[Pacemaker] small gfs question

Andrew Beekhof andrew at beekhof.net
Fri Jun 25 15:13:28 UTC 2010


On Thu, Jun 24, 2010 at 11:49 AM, Robert Lindgren
<robert.lindgren at gmail.com> wrote:
>
>
> On Thu, Jun 24, 2010 at 11:45 AM, Andrew Beekhof <andrew at beekhof.net> wrote:
>>
>> On Thu, Jun 24, 2010 at 10:35 AM, Robert Lindgren
>> <robert.lindgren at gmail.com> wrote:
>> >
>> >
>> > On Thu, Jun 24, 2010 at 9:58 AM, Andrew Beekhof <andrew at beekhof.net>
>> > wrote:
>> >>
>> >> On Thu, Jun 24, 2010 at 9:47 AM, Robert Lindgren
>> >> <robert.lindgren at gmail.com> wrote:
>> >> > Thanks for the hint Andrew,
>> >> >
>> >> > But I feel that I might have missed something with pacemaker config.
>> >> > I
>> >> > see
>> >> > that dlm_controld.pcmk,gfs_controld.pcmk are started OK, which I
>> >> > guess
>> >> > are
>> >> > doing the maintenance of the GFS things from the cluster side. Are
>> >> > there
>> >> > any
>> >> > special params needed for dlm_controld and gfs_controld?
>> >>
>> >> Nope.  The agents set any required options automatically.
>> >> What versions are you running and which distro?
>> >
>> > I'm running on ubuntu lucid, with lucid-cluster ppa gfs stuff.
>> > pacemaker 1.0.8+hg15494-2ubuntu2
>> > corosync 1.2.0-0ubuntu1
>> > drbd8-source 2:8.3.7-1ubuntu2.1
>> > gfs2-pacemaker 3.0.7-0ubuntu0ppa2.2
>> > gfs2-tools 3.0.7-0ubuntu0ppa2.2
>> > libdlm3-pacemaker 3.0.7-0ubuntu0ppa2.2
>> > I'm not really sure but when the "other" node dies from power failure I
>> > see
>> > that gfs looses it's connection to the node (obviously), then I read
>> > something about fencing in linux-cluster, which unlooks gfs. And I'm not
>> > sure what it does and what part of the pacemaker/openais/corosync/drbd
>> > should do this unlooking?
>>
>> You do have stonith configured right?
>
> No :)

Ah, that explains it then

> right now (during test) I don't have hardware with stonith devices,
> like drac5 or something. Is it possible to configure stonith with for
> example external/ssh and make it work?

Well external/ssh isnt going to work if there's no network access to
the "bad" node...




More information about the Pacemaker mailing list