[torqueusers] new build system ready!

Garrick Staples garrick at usc.edu
Fri Feb 3 12:51:19 MST 2006

On Fri, Feb 03, 2006 at 01:23:56PM -0500, Troy Baer alleged:
> On Fri, 2006-02-03 at 11:35 -0500, Bisbal, Prentice wrote:
> > Do the config files go in /etc/ now, or are they still in the spool
> > directory? 
> I'd really like to make sure that is configurable, preferably with the
> historical behavior (config files in spool dir) as the default.  I'm
> having a hard time wrapping my brain around how you would draw the line
> between "config file" and "spool file", particularly for the DB files
> used by the pbs_server daemon.  The number of files that qualify as
> "config files" in my book is pretty small:


> # pbs_mom
> $PBS_HOME/pbs_environment
> $PBS_HOME/mom_priv/config

pbs_environment is used by pbs_mom, pbs_server, and pbs_sched (all
daemons.)  It's location can be changed at compile time with
--set-environ.  Oddly, the compile-time default is hardcoded to be the same
location as $PBS_SERVER_HOME, but does not follow run-time changes when
you redefine home with env or -d.

> # pbs_server
> $PBS_HOME/server_priv/nodes
> $PBS_HOME/torque.conf

server_priv/nodes is borderline "data" since it can be completed handled
through qmgr.  There's no need to ever directly edit it.

torque.conf is actually read by qsub too.  This file needs to go away
(at least, it's present form.)  The SERVERHOST, SUBMITHOSTS, and
ALLOWCOMPUTEHOSTSUBMIT need to be server attributes editable in qmgr.

> # general
> $PBS_HOME/server_name (Does torque.conf obviate the need for this on
> compute nodes and submit hosts?)

No.  server_name is used by clients and pbs_mom to find the name of the
server.  It's location can be changed at compile time with

torque.conf is used by pbs_server to set its own name.  IMHO, this is
broken behaviour.

> (Note that I use Moab and haven't looked at pbs_sched in years, so I
> have no idea what config files it requires.)

It varies by which pbs_sched you build, but it is similar to MOM.

> Not all of these need to live on every host either, and I'd like to make
> sure that how ever this is implemented, it a) doesn't break historical
> practice too badly and b) doesn't unintentionally make life more
> difficult for people with shared root file systems*.
> * Spoken as someone who has upward of a thousand nodes using NFS-root
> file systems.

Agreed.  Other than torque.conf, I'm mostly happy with the current state.

If I ruled the world, PBS_SERVER_HOME's default would be based on
automake's $localstatedir (/var).  $sysconfdir/torque.conf (/etc) would
be used to direct all clients and daemons to the location of
PBS_SERVER_HOME (if not using the compiled default.)  

And finally, server_name and pbs_environment would always follow
PBS_SERVER_HOME at run-time, unless set differently at compile-time.

Garrick Staples, Linux/HPCC Administrator
University of Southern California
-------------- next part --------------
A non-text attachment was scrubbed...
Name: not available
Type: application/pgp-signature
Size: 189 bytes
Desc: not available
Url : http://www.supercluster.org/pipermail/torqueusers/attachments/20060203/600ba1b0/attachment.bin

More information about the torqueusers mailing list