Running R in a HTCondor Pool
It works very well to move the R runtime in either Windows or Linux with the job. Then one actually runs a wrapper script to set up the runtime and place things as needed. The only moderately tricky part is handling non-included R Packages.
Though the commands are the same in both a Wiindows R developement location(One where a matching RTools is installed and R was built using those tools) the environment settings to have a local Package store found is a bit different.
I make sure that the directory RR/library exists and then I run this command and tar up the RR directory afRLIBSter:
R CMD INSTALL --preclean -l ./RR/library package.tar.gx tar -zcvf RLIBS.tar.gz RR
With windows I ship cygwin tools for tar and gzip and associated DLLs so there is some similarity between extractions by the wrapper scripts.
Under Linux using perl I drop R into current directory and set these two things:
$ENV{R_HOME} = "$Rlocation";
$ENV{LD_LIBRARY_PATH} = "$Rlocation/lib";
If there is a library tar ball as mentioned above, the following is done:
# do we have prebuilt libraries?
my $renviron;
if(-f "RLIBS.tar.gz") {
system("tar -zxvf RLIBS.tar.gz");
$ENV{HOME} = $location;
$renviron = "$location/.Renviron";
open(RLIB,">$renviron") or die "Can not create <$renviron>:$!\n";
print RLIB "R_LIBS_USER=$location/RR/library\n";
close(RLIB);
}
Once that is done, the following starts R:
$cmdtorun = "$Rlocation/bin/Rscript --no-save ./$scripts $realarg";
print "about to execute <<$cmdtorun>>\n";
$res = system("$cmdtorun");
Windows is similar but wrapper is a .bat script. Here is what we use in the Center for High Throughput Computing for packages, shared files for sets up jobs, job specific input, the runtime etc.
dir time /t if not exist built-win32-R-2.10.1.tar.gz goto noruntime set R_LIBS_USER=RLIBS tar.exe -zxvf built-win32-R-2.10.1.tar.gz if not exist RLIBS.tar.gz goto nolibs tar.exe -zxvf RLIBS.tar.gz set R_LIBS_USER=RLIBS :nolibs if not exist %2.tar.gz goto nojobfiles tar -xvf %2.tar :nojobfiles if not exist sharedfiles.tar.gz goto nosharedfiles tar -zxvf sharedfiles.tar.gz :nosharedfiles mkdir tmp set PATH=%path%;.\bin dir dir .\bin echo Try job now Rscript.exe soartest.R dir echo Job completed :noruntime echo Failed to find runtime. This is bad!