def __enter__(self): # for cluster execution, the working directory can not be # local. Use a temporary shared location instead and copy # files over after job has completed. Note that this assumes # that all paths of input files are absolute and reside in a # shared directory. if self.work_dir_is_local: self.original_dir = self.work_dir self.work_dir = get_temp_dir(shared=True) return self
def build_job_script(self, statement): '''build job script from statement. returns (name_of_script, stdout_path, stderr_path) ''' tmpfilename = get_temp_filename(dir=self.work_dir, clear=True) tmpfilename = tmpfilename + ".sh" expanded_statement, cleanup_funcs = self.expand_statement(statement) with open(tmpfilename, "w") as tmpfile: # disabled: -l -O expand_aliases\n" ) # make executable tmpfile.write("#!/bin/bash -eu\n") if not self.ignore_pipe_errors: tmpfile.write("set -o pipefail\n") os.chmod(tmpfilename, stat.S_IRWXG | stat.S_IRWXU) tmpfile.write("\ncd {}\n".format(self.work_dir)) if self.output_directories is not None: for outdir in self.output_directories: if outdir: tmpfile.write("\nmkdir -p {}\n".format(outdir)) # create and set system scratch dir for temporary files tmpfile.write("umask 002\n") cluster_tmpdir = get_params()["cluster_tmpdir"] if self.run_on_cluster and cluster_tmpdir: tmpdir = cluster_tmpdir tmpfile.write("TMPDIR=`mktemp -d -p {}`\n".format(tmpdir)) tmpfile.write("export TMPDIR\n") else: tmpdir = get_temp_dir(dir=get_params()["tmpdir"], clear=True) tmpfile.write("mkdir -p {}\n".format(tmpdir)) tmpfile.write("export TMPDIR={}\n".format(tmpdir)) cleanup_funcs.append( ("clean_temp", "{{ rm -rf {}; }}".format(tmpdir))) # output times whenever script exits, preserving # return status cleanup_funcs.append( ("info", "{ echo 'benchmark'; hostname; times; }")) for cleanup_func, cleanup_code in cleanup_funcs: tmpfile.write("\n{}() {}\n".format(cleanup_func, cleanup_code)) tmpfile.write("\nclean_all() {{ {}; }}\n".format("; ".join( [x[0] for x in cleanup_funcs]))) tmpfile.write("\ntrap clean_all EXIT\n\n") if self.job_memory not in("unlimited", "etc") and \ self.options.get("cluster_memory_ulimit", False): # restrict virtual memory # Note that there are resources in SGE which could do this directly # such as v_hmem. # Note that limiting resident set sizes (RSS) with ulimit is not # possible in newer kernels. # -v and -m accept memory in kb requested_memory_kb = max( 1000, int( math.ceil( iotools.human2bytes(self.job_memory) / 1024 * self.job_threads))) # unsetting error exit as often not permissions tmpfile.write("set +e\n") tmpfile.write( "ulimit -v {} > /dev/null \n".format(requested_memory_kb)) tmpfile.write( "ulimit -m {} > /dev/null \n".format(requested_memory_kb)) # set as hard limit tmpfile.write("ulimit -H -v > /dev/null \n") tmpfile.write("set -e\n") if self.shellfile: # make sure path exists that we want to write to tmpfile.write("mkdir -p $(dirname \"{}\")\n".format( self.shellfile)) # output low-level debugging information to a shell log file tmpfile.write('echo "%s : START -> %s" >> %s\n' % (self.job_name, tmpfilename, self.shellfile)) # disabled - problems with quoting # tmpfile.write( '''echo 'statement=%s' >> %s\n''' % # (shellquote(statement), self.shellfile) ) tmpfile.write("set | sed 's/^/%s : /' >> %s\n" % (self.job_name, self.shellfile)) tmpfile.write("pwd | sed 's/^/%s : /' >> %s\n" % (self.job_name, self.shellfile)) tmpfile.write("hostname | sed 's/^/%s: /' >> %s\n" % (self.job_name, self.shellfile)) # cat /proc/meminfo is Linux specific if get_params()['os'] == 'Linux': tmpfile.write( "cat /proc/meminfo | sed 's/^/%s: /' >> %s\n" % (self.job_name, self.shellfile)) elif get_params()['os'] == 'Darwin': tmpfile.write("vm_stat | sed 's/^/%s: /' >> %s\n" % (self.job_name, self.shellfile)) tmpfile.write('echo "%s : END -> %s" >> %s\n' % (self.job_name, tmpfilename, self.shellfile)) tmpfile.write("ulimit | sed 's/^/%s: /' >> %s\n" % (self.job_name, self.shellfile)) job_path = os.path.abspath(tmpfilename) tmpfile.write(expanded_statement) tmpfile.write("\n\n") tmpfile.close() return statement, job_path