[Linux-cluster] ccsd and mounting gfs errors

David Teigland teigland at redhat.com
Wed Oct 27 03:54:52 UTC 2004


On Tue, Oct 26, 2004 at 04:16:39PM -0400, linux-cluster-rhn at chaj.com wrote:
> 
> Hi, similar to the problem below, I'm getting ccsd errs in /v/l/messages and
> get a "Permission denied" error when trying to mount my gfs partition.
> 
> /p/c/nodes:
> Node  Votes Exp Sts  Name
>    1    1    1   M   nfstest2
>    2    1    1   M   nfstest1
> 
> /p/c/services
> Service          Name                              GID LID State     Code
> 
> are the same on both machines.
> 
> /v/l/messages simply shows:
> Oct 24 04:02:00 nfstest1 ccsd[1387]: Error while processing connect:
> Connection
> refused
> Oct 24 04:02:31 nfstest1 last message repeated 31 times
> Oct 24 04:03:32 nfstest1 last message repeated 61 times

This is a very common problem -- probably the most common.  It's caused by
fenced (which was started by running fence_tool join) trying and failing
to connect to ccsd.

What's wrong with ccsd?  Unfortunately, there's nearly no way to figure
that out without literally digging into the code, recompiling with
debugging on, etc.  (And even doing that hasn't always helped when I've
tried going down this path.)  The root problem may not even lie in ccs,
but could be anywhere in the layers of magma code.

So, I can't begin to guess what's wrong or provide any direction for
debugging.  The only suggestion I can make is to rebuild/install
everything from the cluster cvs.  Hopefully ccs will be improved to allow
at least a nominal diagnosis when things to wrong.  There's also an
outstanding request for a ccsd option to bypass all clustering/magma stuff
so you can at least bring the cluster and gfs up if the complex bits of
ccs aren't working.

-- 
Dave Teigland  <teigland at redhat.com>




More information about the Linux-cluster mailing list