[Ocfs2-tools-devel] [Ocfs2-devel] Patch to Pacemaker hooks in ocfs2_controld

Andrew Beekhof beekhof at gmail.com
Sat Mar 7 03:56:12 PST 2009


On Sat, Mar 7, 2009 at 01:41, Joel Becker <Joel.Becker at oracle.com> wrote:
> On Fri, Mar 06, 2009 at 06:56:56PM +0100, Andrew Beekhof wrote:
>> On Fri, Mar 6, 2009 at 00:59, Joel Becker <Joel.Becker at oracle.com> wrote:
>> >        But that's logged in the ais logs, not in the ocfs2_controld
>> > output, right?  Is that in syslog?
>>
>> Its logged via syslog as ocfs2_controld - is there somewhere else that
>> ocfs2_controld logs to?
>
>        ocfs2_controld stores everything from log_debug() and
> log_error() in an internal ring buffer.  So you can do 'debugfs.ocfs2 -R
> "controld dump"' to see it.  Messages from log_error() are also passed
> to syslog, so that high-level issues can be seen in the logs.
>        So anything logged by the internal ais code is going to go to
> syslog but not to the ring buffer.  When I dump the ring buffer, I won't
> see it.  I don't think we need to find a way to get it into the ring
> buffer - I'd just like to see a log_debug() or two at those ais error
> points so that hte ring buffer shows where ais failed or something like
> that.

I think thats fair enough for the sign-in/nodeid parts, but for
process_pcmk() the message basically amounts to "we got a message"
which isn't likely to be useful.
To log more than that, you're basically talking about re-implementing
ais_dispatch() which defeats the point of having the hooks use the
same code-paths as all the other pacemaker daemons.

One thing I'd like to point out though, as a user/integrator though,
is that in practice the ring buffer is very rarely useful.
I like the idea of ring buffers for debugging, but in the case of
ocfs2_controld at the point you realize you want it's contents - the
process is gone (more often than not, the node with it too) and its
contents is lost.

The other thing you might want to consider is adding some granularity
between "this is an error" and "this is debugging that will never be
seen".
Basically support for the other syslog levels.



More information about the Ocfs2-tools-devel mailing list