def run_simulation_with_PJ(self, sh_jobs_scripts):
"""
running simulation from SWEEP dir using PJ
"""
from qcg.pilotjob.api.job import Jobs
jobs = Jobs()
for sh_job_scripts in sh_jobs_scripts:
sweep_dir_name = os.path.basename(os.path.dirname(sh_job_scripts))
jobs.add(
name="SWEEP_{}".format(sweep_dir_name),
exec="bash",
args=["-l", sh_job_scripts],
stdout="{}/{}.stdout".format(os.path.dirname(sh_job_scripts),
"${jname}__${uniq}"),
stderr="{}/{}.stderr".format(os.path.dirname(sh_job_scripts),
"${jname}__${uniq}"),
numCores={"exact": self.cores},
model="default")
print("\nAdd job with :")
print("name=SWEEP_{}".format(sweep_dir_name))
print("args = [-l,{}]".format(sh_job_scripts))
print("stdout = {}/{}.stdout".format(
os.path.dirname(sh_job_scripts), "${jname}__${uniq}"))
print("stderr = {}/{}.stderr".format(
os.path.dirname(sh_job_scripts), "${jname}__${uniq}"))
print("numCores=exact: {}".format(self.cores))
ids = QCG_MANAGER.submit(jobs)
# wait until submited jobs finish
QCG_MANAGER.wait4(ids)
print("\nAll new SWEEP dirs are finished...\n")
def submit(self, fn: Callable[..., Union[str, Tuple[str, Dict[str, Any]]]],
*args, **kwargs):
"""Submits a specific task to the QCG-PJ manager using template-based, executor-like interface.
Parameters
----------
fn : Callable
A callable that returns a tuple representing a task's template.
The first element of the tuple should be a string containing
a QCG-PilotJob task's description with placeholders
(identifiers preceded by $ symbol) and the second a dictionary
that assigns default values for selected placeholders.
*args: variable length list with dicts, optional
A set of dicts which contain parameters that will be used to substitute placeholders
defined in the template.
Note: *args overwrite defaults, but they are overwritten by **kwargs
**kwargs: arbitrary keyword arguments
A set of keyword arguments that will be used to substitute placeholders defined in
the template.
Note: **kwargs overwrite *args and defaults.
Returns
-------
QCGPJFuture
The QCGPJFuture object assigned with the submitted task
"""
template = fn()
if isinstance(template, tuple):
template_str = template[0]
defaults = template[1]
else:
template_str = template
defaults = {}
t = Template(textwrap.dedent(template_str))
substitutions = {}
for a in args:
if a is not None:
substitutions.update(a)
substitutions.update(kwargs)
td_str = t.substitute(defaults, **substitutions)
td = ast.literal_eval(td_str)
if 'env' not in td['execution']:
td['execution']['env'] = {}
td['execution']['env']['QCG_PM_EXEC_API_JOB_ID'] = '${jname}'
jobs = Jobs()
jobs.add_std(td)
jobs_ids = self._qcgpjm.submit(jobs)
return QCGPJFuture(jobs_ids, self._qcgpjm)
def test_slurmenv_api_iteration_simple():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
its = 2
jobs = Jobs(). \
add_std({ 'name': 'host',
'iteration': { 'stop': its },
'execution': { 'exec': 'hostname', 'args': [ '--fqdn' ], 'stdout': 'out' },
'resources': { 'numCores': { 'exact': 1 } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
assert jinfos
jinfo = jinfos['host']
print('jinfo: {}'.format(jinfo))
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0))
its = 2
jobs = Jobs(). \
add_std({ 'name': 'host2',
'iteration': { 'stop': its },
'execution': { 'exec': 'hostname', 'args': [ '--fqdn' ], 'stdout': 'out' },
'resources': { 'numCores': { 'exact': 1 } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos['host2']
print('jinfo: {}'.format(jinfo))
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0))
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format('host2', iteration),
job_it.wdir == tmpdir, job_it.total_cores == 1))
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
def test_slurmenv_api_submit_exceed_total_cores():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
jobs = Jobs(). \
add_std({ 'name': 'date',
'execution': { 'exec': '/bin/date' },
'resources': {
'numCores': { 'exact': resources.total_cores + 1 }
}})
with pytest.raises(ConnectionError, match=r".*Not enough resources.*"):
m.submit(jobs)
assert len(m.list()) == 0
jobs = Jobs(). \
add_std({ 'name': 'date',
'execution': { 'exec': '/bin/date' },
'resources': {
'numNodes': { 'exact': resources.total_nodes + 1 }
}})
with pytest.raises(ConnectionError, match=r".*Not enough resources.*"):
ids = m.submit(jobs)
assert len(m.list()) == 0
jobs = Jobs(). \
add_std({ 'name': 'date',
'execution': {
'exec': '/bin/date',
'stdout': 'std.out',
},
'resources': { 'numCores': { 'exact': resources.total_cores } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
assert jinfos['date'].total_cores == resources.total_cores
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
def test_resume_tracker_files(tmpdir):
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json', '--nodes', '4'], {'wdir': str(tmpdir)})
job_req = {
'name': 'host',
'execution': {
'exec': '/bin/date',
'stdout': 'out',
},
'resources': { 'numCores': { 'exact': 1 } }
}
jobs = Jobs().add_std(job_req)
submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
time.sleep(1)
aux_dir = find_single_aux_dir(str(tmpdir))
print(f'aux_dir content: {str(listdir(aux_dir))}')
assert all(exists(join(aux_dir, fname)) for fname in ['track.reqs', 'track.states']), \
f"missing tracker files in {aux_dir}: {str(listdir(aux_dir))}"
finally:
if m:
m.finish()
m.cleanup()
rmtree(tmpdir)
def test_slurmenv_api_submit_many_cores():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
jobs = Jobs(). \
add_std({ 'name': 'host',
'execution': {
'exec': '/bin/hostname',
'args': [ '--fqdn' ],
'stdout': 'out',
},
'resources': { 'numCores': { 'exact': resources.total_cores } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
# check working directories of job's inside working directory of service
assert tmpdir == jinfos['host'].wdir, str(jinfos['host'].wdir)
assert all((len(jinfos['host'].nodes) == resources.total_nodes,
jinfos['host'].total_cores == resources.total_cores)), str(jinfos['host'])
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
def test_slurmenv_api_submit_simple():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
jobs = Jobs().\
add_std({ 'name': 'host',
'execution': {
'exec': '/bin/hostname',
'args': [ '--fqdn' ],
'stdout': 'std.out',
'stderr': 'std.err'
}})
assert submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
def test_slurmenv_api_cancel_kill_nl():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
print(f'tmpdir: {tmpdir}')
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
iters=10
ids = m.submit(Jobs().
add(script='trap "" SIGTERM; sleep 30s', iteration=iters, stdout='sleep.out.${it}',
stderr='sleep.err.${it}', numCores=1)
)
jid = ids[0]
assert len(m.list()) == 1
list_jid = list(m.list().keys())[0]
assert list_jid == jid
# wait for job to start executing
sleep(2)
m.cancel([jid])
# wait for SIGTERM job cancel
sleep(2)
jinfos = m.info_parsed(ids)
assert all((len(jinfos) == 1, jid in jinfos, jinfos[jid].status == 'QUEUED'))
# wait for SIGKILL job cancel (~ExecutionJob.SIG_KILL_TIMEOUT)
sleep(ExecutionJob.SIG_KILL_TIMEOUT)
jinfos = m.info_parsed(ids, withChilds=True)
assert all((len(jinfos) == 1, jid in jinfos, jinfos[jid].status == 'CANCELED'))
# the canceled iterations are included in 'failed' entry in job statistics
# the cancel status is presented in 'childs/state' entry
assert all((jinfos[jid].iterations, jinfos[jid].iterations.get('start', -1) == 0,
jinfos[jid].iterations.get('stop', 0) == iters, jinfos[jid].iterations.get('total', 0) == iters,
jinfos[jid].iterations.get('finished', 0) == iters, jinfos[jid].iterations.get('failed', -1) == iters))
assert len(jinfos[jid].childs) == iters
for iteration in range(iters):
job_it = jinfos[jid].childs[iteration]
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jid, iteration),
job_it.status == 'CANCELED')), str(job_it)
m.remove(jid)
finally:
m.finish()
m.cleanup()
def test_local_manager_submit_simple(tmpdir):
cores = 4
# switch on debugging (by default in api.log file)
m = LocalManager(['--wd', str(tmpdir), '--nodes',
str(cores)], {'wdir': str(tmpdir)})
try:
res = m.resources()
assert all(
('total_nodes' in res, 'total_cores'
in res, res['total_nodes'] == 1, res['total_cores'] == cores))
ids = m.submit(Jobs().add(name='host',
exec='/bin/hostname',
args=['--fqdn'],
stdout='host.stdout').add(
name='date',
exec='/bin/date',
stdout='date.stdout',
numCores={'exact': 2}))
assert len(m.list()) == 2
m.wait4(ids)
jinfos = m.info(ids)
assert all(
('jobs' in jinfos, len(jinfos['jobs'].keys()) == 2, 'host'
in jinfos['jobs'], 'date' in jinfos['jobs'],
jinfos['jobs']['host'].get('data', {}).get('status',
'') == 'SUCCEED',
jinfos['jobs']['date'].get('data', {}).get('status',
'') == 'SUCCEED'))
aux_dir = find_single_aux_dir(str(tmpdir))
assert all(
(exists(tmpdir.join('.qcgpjm-client', 'api.log')),
exists(join(aux_dir,
'service.log')), exists(tmpdir.join('host.stdout')),
exists(tmpdir.join('date.stdout'))))
finally:
m.finish()
# m.stopManager()
m.cleanup()
def test_slurmenv_api_std_streams_many_cores():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
jobs = Jobs(). \
add_std({ 'name': 'host',
'execution': {
'exec': 'cat',
'stdin': '/etc/system-release',
'stdout': 'out',
'stderr': 'err'
},
'resources': {
'numCores': { 'exact': 2 }
}
})
assert submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
assert all((exists(join(tmpdir, 'out')), exists(join(tmpdir, 'err'))))
with open(join(tmpdir, 'out'), 'rt') as out_f:
out = out_f.read()
with open(join('/etc/system-release'), 'rt') as sr_f:
system_release = sr_f.read()
assert system_release in out
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
def test_resume_simple(tmpdir):
try:
ncores = 4
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json', '--nodes', str(ncores)],
{'wdir': str(tmpdir)})
its = 10
job_req = {
'name': 'sleep',
'execution': {
'exec': '/bin/sleep',
'args': [ '4s' ],
'stdout': 'out',
},
'iteration': { 'stop': its },
'resources': { 'numCores': { 'exact': 1 } }
}
jobs = Jobs().add_std(job_req)
job_ids = m.submit(jobs)
# because job iterations executes in order, after finish of 4th iteration, the three previous should also finish
m.wait4('sleep:3')
jinfos = m.info_parsed(job_ids, withChilds=True)
assert jinfos
jinfo = jinfos['sleep']
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == ncores, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
exp_status = ['SUCCEED']
if iteration > 3:
exp_status = ['EXECUTING', 'SCHEDULED', 'QUEUED']
assert all((job_it.iteration == iteration,
job_it.name == '{}:{}'.format('sleep', iteration),
job_it.status in exp_status)),\
f"{job_it.iteration} != {iteration}, {job_it.name} != {'{}:{}'.format('sleep', iteration)}, {job_it.status} != {exp_status}"
# kill process
m.kill_manager_process()
m.cleanup()
ncores = 4
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json', '--nodes', str(ncores),
'--resume', tmpdir],
{'wdir': str(tmpdir)})
m.wait4all()
jinfos = m.info_parsed(job_ids, withChilds=True)
assert jinfos
jinfo = jinfos['sleep']
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
assert all((job_it.iteration == iteration,
job_it.name == '{}:{}'.format('sleep', iteration),
job_it.status == 'SUCCEED')), \
f"{job_it.iteration} != {iteration}, {job_it.name} != {'{}:{}'.format('sleep', iteration)}, {job_it.status} != SUCCEED"
finally:
if m:
m.finish()
m.cleanup()
def test_slurmenv_api_iteration_node_scheduling():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
# TODO: it's hard to write comprehensive iteration scheduling node tests on only two nodes (in slurm's \
# development docker)
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
# in that case the 'split-into' is default the number of iterations
# so total available resources should be splited into two partitions and each of the
# iteration should run on its own partition
jname = 'host'
its = 2
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out_${it}', 'stderr': 'err_${it}' },
'resources': { 'numCores': { 'exact': resources.nodes[0].total },
'numNodes': { 'min': 1,
'scheduler': { 'name': 'split-into' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores == resources.nodes[0].total, len(job_it.nodes) == 1)), str(job_it)
# all iterations has been scheduled across all nodes
assert sum([ len(child.nodes) for child in jinfo.childs ]) == resources.total_nodes
# the iterations should execute on different node
assert list(jinfo.childs[0].nodes)[0] != list(jinfo.childs[1].nodes)[0]
# we explicity specify the 'split-into' parameter to 2, behavior should be the same as in the
# previous example
jname = 'host2'
its = 2
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'exact': resources.nodes[0].total },
'numNodes': { 'min': 1,
'scheduler': { 'name': 'split-into', 'params': { 'parts': 2 } } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores == resources.nodes[0].total, len(job_it.nodes) == 1)), str(job_it)
# all iterations has been scheduled across all nodes
assert sum([ len(child.nodes) for child in jinfo.childs ]) == resources.total_nodes
# the iterations should execute on different node
assert list(jinfo.childs[0].nodes)[0] != list(jinfo.childs[1].nodes)[0]
# the 'maximum-iters' scheduler is trying to launch as many iterations in the same time on all available
# resources
jname = 'host3'
its = 4
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'exact': resources.nodes[0].total },
'numNodes': { 'min': 1,
'scheduler': { 'name': 'maximum-iters' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores == resources.nodes[0].total, len(job_it.nodes) == 1)), str(job_it)
assert sum([len(child.nodes) for child in jinfo.childs]) == its
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
def test_slurmenv_api_iteration_core_scheduling():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
# in that case the 'split-into' is default the number of iterations
# so total available resources should be splited into two partitions and each of the
# iteration should run on its own partition
jname = 'host'
its = 2
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'hostname', 'args': [ '--fqdn' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'split-into' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
# all iterations has been scheduled across all resources
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores
assert all(child.total_cores == resources.total_cores / its for child in jinfo.childs)
# we explicity specify the 'split-into' parameter to 2, behavior should be the same as in the
# previous example
jname = 'host2'
its = 2
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'hostname', 'args': [ '--fqdn' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'split-into', 'params': { 'parts': 2 } } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
# all iterations has been scheduled across all resources
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores
assert all(child.total_cores == resources.total_cores / 2 for child in jinfo.childs)
# we explicity specify the 'split-into' parameter to 4, the two iterations should be sheduled
# on half of the available resources
jname = 'host3'
its = 2
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'hostname', 'args': [ '--fqdn' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'split-into', 'params': { 'parts': 4 } } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
# all iterations has been scheduled across all resources
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores / 2
assert all(child.total_cores == resources.total_cores / 4 for child in jinfo.childs)
# we explicity specify the 'split-into' parameter to 2, but the number of iterations is larger than
# available partitions in the same time, so they should be executed serially (by parts)
jname = 'host4'
its = 10
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'hostname', 'args': [ '--fqdn' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'split-into', 'params': { 'parts': 2 } } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
assert all(child.total_cores == resources.total_cores / 2 for child in jinfo.childs)
# the 'maximum-iters' scheduler is trying to launch as many iterations in the same time on all available
# resources
jname = 'host5'
its = 2
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'maximum-iters' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores
# the 'maximum-iters' scheduler is trying to launch as many iterations in the same time on all available
# resources
jname = 'host6'
its = resources.total_cores
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'maximum-iters' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores
# in case where number of iterations exceeds the number of available resources, the 'maximum-iters' schedulers
# splits iterations into 'steps' minimizing this number, and allocates as many resources as possible for each
# iteration inside 'step'
jname = 'host7'
its = resources.total_cores
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'maximum-iters' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
assert (child.total_cores == 1 for child in jinfo.childs)
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores
# in case where number of iterations exceeds the number of available resources, the 'maximum-iters' schedulers
# splits iterations into 'steps' minimizing this number, and allocates as many resources as possible for each
# iteration inside 'step'
jname = 'host8'
its = resources.total_cores * 2
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'maximum-iters' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1, job_it.total_cores < resources.total_cores)), str(job_it)
assert (child.total_cores == 1 for child in jinfo.childs)
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores * 2
# in case where number of iterations exceeds the number of available resources, the 'maximum-iters' schedulers
# splits iterations into 'steps' minimizing this number, and allocates as many resources as possible for each
# iteration inside 'step'
jname = 'host9'
its = resources.total_cores + 1
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': 1,
'scheduler': { 'name': 'maximum-iters' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores >= 1)), str(job_it)
assert (child.total_cores == 1 for child in jinfo.childs)
# because all iterations will be splited in two 'steps' and in each step the iterations that has been assigned
# for the step should usage maximum available resources
assert sum([ child.total_cores for child in jinfo.childs ]) == resources.total_cores * 2
# in this case where two iterations can't fit at once on resources, all the iterations should be scheduled
# serially on all available resources
jname = 'host10'
its = resources.total_nodes
jobs = Jobs(). \
add_std({ 'name': jname,
'iteration': { 'stop': its },
'execution': { 'exec': 'sleep', 'args': [ '2s' ], 'stdout': 'out' },
'resources': { 'numCores': { 'min': resources.total_cores - 1,
'scheduler': { 'name': 'maximum-iters' } } }
})
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED', withChilds=True)
assert jinfos
jinfo = jinfos[jname]
assert all((jinfo.iterations, jinfo.iterations.get('start', -1) == 0,
jinfo.iterations.get('stop', 0) == its, jinfo.iterations.get('total', 0) == its,
jinfo.iterations.get('finished', 0) == its, jinfo.iterations.get('failed', -1) == 0)), str(jinfo)
assert len(jinfo.childs) == its
for iteration in range(its):
job_it = jinfo.childs[iteration]
print('job iteration {}: {}'.format(iteration, str(job_it)))
assert all((job_it.iteration == iteration, job_it.name == '{}:{}'.format(jname, iteration),
job_it.total_cores == resources.total_cores, len(job_it.nodes) == resources.total_nodes)),\
str(job_it)
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
def test_slurmenv_api_submit_resource_ranges():
if not in_slurm_allocation() or get_num_slurm_nodes() < 2:
pytest.skip('test not run in slurm allocation or allocation is smaller than 2 nodes')
resources, allocation = get_slurm_resources_binded()
set_pythonpath_to_qcg_module()
tmpdir = str(tempfile.mkdtemp(dir=SHARED_PATH))
try:
m = LocalManager(['--log', 'debug', '--wd', tmpdir, '--report-format', 'json'], {'wdir': str(tmpdir)})
jobs = Jobs(). \
add_std({ 'name': 'host',
'execution': {
'exec': '/bin/hostname',
'args': [ '--fqdn' ],
'stdout': 'out',
},
'resources': { 'numCores': { 'min': 1 } }
})
# job should faile because of missing 'max' parameter
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'FAILED')
jinfo = jinfos['host']
assert "Both core's range boundaries (min, max) must be defined" in jinfo.messages, str(jinfo)
jobs = Jobs(). \
add_std({ 'name': 'host2',
'execution': {
'exec': '/bin/hostname',
'args': [ '--fqdn' ],
'stdout': 'out',
},
'resources': {
'numNodes': { 'exact': 1 },
'numCores': { 'min': 1, 'max': resources.nodes[0].total + 1 } }
})
# job should run on single node (the first free) with all available cores
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
jinfo = jinfos['host2']
assert all((len(jinfo.nodes) == 1, jinfo.total_cores == resources.nodes[0].total)), str(jinfo)
jobs = Jobs(). \
add_std({ 'name': 'host3',
'execution': {
'exec': '/bin/hostname',
'args': [ '--fqdn' ],
'stdout': 'out',
},
'resources': {
'numCores': { 'min': 1, 'max': resources.nodes[0].total + 1 } }
})
# job should run on at least two nodes with total maximum given cores
jinfos = submit_2_manager_and_wait_4_info(m, jobs, 'SUCCEED')
jinfo = jinfos['host3']
assert all((len(jinfo.nodes) == 2, jinfo.total_cores == resources.nodes[0].total + 1)), str(jinfo)
finally:
if m:
m.finish()
# m.stopManager()
m.cleanup()
rmtree(tmpdir)
######################################################################
config PilotJob
######################################################################
"""
from qcg.pilotjob.api.manager import Manager
from qcg.pilotjob.api.manager import LocalManager
from qcg.pilotjob.api.job import Jobs
# m = LocalManager(cfg={'log_level': 'DEBUG'}, server_args=['--log', 'debug'])
m = LocalManager()
# get available resources
print("\n\navailable resources:\n%s\n" % str(m.resources()))
# submit jobs and save their names in 'ids' list
jobs = Jobs()
print("Start Adding jobs . . .\n\n")
INSTANCE_INDEX = 0
for i in range(NUM_INSTANCES):
for SUBMODEL in ['macro', 'micro']:
cmd = '%s run_mscale.py --submodel %s --data_dir=%s --instance_index %d --coupling_type %s --num_instances %d --weather_coupling %s' % (
PYTHON_CMD, SUBMODEL, DATA_DIR, INSTANCE_INDEX, COUPLING_TYPE,
NUM_INSTANCES, WEATHER_COUPLING)
print("\tAdd job with cmd = %s" % (cmd))
TaskID = 'TaskID%d_%s' % (INSTANCE_INDEX + 1, SUBMODEL)
stderr = 'log_task/%s_${jname}__${uniq}.stderr' % (TaskID)
stdout = 'log_task/%s_${jname}__${uniq}.stdout' % (TaskID)
from qcg.pilotjob.api.manager import Manager
from qcg.pilotjob.api.manager import LocalManager
from qcg.pilotjob.api.job import Jobs
m = LocalManager(cfg={'log_level': 'DEBUG'}, server_args=['--log', 'debug'])
# m = Manager(cfg={'log_level': 'DEBUG'}, server_args=['--log', 'debug'])
# get available resources
print("available resources:\n%s\n" % str(m.resources()))
# submit jobs and save their names in 'ids' list
jobs = Jobs()
$submitted_jobs_list
ids = m.submit(jobs)
# wait until submited jobs finish
m.wait4(ids)
# get detailed information about submited and finished jobs
print("jobs details:\n%s\n" % str(m.info(ids)))
m.finish()
m.kill_manager_process()
m.cleanup()