After looking at the code of CondorJob in GridScala again, I can tell that such requirement specification is not possible at the moment. I’ve thus created an issue for this: https://github.com/openmole/gridscale/issues/5 <https://github.com/openmole/gridscale/issues/5> .
> On 11 May 2015, at 12:19, Andreas Schuh <[email protected]> wrote: > > Hi, > > many of my jobs are failing due to the currently missing requirements. How > would I specify a requirement such as the following in the CondorEnvironment? > The need for a list of 2-tuples makes me wonder... > > Arch == "X86_64" && OpSysShortName =?= “ubuntu” && OpSysMajorVer >= 12 > > where I believe the “Arch” requirement is already set by OpenMOLE. > > Thanks, > Andreas > >> On 8 May 2015, at 09:38, Andreas Schuh <[email protected] >> <mailto:[email protected]>> wrote: >> >> >>> On 8 May 2015, at 08:58, Jonathan Passerat-Palmbach >>> <[email protected] >>> <mailto:[email protected]>> wrote: >>> >>> Hi Andreas, >>> >>> For Condor, the requirements are defined in a list which elements are >>> combined by OpenMOLE using &&. >>> If you want an OR behaviour, you'd have to express it literally in your >>> requirement string. >> >> Ok, but wouldn’t that be a bit nasty ? Because I would have to say >> >> CondorRequirement(“OpSysMajorVer”, “12 || OpSysMajorVer == 13 || >> OpSysMajorVer == 14”) >> >> Or did I misunderstand the arguments of the CondorRequirement 2-tuple ? >> Also, it seems this way I couldn’t use a non-equality or other comparison. >> Compared to the SLURMEnvironment, which takes just simple strings for each >> constraint (1-tuple), the CondorEnvironment requires a special type which is >> effectively a 2-tuple… maybe it would make more sense to have the >> CondorEnvironment behave equivalent to the SLURMEnvironment which would make >> it also more flexible when I can specify the operator between key >> (requirement name) and value myself. >> >>> >>> As for the environment, we don't specify any particular flag. >>> I guess it works because you define this environment variable in your >>> .bashrc and we force it to be loaded before the execution starts. >> >> Alright, fair enough. I then just need a way to figure out on which machines >> in our lab this doesn’t work because maybe they run a different OS or have >> not mounted the shared directories where my libraries are located… >> >> Does OpenMOLE report on which compute nodes tasks failed ? This would be >> helpful in debugging such issues. >> >>> >>> Cheers >>> J. >>> >>> On 7 May 2015 22:38:05 CEST, Andreas Schuh <[email protected] >>> <mailto:[email protected]>> wrote: >>> Hi, >>> >>> the CondorEnvironment has a requirements argument which is a list of >>> key/value tuples. I assume these are combined using a logical AND. Is it >>> also possible to have some requirements combined using a logical OR ? In >>> particular, I have used the following for our HTCondor environment in my >>> scripts: >>> >>> requirements = OpSysShortName == "Ubuntu" && (OpSysMajorVer == 12 || >>> OpSysMajorVer == 13 || OpSysMajorVer == 14) >>> >>> (although I would be fine with something like OpSysMajorVer >= 12 instead…) >>> >>> Further, I also had to set the LD_LIBRARY_PATH in my own condor_submit >>> scripts. As my programs run fine on most machines using OpenMOLE without >>> specifying it I am wondering if it uses some condor_submit flag that asks >>> it to pass all environment variables on ? >>> >>> environment = >>> LD_LIBRARY_PATH=/homes/as12312/opt/lib64:/homes/as12312/opt/lib:/usr/lib/matlab/R2012a/sys/os:/usr/lib/matlab/R2012a/runtime/glnxa64:/usr/lib/matlab/R2012a/bin/glnxa64 >>> >>> Andreas >>> >>> OpenMOLE-users mailing list >>> [email protected] <mailto:[email protected]> >>> http://fedex.iscpif.fr/mailman/listinfo/openmole-users >>> <http://fedex.iscpif.fr/mailman/listinfo/openmole-users> >>> >>> -- >>> Jonathan Passerat-Palmbach, PhD >>> Research Associate >>> Department of Computing >>> Imperial College London >>> >>> South Kensigton Campus >>> Huxley Building - room 344 >>> 180 Queen's Gate >>> London SW7 2AZ >> >
_______________________________________________ OpenMOLE-users mailing list [email protected] http://fedex.iscpif.fr/mailman/listinfo/openmole-users
