[Samba] CTDB starting statd without -n gfs -H /etc/ctdb/statd-callout

[Chris Walker]

 Hello,

First and foremost, thanks *very* much for ctdb.  It's a joy to use
after banging around with other HA solutions.  We're planning to use
it to export Samba and NFS shares throughout campus.

I'm having one problem with the NFS part though.  When ctdbd first
starts statd (we're using CTDB_MANAGES_NFS=yes), it does so without
appending the stuff in the STATD_HOSTNAME variable in
/etc/sysconfig/nfs, which is where the statd-callout script is passed
to statd.  In our case, this means that statd is running as

rpc.statd -p 662 -o 2020

instead of

rpc.statd -n gfs -H /etc/ctdb/statd-callout -p 662 -o 2020

I could be wrong, but it looks to me that ctdb is using the nfslock
init script to start statd.  This script doesn't use $STATD_HOSTNAME
at all, so it follows that the statd-callout script isn't passed to
statd.

If I kill statd and let ctdb start the 60.nfs script restart it when
it monitors, then statd is run with the correct statd-callout script,
since 60.nfs does append the $STATD_HOSTNAME variable when rpc.statd
is invoked.  And the same is true if I change the nfslock init script
so that it appends the $STATD_HOSTNAME.

This is an up-to-date CentOS 5.5 OS, with CTDB pulled from the git
repository last week.


One quick unrelated question about CTDB -- the documentation states
that the CTDB_NODES IP addresses should live on a "private
non-routable subnet which is only used for internal cluster traffic".
This this a requirement?  I have our cluster nodes on one part of a
/24 (which is routable to our organization, but not to the internet),
and the CTDB_PUBLIC_ADDRESSES on another part.  This seems to be
working fine, but I wanted to check that I wasn't doing something that
would bite us later.

Thanks again for CTDB and Samba!

Best,
Chris