[Pkg-gridengine-devel] pkg-gridengine first steps

[Michael Banck]

On Tue, Apr 24, 2007 at 11:17:05AM +0200, Andreas Beckmann wrote:
> > On Tue, 24, Apr, 2007 at 12:03:37AM +0200, Michael Banck spoke thus..
> >> There's also debian/TODO which lists the (IMHO) missing stuff, feel free
> >> to add to it.
> For now, I wouldn't move around stuff in /var/spool/gridengine as this
> might break upstream's setup suggestion of sharing SGE_ROOT between all
> hosts. Although we won't recommend (and support) this practise, we
> shouldn't make it harder to do this.

Hrm.  There's three things in SPOOL_DIR/SGE_ROOT I guess which don't
really belong there:

 * The initial configuration ($SGE_ROOT/default/common/bootstrap)
 * The log files ($SPOOL_DIR/{qmaster,execd,schedd}/messages)
 * The pid files ($SPOOL_DIR/{qmaster,execd,schedd}/sge_{qmaster,execd,schedd}.pid)

Actually, right now I'm keeping $SGE_ROOT at /var/lib/gridengine and
$SPOOL_DIR at /var/spol/gridengine.  As the servers chdir to their
respective spool directory on startup, $SGE_ROOT is mostly unused (most
notably contains the act_qmaster file), so maybe moving $SGE_ROOT over
to /var/spool/gridengine might make sense in above shared $SGE_ROOT
context.

I really think the log files should be accessible under
/var/log/gridengine/ or so as well at least; maybe we could have
symlinks from /var/spool/gridengine/{qmaster,execd,schedd}.log pointing
to /var/spool/gridengine/{qmaster,execd,schedd}/messages.

The pid files aren't that much of a problem, they are generated on
startup by the servers but only used by the rc scripts on stop, we can
just put them anywhere and I'm not sure whether they are critical to a
shared $SGE_ROOT.  I'm fine with keeping them under $SPOOL_DIR for now,
but it looks a bit ugly.

In the end, I think people who want to share SGE_ROOT via NFS will
likely just download the upstream release and run their installation
scripts.

What we should check is what needs to be done for -client to find its
qmaster.  AFAICT, they checks for $SGE_ROOT, then read
$SGE_ROOT/default/common/bootstrap to figure out something (what
exactly?) and then read $SGE_ROOT/default/common/act_qmaster to get the
hostname of the master (they also get the port via SGE_QMASTER_PORT or
/etc/services, but we have this covered now).

We could either patch their startup routines to (alternatively) look for
something like /etc/gridengine/act_qmaster, make /var/lib/gridengine a
link to /etc/gridengine/act_qmaster, or maybe even ask the admin via
debconf about the qmaster hostname.  Thoughts?


Michael
Michael