def __init__(self, **kwargs):
self.logger = get_logger()
self.queue_manager = None
self.run_on_cluster = will_run_on_cluster(kwargs)
self.job_threads = kwargs.get("job_threads", 1)
if "job_memory" in kwargs and "job_total_memory" in kwargs:
raise ValueError(
"both job_memory and job_total_memory have been given")
self.job_total_memory = kwargs.get('job_total_memory', None)
self.job_memory = kwargs.get('job_memory', None)
if self.job_total_memory == "unlimited" or self.job_memory == "unlimited":
self.job_total_memory = self.job_memory = "unlimited"
else:
if self.job_total_memory:
self.job_memory = iotools.bytes2human(
iotools.human2bytes(self.job_total_memory) /
self.job_threads)
elif self.job_memory:
self.job_total_memory = self.job_memory * self.job_threads
else:
self.job_memory = get_params()["cluster"].get(
"memory_default", "4G")
if self.job_memory == "unlimited":
self.job_total_memory = "unlimited"
else:
self.job_total_memory = self.job_memory * self.job_threads
self.ignore_pipe_errors = kwargs.get('ignore_pipe_errors', False)
self.ignore_errors = kwargs.get('ignore_errors', False)
self.job_name = kwargs.get("job_name", "unknow_job_name")
self.task_name = kwargs.get("task_name", "unknown_task_name")
# deduce output directory/directories, requires somewhat
# consistent naming in the calling function.
outfiles = []
if "outfile" in kwargs:
outfiles.append(kwargs["outfile"])
if "outfiles" in kwargs:
outfiles.extend(kwargs["outfiles"])
self.output_directories = set(
sorted([os.path.dirname(x) for x in outfiles]))
self.options = kwargs
self.work_dir = get_params()["work_dir"]
self.shellfile = kwargs.get("shell_logfile", None)
if self.shellfile:
if not self.shellfile.startswith(os.sep):
self.shellfile = os.path.join(self.work_dir,
os.path.basename(self.shellfile))
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