Hi,
I have trouble submitting a job from the UI which may (may not?) be
related the home account changes (see below).
However, I see that we have gone from red to green so something is
working :-)
Regards, Dave.
[kant@hepbf3]$ globus-job-run hepbf4.ph.qmul.ac.uk:2119/jobmanager-pbs /bin/hostname
GRAM Job submission failed because cannot access cache files in
~/.globus/.gass_cache, check permissions, quota, and disk space (error code 76)
=================================================================================
To deal with the home accounts issue, I did the following:
1. Created a CE_HOME in site-cfg.h file
/* Global HomeDir */
#define CE_HOME /home/edg
2. Changed all occurences of /home to CE_HOME in the following files:
WorkerNode-cfg.h
nfsmount.nfsdetails_home CE_HOME CE_HOSTNAME:CE_HOME rw
StorageElement-cfg.h
nfsmount.nfsdetails_home CE_HOME CE_HOSTNAME:CE_HOME rw
ComputingElement-cfg.h
EXTRA(nfs.exports) home
+nfs.fs_home CE_HOME
Users-cfg.h
auth.usercomment_alice001 mapped user for alice
+auth.userhome_alice001 CE_HOME/alice001
+auth.usergroup_alice001 alice
+auth.usersuppgroups_alice001 iteam
+auth.useruid_alice001 1001
User-cfg.h has loads of /home entries so I used sed to do a global replace
sed s/'\/home'/'CE_HOME'/ Users-cfg.h > Users-cfg.qmul.h
=================================================================================
On Tue, 20 May 2003, Steve Traylen wrote:
> Hi Dave,
>
> I think you may be okay with this , providing the password
> file(or nis) has the correct entry on each worker node, ie
> /home/edg/wpsix001 then everything should just work. I believe
> scotgrid has something similar. $HOME will be evaluated on
> each node.
>
> If it does not work then altering the Users-cfg.h file to use
> /home/edg rather than home is the sure way to solve this.
>
> Steve
>
> On Thu, 15 May 2003, D.Kant wrote:
>
> > Hi Everyone,
> >
> > We have an e-science farm which has its own front-end running a
> > PBS job manager and some worker nodes. I want our edg front-end to
> > pass job requests to the e-science front-end in much the same way
> > as the prescription described by Andrew Mcnab in reference to Babar.
> >
> > The e-science farm has different home directories for different
> > applications e.g /home/hep, /home/astro and /home/edg
> >
> > Now,
> > an edg worker node typically has mount points which look like
> > this:
> >
> > hepbf4:/home /home
> > hepbf2:/flatfiles /flatfiles
> > hepbf4:/share/grid-security /share/grid-security
> >
> > but an e-science worker node, configured to run edg jobs passed to it
> > from the e-science front-end would have mounts points like this:
> >
> > hepbf4:/home /home/edg
> > hepbf2:/flatfiles /flatfiles
> > hepbf4:/share/grid-security /share/grid-security
> >
> > Now lets assume that someone submits a job to our CE (hepbf4) and after
> > successful authentication the job gets mapped to $HOME/tutor01. Now on
> > the edg side $HOME is /home and the automatically created accounts are
> > visible, so no problem there. However, on the e-science side, $HOME is
> > also /home but no accounts are visible because the edg home directories
> > are mounted elsewhere! Clearly there will be a problem !!!
> >
> > Perhaps there is a way in site-cfg.h to define a CE_HOME ???
> >
> > Thanks, Dave.
> >
> > --------------------------------------------------------------
> > Department of Physics | Dr Dave Kant
> > Queen Mary College | TEL/FaX: +44 (0)20 7882 5054
> > Mile End Road London E1 4NS | e-mail : [log in to unmask]
> > --------------------------------------------------------------
> >
>
> --
> Steve Traylen
> [log in to unmask]
> http://www.gridpp.ac.uk/
>
--
--------------------------------------------------------------
Department of Physics | Dr Dave Kant
Queen Mary College | TEL/FaX: +44 (0)20 7882 5054
Mile End Road London E1 4NS | e-mail : [log in to unmask]
--------------------------------------------------------------
|