probleme at obelix : RunChecker and TS

[jgipsl]

using libIGCM trunk rev 964

At obelix, the RunChecker.job crash with following error message :

./RunChecker.job[167]: IGCM_check_SearchCatalog[114]: unset: warning: SUBMIT_DIR: is read only
./RunChecker.job[167]: IGCM_check_SearchCatalog: line 192: SUBMIT_DIR: is read only

Find exemple here :

cd /home/scratch01/jghattas/MYPOSTTEST/modipsl_libItrunk/libIGCM
./RunChecker.job -u jghattas -p /home/scratch01/jghattas/MYPOSTTEST/modipsl_libItrunk/config/ORCHIDEE_OL/testrun

At obelix, also the TS do not work, they are not even launched because the rebuild_fromWorkdir crash in the end on an "export SUBMIT_DIR". Error message :

# Exemple from file /home/scratch01/jghattas/IGCM_OUT/OL2/TEST/test/testrun/Out/rebuild_fromWorkdir.19821231.out
/var/spool/torque/mom_priv/jobs/520464.obelix1.SC[156]: IGCM_FlushRebuild: line 38: SUBMIT_DIR: is read only

And in Script_Output, there is another error message, minor I think :

# exemple from /home/scratch01/jghattas/IGCM_OUT/OL2/TEST/test/testrun/Out/Script_Output_testrun.000001.30514
cat: write error: Broken pipe

[jgipsl]

Commit [968] corrects error on RunChecker and create_TS.
Last error in Script_output remains.

[jgipsl]

The remaining problem occures when concating the out_orchidee_000x files into one file. The file is producced and looks ok so I don't know if the error message makes a difference.

The probleme occures when the files($file) are to big.
The limit in size is between 175171 and 227169.

The corresponing line is 1057 in libIGCM_comp.ksh :

cat ${file} | sed "s/\(.*\)/${i_}\1/" ${file} >> ${file_out}

[sdipsl]

try this:

cat ${file} | xargs sed "s/\(.*\)/${i_}\1/" ${file} >> ${file_out}

[jgipsl]

Doesn't work...

[sdipsl]

One possible work-around would be to try to increase ulimit -s. it is 10240 by default. On Linux, the maximum amount of space for command arguments is 1/4th of the amount of available stack space. So on obelix what is the value of this.
ulimit -s

If not 10240 then change that
ulimit -s 10240

If you are not allowed to do that then ask the obelix sys admin to do it for you

Trop bizarre de se parler en anglais :-)

[jgipsl]

C'est déjà mis a unlimited sur obelix... donc pas de chance, pas d'amélioration...

[sdipsl]

Est ce toujours un problème?

Si oui qu'en pense les admins de obelix? Je trouve que la balle est dans leur camp étant la taille de fichier à partir de laquelle le problème se pose.

[jgipsl]

Francois Marabelle(LSCE) dit :

Je pense que le script est faux. il faudrait faire:

cat ${file} | sed "s/\(.*\)/${i_}\1/"  >> ${file_out}

au lieu de:

cat ${file} | sed "s/\(.*\)/${i_}\1/" ${file} >> ${file_out}

Josefine : J'ai testé sur obelix, ca marche. Ca ne donne plus des messages d'erreur et le fichier out_orchidee est identique a celui d'avant.