works.
> 
>>> colocation colDLMDRBD inf: cloneDRBD msDRBD:Master
>> "cloneDLM"?
> 
> Oops, yes. Thanks.
> 
>>> # mkfs.ocfs2 -N 16 /dev/drbd0
>> 16 node slots for a filesystem that can only ever run on 2 nodes?
> 
> Copy and paste form the Novell Doc. Of course thes do not use DRBD. So I will 
> change.
> 
>>> primitive resFS ocf:heartbeat:Filesystem \
>>>
>>>     params device="/dev/drbd0" directory="/mnt" fstype="ocfs2" \
>>>
>>>     op monitor interval="120s"
>> Use device="/dev/drbd/by-res/<resourcename>" please, makes the config
>> much more transparent and easier to follow.
> 
> Ok
> 
> Thanks for all your comments!
> 




More information about the Pacemaker mailing list