def test_job_script():
with HTCondorCluster(
cores=4,
processes=2,
memory="100MB",
disk="100MB",
env_extra=[
'export LANG="en_US.utf8"', 'export LC_ALL="en_US.utf8"'
],
job_extra={"+Extra": "True"},
) as cluster:
job_script = cluster.job_script()
assert "RequestCpus = MY.DaskWorkerCores" in job_script
assert "RequestDisk = floor(MY.DaskWorkerDisk / 1024)" in job_script
assert "RequestMemory = floor(MY.DaskWorkerMemory / 1048576)" in job_script
assert "MY.DaskWorkerCores = 4" in job_script
assert "MY.DaskWorkerDisk = 100000000" in job_script
assert "MY.DaskWorkerMemory = 100000000" in job_script
assert 'MY.JobId = "$(ClusterId).$(ProcId)"' in job_script
assert "LANG=en_US.utf8" in job_script
assert "LC_ALL=en_US.utf8" in job_script
assert "export" not in job_script
assert "+Extra = True" in job_script
assert ("{} -m distributed.cli.dask_worker tcp://".format(
sys.executable) in job_script)
assert "--memory-limit 50.00MB" in job_script
assert "--nthreads 2" in job_script
assert "--nprocs 2" in job_script
def test_basic(loop):
with HTCondorCluster(cores=1, memory="100MB", disk="100MB",
loop=loop) as cluster:
with Client(cluster) as client:
cluster.scale(2)
start = time()
while not (cluster.pending_jobs or cluster.running_jobs):
sleep(0.100)
assert time() < start + QUEUE_WAIT
future = client.submit(lambda x: x + 1, 10)
assert future.result(QUEUE_WAIT) == 11
assert cluster.running_jobs
workers = list(client.scheduler_info()["workers"].values())
w = workers[0]
assert w["memory_limit"] == 1e8
assert w["nthreads"] == 1
cluster.scale(0)
start = time()
while cluster.running_jobs:
sleep(0.100)
assert time() < start + QUEUE_WAIT
def test_extra_args_broken_cancel(loop):
with HTCondorCluster(
cores=1,
memory="100MB",
disk="100MB",
loop=loop,
cancel_command_extra=["-name", "wrong.docker"],
) as cluster:
with Client(cluster) as client:
cluster.scale(2)
client.wait_for_workers(2)
workers = Job._call(["condor_q", "-af", "jobpid"]).strip()
assert workers, "we got dask workers"
cluster.scale(0)
start = time()
while client.scheduler_info()["workers"]:
sleep(0.100)
workers = Job._call(["condor_q", "-af", "jobpid"]).strip()
assert workers, "killing workers with broken cancel_command didn't fail"
if time() > start + QUEUE_WAIT // 3:
return
def test_header():
with HTCondorCluster(cores=1, memory="100MB", disk="100MB") as cluster:
assert cluster._dummy_job.job_header_dict["MY.DaskWorkerCores"] == 1
assert cluster._dummy_job.job_header_dict[
"MY.DaskWorkerDisk"] == 100000000
assert cluster._dummy_job.job_header_dict[
"MY.DaskWorkerMemory"] == 100000000
def test_job_script():
with HTCondorCluster(cores=4,
processes=2,
memory='100MB',
disk='100MB',
env_extra=[
'export LANG="en_US.utf8"',
'export LC_ALL="en_US.utf8"'
],
job_extra={'+Extra': "True"}) as cluster:
job_script = cluster.job_script()
assert 'RequestCpus = MY.DaskWorkerCores' in job_script
assert 'RequestDisk = floor(MY.DaskWorkerDisk / 1024)' in job_script
assert 'RequestMemory = floor(MY.DaskWorkerMemory / 1048576)' in job_script
assert 'MY.DaskWorkerCores = 4' in job_script
assert 'MY.DaskWorkerDisk = 100000000' in job_script
assert 'MY.DaskWorkerMemory = 100000000' in job_script
assert 'MY.JobId = "$(ClusterId).$(ProcId)"' in job_script
assert 'LANG=en_US.utf8' in job_script
assert 'LC_ALL=en_US.utf8' in job_script
assert 'JOB_ID=$F(MY.JobId)' in job_script
assert 'export' not in job_script
assert '+Extra = True' in job_script
assert '{} -m distributed.cli.dask_worker tcp://'.format(
sys.executable) in job_script
assert '--memory-limit 50.00MB' in job_script
assert '--nthreads 2' in job_script
assert '--nprocs 2' in job_script
def test_config_name_htcondor_takes_custom_config():
conf = {
"cores": 1,
"memory": "120 MB",
"disk": "120 MB",
"job-extra": [],
"name": "myname",
"processes": 1,
"interface": None,
"death-timeout": None,
"extra": [],
"env-extra": [],
"log-directory": None,
"shebang": "#!/usr/bin/env condor_submit",
"local-directory": "/tmp",
}
with dask.config.set({"jobqueue.htcondor-config-name": conf}):
with HTCondorCluster(config_name="htcondor-config-name") as cluster:
assert cluster.job_name == "myname"
def test_config_name_htcondor_takes_custom_config():
conf = {
'cores': 1,
'memory': '120 MB',
'disk': '120 MB',
'job-extra': [],
'name': 'myname',
'processes': 1,
'interface': None,
'death-timeout': None,
'extra': [],
'env-extra': [],
'log-directory': None,
'shebang': "#!/usr/bin/env condor_submit",
'local-directory': "/tmp",
}
with dask.config.set({'jobqueue.htcondor-config-name': conf}):
with HTCondorCluster(config_name='htcondor-config-name') as cluster:
assert cluster.name == 'myname'
def main():
n_port = 8786
with HTCondorCluster(cores=1,
memory='100MB',
disk='100MB',
death_timeout='60',
nanny=True,
scheduler_options={
'port': n_port,
'host': socket.gethostname()
},
job_extra={
'should_transfer_files': 'Yes',
'when_to_transfer_output': 'ON_EXIT'
},
extra=['--worker-port {}'.format(n_port)]) as cluster:
with Client(cluster) as client:
cluster.scale(1)
future = client.submit(lambda x: x + 1, 10)
print('Result is {}'.format(future.result()))
def main(args):
base_dir = args.base_dir
n_workers = args.n_workers
input_files = [base_dir + "/" + fl for fl in os.listdir(base_dir)]
print("Found {} files in {}".format(len(input_files), base_dir))
start = time()
n_port = 8786
with HTCondorCluster(cores=1,
memory="500MB",
disk="500MB",
death_timeout="60",
nanny=False,
scheduler_options={
"port": n_port,
"host": socket.gethostname()
},
job_extra={
"should_transfer_files": "Yes",
"when_to_transfer_output": "ON_EXIT",
"+JobFlavour": "espresso"
},
extra=["--worker-port {}".format(n_port)]) as cluster:
#print(cluster.job_script())
with Client(cluster) as client:
cluster.scale(n_workers)
futures = client.map(dummy_extractor, input_files)
arrays = client.gather(futures)
end = time()
final_arr = ak.concatenate(arrays)
print("Done concatenating")
print(final_arr.type)
print("Computation time: {}".format(
str(timedelta(seconds=int(end - start)))))
def test_job_script():
with HTCondorCluster(
cores=4,
processes=2,
memory="100MB",
disk="100MB",
env_extra=[
'export LANG="en_US.utf8"', 'export LC_ALL="en_US.utf8"'
],
job_extra={"+Extra": "True"},
submit_command_extra=["-verbose"],
cancel_command_extra=["-forcex"],
) as cluster:
job_script = cluster.job_script()
assert "RequestCpus = MY.DaskWorkerCores" in job_script
assert "RequestDisk = floor(MY.DaskWorkerDisk / 1024)" in job_script
assert "RequestMemory = floor(MY.DaskWorkerMemory / 1048576)" in job_script
assert "MY.DaskWorkerCores = 4" in job_script
assert "MY.DaskWorkerDisk = 100000000" in job_script
assert "MY.DaskWorkerMemory = 100000000" in job_script
assert 'MY.JobId = "$(ClusterId).$(ProcId)"' in job_script
assert "LANG=en_US.utf8" in job_script
assert "LC_ALL=en_US.utf8" in job_script
assert "export" not in job_script
assert "+Extra = True" in job_script
assert re.search(r"condor_submit\s.*-verbose",
cluster._dummy_job.submit_command)
assert re.search(r"condor_rm\s.*-forcex",
cluster._dummy_job.cancel_command)
assert ("{} -m distributed.cli.dask_worker tcp://".format(
sys.executable) in job_script)
formatted_bytes = format_bytes(parse_bytes("50MB")).replace(" ", "")
assert f"--memory-limit {formatted_bytes}" in job_script
assert "--nthreads 2" in job_script
assert "--nprocs 2" in job_script
require_encryption=True)
HTCondorJob.submit_command = "condor_submit -spool"
cluster = HTCondorCluster(
cores=4,
memory="2GB",
disk="1GB",
log_directory="logs",
silence_logs="debug",
scheduler_options={
"dashboard_address": "8786",
"port": 8787,
"external_address": "129.93.183.33:8787"
},
# HTCondor submit script
job_extra={
"universe": "docker",
# To be used with coffea-casa:0.1.11
"encrypt_input_files": "/etc/cmsaf-secrets/xcache_token",
#"docker_network_type": "host",
"docker_image": "coffeateam/coffea-casa-analysis:0.1.11",
"container_service_names": "dask",
"dask_container_port": "8787",
"should_transfer_files": "YES",
"when_to_transfer_output": "ON_EXIT",
"+DaskSchedulerAddress": '"129.93.183.33:8787"',
})
cluster.adapt(minimum_jobs=5, maximum_jobs=100, maximum_memory="4 GB"
) # auto-scale between 5 and 100 jobs (maximum_memory="4 GB")
from tdub import setup_logging
from tdub.train import prepare_from_root
from tdub.utils import get_selection, get_features, quick_files
import lightgbm as lgbm
from sklearn.model_selection import train_test_split
from dask_jobqueue import HTCondorCluster
from dask.distributed import Client
from dask_ml.model_selection import GridSearchCV
cluster = HTCondorCluster(cores=2, disk="4GB", memory="8GB")
client = Client(cluster)
cluster.adapt(maximum_jobs=200)
setup_logging()
qf = quick_files("/atlasgpfs01/usatlas/data/ddavis/wtloop/v29_20191111")
df, y, w = prepare_from_root(qf["tW_DR"], qf["ttbar"], "1j1b")
X_train, X_test, y_train, y_test, w_train, w_test = train_test_split(
df, y, w, train_size=0.8, random_state=414, shuffle=True)
n_sig = y_train[y_train == 1].shape[0]
n_bkg = y_train[y_train == 0].shape[0]
spw = n_bkg / n_sig
n_sig = y[y == 1].shape[0]
n_bkg = y[y == 0].shape[0]
spw = n_bkg / n_sig
def test_header():
with HTCondorCluster(cores=1, memory='100MB', disk='100MB') as cluster:
assert cluster.job_header_dict['MY.DaskWorkerCores'] == 1
assert cluster.job_header_dict['MY.DaskWorkerDisk'] == 100000000
assert cluster.job_header_dict['MY.DaskWorkerMemory'] == 100000000
if not check_port(8786):
raise RuntimeError(
"Port '8786' is not occupied on this node. Try another one."
)
import socket
cluster = HTCondorCluster(
cores=1,
memory='2GB', # hardcoded
disk='1GB',
death_timeout='60',
nanny=False,
scheduler_options={
'port': n_port,
'host': socket.gethostname()
},
job_extra={
'log': 'dask_job_output.log',
'output': 'dask_job_output.out',
'error': 'dask_job_output.err',
'should_transfer_files': 'Yes',
'when_to_transfer_output': 'ON_EXIT',
'+JobFlavour': '"workday"',
},
extra=['--worker-port {}'.format(n_port)],
env_extra=env_extra,
)
elif 'slurm' in args.executor:
cluster = SLURMCluster(
queue='all',
cores=args.workers,
processes=args.workers,
HTCondorJob.submit_command = "condor_submit -spool"
cluster = HTCondorCluster(
cores=4,
memory="2GB",
disk="1GB",
log_directory="logs",
silence_logs="debug",
scheduler_options={
"dashboard_address": "8786",
"port": 8787,
"external_address": "129.93.183.33:8787"
},
# HTCondor submit script
job_extra={
"universe": "docker",
# To be used with coffea-casa:0.1.7
"transfer_input_files":
"/etc/cmsaf-secrets/xcache_token,/etc/cmsaf-secrets/ca.pem,/etc/cmsaf-secrets/usercert.pem",
"encrypt_input_files":
"/etc/cmsaf-secrets/xcache_token,/etc/cmsaf-secrets/ca.pem,/etc/cmsaf-secrets/usercert.pem",
#"docker_network_type": "host",
"docker_image": "oshadura/coffea-casa-analysis:0.1.7",
"container_service_names": "dask",
"dask_container_port": "8787",
"should_transfer_files": "YES",
"when_to_transfer_output": "ON_EXIT",
"+DaskSchedulerAddress": '"129.93.183.33:8787"',
})
cluster.scale(jobs=2)
from dask.distributed import performance_report
if 'slurm' in args.executor:
cluster = SLURMCluster(
queue='all',
cores=args.workers,
processes=args.workers,
memory="200 GB",
retries=10,
walltime='00:30:00',
env_extra=env_extra,
)
elif 'condor' in args.executor:
cluster = HTCondorCluster(
cores=args.workers,
memory='4GB',
disk='4GB',
env_extra=env_extra,
)
if args.executor == 'dask/casa':
client = Client("tls://localhost:8786")
# client = Client("tls://andrzej-2enovak-40cern-2ech.dask.coffea.casa:8786")
# import shutil
# shutil.make_archive("workflows", "zip", base_dir="workflows")
# client.upload_file("workflows.zip")
from distributed.diagnostics.plugin import UploadDirectory
# client.register_worker_plugin(UploadDirectory("workflows", restart=True, update_path=True), nanny=True)
client.register_worker_plugin(UploadDirectory("workflows",
restart=False,
update_path=True),
nanny=True)