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_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()
class QCGPJExecutor(Executor):
"""QCG-PilotJob Executor. It provides simplified interface for common uses of QCG-PilotJob
Parameters
----------
wd : str, optional
Working directory where QCG-PilotJob manager should be started, by default it is
a current directory
resources : str, optional
The resources to use. If specified forces usage of Local mode of QCG-PilotJob Manager.
The format is compliant with the NODES format of QCG-PilotJob, i.e.:
[node_name:]cores_on_node[,node_name2:cores_on_node][,...].
Eg. to define 4 cores on an unnamed node use `resources="4"`,
to define 2 nodes: node_1 with 2 cores and node_2 with 3 cores, use `resources="node_1:2,node_2:3"`
reserve_core : bool, optional
If True reserves a core for QCG-PilotJob Manager instance,
by default QCG-PilotJob Manager shares a core with computing tasks
Parameters.
enable_rt_stats : bool, optional
If True, QCG-PilotJob Manager will collect its runtime statistics
wrapper_rt_stats : str, optional
The path to the QCG-PilotJob Manager tasks wrapper program used for collection of statistics
log_level : str, optional
Logging level for QCG-PilotJob Manager (for both service and client part).
other_args : optional
Optional list of additional arguments for initialisation of QCG-PilotJob Manager
Returns
-------
None
"""
def __init__(self,
wd=".",
resources=None,
reserve_core=False,
enable_rt_stats=False,
wrapper_rt_stats=None,
log_level='info',
*other_args):
self.finished = False
# ---- QCG PILOT JOB INITIALISATION ---
# Establish logging levels
service_log_level, client_log_level = self._setup_qcgpj_logging(
log_level)
# Prepare input arguments for QCG-PJM
args = ['--log', service_log_level, '--wd', wd]
if resources:
args.append('--nodes')
args.append(str(resources))
if reserve_core:
args.append('--system-core')
if enable_rt_stats:
args.append('--enable-rt-stats')
if wrapper_rt_stats:
args.append('--wrapper-rt-stats')
args.append(wrapper_rt_stats)
if other_args:
args.append(other_args)
client_conf = {
'log_file': wd + '/api.log',
'log_level': client_log_level
}
_logger.info(f'Starting QCG-PJ Manager with arguments: {args}')
# create QCGPJ Manager (service part)
self._qcgpjm = LocalManager(args, client_conf)
def __enter__(self):
return self
def __exit__(self, exc_type, exc_value, traceback):
self.shutdown()
def shutdown(self, wait=True):
"""Shutdowns the QCG-PJ manager service. If it is already closed, the method has no effect.
"""
if not self.finished:
self._qcgpjm.finish()
self.finished = True
else:
pass
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)
@property
def qcgpj_manager(self):
"""Returns current QCG-PilotJob manager instance
"""
return self._qcgpjm
@staticmethod
def _setup_qcgpj_logging(log_level):
log_level = log_level.upper()
try:
service_log_level = ServiceLogLevel[log_level].value
except KeyError:
service_log_level = ServiceLogLevel.DEBUG.value
try:
client_log_level = ClientLogLevel[log_level].value
except KeyError:
client_log_level = ClientLogLevel.DEBUG.value
return service_log_level, client_log_level
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()
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)
jobs.add(name=TaskID,
exec='bash',
args=['-c', cmd],
stdout=stdout,
stderr=stderr,
numCores={'exact': INSTANCE_CORES},
model='default')
INSTANCE_INDEX = INSTANCE_INDEX + 1
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()
