[Pacemaker] RHEL/CentOS 6.4: corosync -> CMAN migration
Andrew Beekhof
andrew at beekhof.net
Mon Mar 18 00:46:51 UTC 2013
On Sun, Mar 17, 2013 at 8:57 PM, Stefan Förster
<cite+pacemaker at incertum.net> wrote:
> Hello Andrew,
>
> * Andrew Beekhof <andrew at beekhof.net>:
>> On Sun, Mar 10, 2013 at 8:35 PM, Stefan Förster
>> > 1. In step (2), will pacemaker stop resources when it's brought down,
>> > despite being in maintenance mode?
>>
>> No.
>> _However_, you will want to stop any cluster filesystems as they're
>> going die anyway.
>> Both OCFS2 and GFS2 require the dlm which requires corosync/cman.
>> Without those running, the filesystems can't function and will panic.
>
> Since we don't use anything that needs global locking we will be fine.
>
>> > 2. In step (5), will the pacemaker resource configuration be
>> > preserved throughout this migration?
>>
>> Absolutely. The only thing that changes is corosync.conf
>> All the resource related configuration stays in cib.xml and is unchanged.
>
> This is something I totally forgot - I should probably take not only
> backups of "crm configure show" and the cluster's XML config
> ("cibadmin -Q" or something like that, I alway forget), but also
> backup cib.xml - thanks for reminding me.
>
> If anyone else has to migrate from the corosync plugin to CMAN on
> RHEL/CentOS 6.4: According to the extensive testing I performed this
> week, it's basically very painfree. The important things to remember,
> IMHO:
>
> 1. Migrate _all_ of your corosync networking configuration, i.e:
> - secondary communication rings, DLM protocol SCTP, if used
> - port and multicast address settings (_especially_ if your
> network guys are good and give higher priority to your cluster
> communication)
> - encryption key
> - stuff that I forgot and that will bite me
> 2. Check startup configuration of "cman" _and_ "pacemaker". Some
> people like to start their clusters at reboot, in this case, both
> services need to be enabled using "chkconfig" (I don't really
> recommend this, but whatever).
Agreed.
> 3. Make sure to redirect CMAN fencing to pacemaker!
> 4. If you had non-default vote settings, migrate them!
> 5. If you onle have two nodes, set two_node="1" and
> expected_votes="1".
> 6. Migrate the logging section.
>
> Thanks for your input on this, guys.
Glad to hear it went well.
Its sounds much scarier than it really is :)
More information about the Pacemaker
mailing list