[Pacemaker] benefits of cman?

[Andrew Beekhof]

On Wed, May 23, 2012 at 5:37 AM, Matthew O'Connor <matt at ecsorl.com> wrote:
>
> On 05/22/2012 03:30 AM, Andrew Beekhof wrote:
>
> We were talking about GFS2 and Pacemaker but the same applies to OCFS2.
> If you're just using ocfs2 there is no need for cman.  But if you want
> ocfs2 /and/ a cluster manager - you want them all using the same
> membership and quorum data.
>
> Yes, I remember reading that.  In my situation, I am torn between what works
> now and what "will" work tomorrow.  My current choice of distro has some
> interesting...irregularities, shall we say?  Not to throw myself into the
> middle of a "get a real distro" holy war, but:  one release has working
> Corosync/Pacemaker but doesn't seem to support cman integration (even though
> it carries the packages for it), while a later release supports cman and
> (apparently) not corosync/pacemaker alone,

cman is "just" a corosync plugin.  so for basic cluster, if cman is
supported so is plain corosync.
You "just" shouldn't use it with a cluster filesystem.

> but pacemaker dies a horrible
> death when I put nodes into standby (not necessarily cman-related, I
> realize).

Um, that shouldn't happen.  Did you file a bug for that?

> Part of me wants to just build the whole stack from the sources
> and get exactly what I want, while other part wants to save myself the
> angst, time, and human-error of doing such a thing, and make the provided
> packages work.
>
> I am probably trying too hard to future-proof, and either need to stick with
> what works now, switch distributions or work harder to solve the outstanding
> problems of the latest distro release's cluster packages.
>
> Anyway, thanks for the info - it is most happily received!
> -- Matthew
>