Hi all,
For those sites using SGE, I was wondering if someone could give me some
advice on how to best to stop grid jobs writing scratch files out to /tmp.
Currently our site's grid queues in SGE still have the 'tmpdir'
attribute set to the default of /tmp. So I suppose I could change this
attribute to the path where we want jobs to write scratch files to.
However I can see that there is a certain amount of grid environment
setup by yaim on the worker nodes under /etc/profile.d/grid-env.sh , but
I'm not yet familiar enough with yaim to see where a lot of it comes from.
On our WN's /etc/yaim/site-info.def, there is the promisingly named
variable:
TMPDIR="/mnt/lustre/grid/scratch"
however that doesn't seem to be used anywhere (not in grid-env.sh) and I
can't see any reference to such a variable under
/opt/glite/yaim/examples/siteinfo/*
Also looking over this page of yaim site-info variables:
https://twiki.cern.ch/twiki/bin/view/LCG/Site-info_configuration_variables#site_info_def
there doesn't appear to be such a TMPDIR variable. There is however:
SUBCLUSTER_<subcluster-identifier>_SUBCLUSTER_TmpDir
SUBCLUSTER_<subcluster-identifier>_SUBCLUSTER_WNTmpDir
Can anyone advise whether this is on the right track, and whether I can
just add these variables to my site-info.def on the WNs?
Also possibly relevant, when running a yaim on the work nodes:
/opt/glite/yaim/bin/yaim -c -s /etc/yaim/site-info.def -n WN -n GLEXEC_wn
I get some warnings:
WARNING: No subcluster has been defined for the WN in the WN_LIST file
/etc/yaim/wn-list.conf
WARNING: YAIM will use the default subcluster id: CE_HOST ->
grid-cream-01.hpc.susx.ac.uk
So if I *do* use the above two variables I suppose I should use
'CE_HOST' as the <subcluster-identifier>?
Any advice on this would be much appreciated as I only have a very thing
grasp of what I am doing...!
Matt
--
Matt Raso-Barnett
Linux Systems Administrator -- MPS
University of Sussex
|