def run_HEJFOG(args):
gf.print_flush("TODO HEJFOG not implemented yet")
command = "HEJ/bin/HEJFOG"
gf.do_shell("chmod +x {0}".format(command))
# TODO:
# parse runcard
# run HEJ-FOG (with chmod)
return 1
def bring_nnlojet(input_grid, runcard, runname):
# Todo: this is not very general, is it?
tmp_tar = "nnlojet.tar.gz"
input_name = "{0}/{1}{2}.tar.gz".format(input_grid, runcard, runname)
stat = gf.copy_from_grid(input_name, tmp_tar, args)
stat += gf.untar_file(tmp_tar)
stat += gf.do_shell("rm {0}".format(tmp_tar))
stat += gf.do_shell("ls")
return stat
def download_program(source):
tar_name = os.path.basename(source)
if not tar_name.endswith("tar.gz"):
gf.print_flush("{0} is not a valid path to download".format(source))
return 1
stat = gf.copy_from_grid(source, tar_name, args)
stat += gf.untar_file(tar_name)
stat += gf.do_shell("rm {0}".format(tar_name))
if gf.DEBUG_LEVEL > 2:
gf.do_shell("ls -l")
return stat
def print_copy_status(args, status_copy):
if status_copy == 0:
gf.print_flush("Copied over to grid storage!")
elif args.Sockets:
gf.print_flush(
"This was a socketed run so we are copying the grid to stderr just "
"in case")
gf.do_shell("cat $(ls *.y* | grep -v .txt) 1>&2")
status_copy = 0
elif args.Warmup:
gf.print_flush("Failure! Outputing vegas warmup to stdout")
gf.do_shell("cat $(ls *.y* | grep -v .txt)")
def bring_files(args):
bring_status = 0
if not args.use_cvmfs_lhapdf:
gf.print_flush("Using own version of LHAPDF")
bring_status += bring_lhapdf(args.lhapdf_grid)
bring_status += bring_nnlojet(args.input_folder, args.runcard,
args.runname)
gf.do_shell("chmod +x {0}".format(args.executable))
if bring_status != 0:
gf.print_flush("Not able to bring data from storage. Exiting now.")
sys.exit(-95)
return bring_status
def run_HEJ(args):
config = config_name(args.runname)
seed = args.seed
status = gf.do_shell(
'sed -i -e "s/seed:.1/seed: {0}/g" {1}'.format(seed, config))
status += gf.do_shell(
'sed -i -e "s/output:.HEJ/output: {0}/g" {1}'.format(
yoda_name(seed), config))
status += gf.do_shell("chmod +x {0}".format(args.executable))
if status == 0:
status += gf.run_command(
"{0} {1} {2}".format(args.executable, config, LHE_FILE))
return status
def download_runcard(input_folder, runcard, runname):
tar = warmup_name(runcard, runname)
gf.print_flush("downloading " + input_folder + "/" + tar)
stat = gf.copy_from_grid(input_folder + "/" + tar, tar, args)
gf.print_flush("finished copying from grid storage... untarring now")
stat += gf.untar_file(tar)
return gf.do_shell("rm {0}".format(tar)) + stat
def run_example(args):
status = gf.do_shell("chmod +x {0}".format(args.executable))
if status == 0:
status += gf.run_command("./{executable} {runcard} {outfile}".format(
executable=args.executable,
runcard=args.runname,
outfile="{0}.out".format(args.seed)))
return status
def download_runcard(input_folder, runcard, runname):
tar = warmup_name(runcard, runname)
gf.print_flush("downloading "+input_folder+"/"+tar)
stat = gf.copy_from_grid(input_folder+"/"+tar, tar, args)
stat += gf.untar_file(tar)
# TODO download:
# Scale setters
return gf.do_shell("rm {0}".format(tar))+stat
def download_rivet(rivet_folder):
tar = os.path.basename(rivet_folder)
gf.print_flush("downloading "+rivet_folder)
stat = gf.copy_from_grid(rivet_folder, "", args)
stat += gf.untar_file(tar)
rivet_dir = os.path.basename(os.path.splitext(rivet_folder)[0])
os.environ['RIVET_ANALYSIS_PATH'] = os.getcwd()+"/"+rivet_dir
return gf.do_shell("rm {0}".format(tar))+stat
def run_executable(nnlojet_command):
# TODO replace by gf.run_command
gf.print_flush(" > Executing command: {0}".format(nnlojet_command))
# Run command
status_nnlojet = gf.do_shell(nnlojet_command)
if status_nnlojet == 0:
gf.print_flush("Command successfully executed")
else:
gf.print_flush("Something went wrong")
gf.DEBUG_LEVEL = 9999
return status_nnlojet
def run_example(args):
status = gf.do_shell("chmod +x {0}".format(args.executable))
if status == 0:
activate_environment()
os.system("echo Attempting to activate venv and run simulation...")
status += gf.run_command(
"source JUNE_env/bin/activate && export PYTHONPATH=$PYTHONPATH:`pwd`/JUNE/ && ./{executable} --idx={idx} --world={world} --latin_hypercube={latin_hypercube} {runcard} {outfile}"
.format(executable=args.executable,
idx=args.seed,
world=args.world,
latin_hypercube=args.latin_hypercube,
runcard=args.runname,
outfile="{0}.out".format(args.seed)))
return status
def store_output(args, socketed=False, socket_config=""):
# Copy stuff to grid storage, remove executable and lhapdf folder
# Core files can be upwards of 6G - make sure they're deleted!
os.system("rm core*")
if not args.use_cvmfs_lhapdf:
gf.do_shell("rm -rf {0} {1}".format(args.executable, args.lhapdf_local))
if args.Production:
local_out = output_name(args.runcard, args.runname, args.seed)
output_file = os.path.join(args.output_folder, args.runname, local_out)
elif args.Warmup:
local_out = warmup_name(args.runcard, args.runname)
output_file = os.path.join(args.warmup_folder, local_out)
tar_status = gf.tar_this(local_out, "*")
if gf.DEBUG_LEVEL > 1:
gf.do_shell("ls")
if socketed:
# Avoid conflicting simultaneous copies and resulting error messages:
# - first node attempts copy to warmup_name(...);
# - other nodes attempt copy to warmup_name_ns(...) as backup.
# Still results in large redundancy.
socket_no = int(socket_config.split()[-1].strip())
if socket_no == 1:
status_copy = gf.copy_to_grid(local_out, output_file, args)
else:
subfolder = os.path.splitext(os.path.basename(output_file))[
0].replace(".tar", "")
backup_name = warmup_name_ns(args.runcard, args.runname, socket_no)
backup_fullpath = os.path.join(
args.warmup_folder, subfolder, backup_name)
status_copy = gf.copy_to_grid(local_out, backup_fullpath, args)
else:
status_copy = gf.copy_to_grid(local_out, output_file, args)
return status_copy, tar_status
def setup():
start_time = datetime.datetime.now()
gf.print_flush("Start time: {0}".format(
start_time.strftime("%d-%m-%Y %H:%M:%S")))
args = gf.parse_arguments()
setup_environment(args, args.lhapdf_local)
socket_config = None
if gf.DEBUG_LEVEL > 1:
# Architecture info
gf.print_flush("Python version: {0}".format(sys.version))
gf.print_node_info("node_info.log")
os.system("lsb_release -a")
gf.do_shell("env")
gf.do_shell("voms-proxy-info --all")
if args.copy_log:
# initialise with node name
gf.do_shell("hostname >> {0}".format(gf.COPY_LOG))
return args, socket_config
args = gf.parse_arguments()
lhapdf_local = ""
if args.use_cvmfs_lhapdf:
lhapdf_local = args.cvmfs_lhapdf_location
set_environment(args, lhapdf_local)
if gf.DEBUG_LEVEL > -1:
# Architecture info
gf.print_flush("Python version: {0}".format(sys.version))
gf.print_node_info("node_info.log")
if args.copy_log:
# initialise with node name
gf.do_shell("hostname >> {0}".format(gf.COPY_LOG))
# Debug info
if gf.DEBUG_LEVEL > 16:
gf.do_shell("env")
gf.do_shell("voms-proxy-info --all")
setup_time = datetime.datetime.now()
# Download components
status = download_program()
# uncomment for downloaded exe
# gf.do_shell("chmod +x {0}".format(args.executable))
# uncomment for downloaded exe
# if gf.DEBUG_LEVEL > 8:
# ------------------------- MAIN -------------------------
if __name__ == "__main__":
args, socket_config = setup()
bring_status = bring_files(args)
nnlojet_command = RUN_CMD.format(
args.threads, args.executable, args.runcard)
if args.Sockets:
nnlojet_command, socket_config = setup_sockets(
args, nnlojet_command, bring_status)
if args.Production: # Assume sockets does not work with production
nnlojet_command += " -iseed {0}".format(args.seed)
if gf.DEBUG_LEVEL > 1:
gf.do_shell("ls")
gf.do_shell("ldd -v {0}".format(args.executable))
# Run executable
status_nnlojet = run_executable(nnlojet_command)
# Store output
status_copy, status_tar = store_output(
args, socketed=args.Sockets, socket_config=socket_config)
print_copy_status(args, status_copy)
if args.Sockets:
try: # only the first one arriving will go through!
gf.print_flush("Close Socket connection")
_ = socket_sync_str(args.Host, args.port, "bye!") # Be polite
except socket.error:
def bring_lhapdf(lhapdf_grid):
tmp_tar = "lhapdf.tar.gz"
stat = gf.copy_from_grid(lhapdf_grid, tmp_tar, args)
gf.print_flush("LHAPDF copy from GRID status: {0}".format(stat))
stat += gf.untar_file(tmp_tar)
return gf.do_shell("rm {0}".format(tmp_tar))+stat
args = gf.parse_arguments()
lhapdf_local = ""
if args.use_cvmfs_lhapdf:
lhapdf_local = args.cvmfs_lhapdf_location
set_environment(args, lhapdf_local)
if gf.DEBUG_LEVEL > -1:
# Architecture info
gf.print_flush("Python version: {0}".format(sys.version))
gf.print_node_info("node_info.log")
if args.copy_log:
# initialise with node name
gf.do_shell("hostname >> {0}".format(gf.COPY_LOG))
# Debug info
if gf.DEBUG_LEVEL > 16:
gf.do_shell("env")
gf.do_shell("voms-proxy-info --all")
setup_time = datetime.datetime.now()
# Download components
# we are abusing "args.warmup_folder", which is otherwise not needed for HEJ
status = download_program(args.warmup_folder)
gf.do_shell("chmod +x {0}".format(args.executable))
if gf.DEBUG_LEVEL > 8:
gf.do_shell("ldd {0}".format(args.executable))