def test_env(env):
# Force asserts to show the full file when failures occur.
# Useful to debug errors that arise.
# Must import the data into the project.
with signac.TemporaryProject(name=gen.PROJECT_NAME) as p:
fp = gen.get_masked_flowproject(p)
fp.import_from(origin=gen.ARCHIVE_DIR)
jobs = fp.find_jobs(dict(environment=_env_name(env)))
if not len(jobs):
raise RuntimeError(
"No reference data for environment {}!".format(_env_name(env))
)
reference = []
generated = []
for job in jobs:
parameters = job.sp.parameters()
if 'bundle' in parameters:
bundle = parameters.pop('bundle')
tmp_out = io.TextIOWrapper(
io.BytesIO(), sys.stdout.encoding)
with open(os.devnull, 'w') as devnull:
with redirect_stderr(devnull):
with redirect_stdout(tmp_out):
fp.submit(
env=env, jobs=[job], names=bundle, pretend=True,
force=True, bundle_size=len(bundle), **parameters)
tmp_out.seek(0)
msg = "---------- Bundled submission of job {}".format(job)
generated.extend([msg] + tmp_out.read().splitlines())
with open(job.fn('script_{}.sh'.format('_'.join(bundle)))) as file:
reference.extend([msg] + file.read().splitlines())
else:
for op in fp.operations:
if 'partition' in parameters:
# Don't try to submit GPU operations to CPU partitions
# and vice versa. We should be able to relax this
# requirement if we make our error checking more
# consistent.
if operator.xor(
'gpu' in parameters['partition'].lower(),
'gpu' in op.lower()):
continue
tmp_out = io.TextIOWrapper(
io.BytesIO(), sys.stdout.encoding)
with open(os.devnull, 'w') as devnull:
with redirect_stderr(devnull):
with redirect_stdout(tmp_out):
fp.submit(
env=env, jobs=[job],
names=[op], pretend=True, force=True, **parameters)
tmp_out.seek(0)
msg = "---------- Submission of operation {} for job {}.".format(op, job)
generated.extend([msg] + tmp_out.read().splitlines())
with open(job.fn('script_{}.sh'.format(op))) as file:
reference.extend([msg] + file.read().splitlines())
assert '\n'.join(reference) == '\n'.join(generated)
def test_run_operations_implicit_argument(self):
project = self.mock_project()
for job in project:
ops = project.next_operations(job)
with suspend_logging():
with redirect_stderr(StringIO()):
project.run(ops)
for job in project:
self.assertIn('said_hello', list(project.labels(job)))
def test_print_status(self):
project = self.mock_project()
for job in project:
list(project.classify(job))
self.assertEqual(project.next_operation(job).name, 'a_op')
self.assertEqual(project.next_operation(job).job, job)
fd = StringIO()
with redirect_stderr(StringIO()):
with redirect_stdout(StringIO()):
project.print_status(file=fd, err=fd)
def test_submit_limited(self):
env = get_environment()
sched = env.scheduler_type()
sched.reset()
project = self.mock_project()
self.assertEqual(len(list(sched.jobs())), 0)
with suspend_logging():
with redirect_stderr(StringIO()):
project.submit(env, num=1)
self.assertEqual(len(list(sched.jobs())), 1)
project.submit(env, num=1)
self.assertEqual(len(list(sched.jobs())), 2)
def test_bundles(self):
env = get_environment()
sched = env.scheduler_type()
sched.reset()
project = self.mock_project()
self.assertEqual(len(list(sched.jobs())), 0)
with suspend_logging():
with redirect_stderr(StringIO()):
project.submit(bundle_size=2, num=2)
self.assertEqual(len(list(sched.jobs())), 1)
project.submit(bundle_size=2, num=4)
self.assertEqual(len(list(sched.jobs())), 3)
sched.reset()
project._fetch_scheduler_status(file=StringIO())
project.submit(bundle_size=0)
self.assertEqual(len(list(sched.jobs())), 1)
def test_submit_operations(self):
env = get_environment()
sched = env.scheduler_type()
sched.reset()
project = self.mock_project()
operations = []
for job in project:
operations.extend(project.next_operations(job))
self.assertEqual(len(list(sched.jobs())), 0)
cluster_job_id = project._store_bundled(operations)
with suspend_logging():
with redirect_stderr(StringIO()):
project.submit_operations(_id=cluster_job_id,
env=env,
operations=operations)
self.assertEqual(len(list(sched.jobs())), 1)
sched.reset()
def test_run(self):
project = self.mock_project()
with redirect_stderr(StringIO()):
project.run()
for job in project:
self.assertIn('said_hello', list(project.labels(job)))
def test_env(env, monkeypatch):
monkeypatch.setattr(flow.FlowProject, "_store_bundled", gen._store_bundled)
# Force asserts to show the full file when failures occur.
# Useful to debug errors that arise.
# Must import the data into the project.
with signac.TemporaryProject(name=gen.PROJECT_NAME) as p:
fp = gen.get_masked_flowproject(p)
# Here we set the appropriate executable for all the operations. This
# is necessary as otherwise the default executable between submitting
# and running could look different depending on the environment.
executable = "/usr/local/bin/python"
for group in fp.groups.values():
for op_key in group.operations:
if op_key in group.operation_directives:
group.operation_directives[op_key][
"executable"] = executable
fp.import_from(origin=gen.ARCHIVE_DIR)
jobs = fp.find_jobs(dict(environment=_env_name(env)))
if not len(jobs):
raise RuntimeError("No reference data for environment {}!".format(
_env_name(env)))
reference = []
generated = []
for job in jobs:
parameters = job.sp.parameters()
if "bundle" in parameters:
bundle = parameters.pop("bundle")
tmp_out = io.TextIOWrapper(io.BytesIO(), sys.stdout.encoding)
with open(os.devnull, "w") as devnull:
with redirect_stderr(devnull):
with redirect_stdout(tmp_out):
fp.submit(
env=env,
jobs=[job],
names=bundle,
pretend=True,
force=True,
bundle_size=len(bundle),
**parameters,
)
tmp_out.seek(0)
msg = f"---------- Bundled submission of job {job}"
generated.extend([msg] + tmp_out.read().splitlines())
with open(job.fn("script_{}.sh".format(
"_".join(bundle)))) as file:
reference.extend([msg] + file.read().splitlines())
else:
for op in {**fp.operations, **fp.groups}:
if "partition" in parameters:
# Don't try to submit GPU operations to CPU partitions
# and vice versa. We should be able to relax this
# requirement if we make our error checking more
# consistent.
if operator.xor(
"gpu" in parameters["partition"].lower(),
"gpu" in op.lower(),
):
continue
tmp_out = io.TextIOWrapper(io.BytesIO(),
sys.stdout.encoding)
with open(os.devnull, "w") as devnull:
with redirect_stderr(devnull):
with redirect_stdout(tmp_out):
fp.submit(
env=env,
jobs=[job],
names=[op],
pretend=True,
force=True,
**parameters,
)
tmp_out.seek(0)
msg = f"---------- Submission of operation {op} for job {job}."
generated.extend([msg] + tmp_out.read().splitlines())
with open(job.fn(f"script_{op}.sh")) as file:
reference.extend([msg] + file.read().splitlines())
assert "\n".join(reference) == "\n".join(generated)
def test_env(env, monkeypatch):
monkeypatch.setattr(flow.FlowProject, "_store_bundled", gen._store_bundled)
# We need to set the scheduler manually. The FakeScheduler is used for two
# reasons. First, the FakeScheduler prints scripts to screen on submission
# and we can capture that output. Second, the FakeScheduler won't try to
# call any cluster executable (e.g. squeue) associated with the real
# schedulers used on supported clusters. Otherwise submission would fail
# when attempting to determine what jobs already exist on the scheduler.
monkeypatch.setattr(env, "scheduler_type", FakeScheduler)
# Force asserts to show the full file when failures occur.
# Useful to debug errors that arise.
# Must import the data into the project.
with signac.TemporaryProject(name=gen.PROJECT_NAME) as p:
with gen.get_masked_flowproject(p, environment=env) as fp:
# Here we set the appropriate executable for all the operations. This
# is necessary as otherwise the default executable between submitting
# and running could look different depending on the environment.
for group in fp.groups.values():
for op_key in group.operations:
if op_key in group.operation_directives:
monkeypatch.setitem(
group.operation_directives[op_key],
"executable",
gen.MOCK_EXECUTABLE,
)
fp.import_from(origin=gen.ARCHIVE_DIR)
jobs = fp.find_jobs(dict(environment=_env_name(env)))
if not len(jobs):
raise RuntimeError(
"No reference data for environment {}!".format(
_env_name(env)))
reference = []
generated = []
for job in jobs:
parameters = job.sp.parameters()
if "bundle" in parameters:
bundle = parameters.pop("bundle")
tmp_out = io.TextIOWrapper(io.BytesIO(),
sys.stdout.encoding)
with open(os.devnull, "w") as devnull:
with redirect_stderr(devnull):
with redirect_stdout(tmp_out):
fp.submit(
jobs=[job],
names=bundle,
pretend=True,
force=True,
bundle_size=len(bundle),
**parameters,
)
tmp_out.seek(0)
msg = f"---------- Bundled submission of job {job}"
generated.extend([msg] + tmp_out.read().splitlines())
with open(job.fn("script_{}.sh".format(
"_".join(bundle)))) as file:
reference.extend([msg] + file.read().splitlines())
else:
for op in {**fp.operations, **fp.groups}:
if "partition" in parameters:
# Don't try to submit GPU operations to CPU partitions
# and vice versa. We should be able to relax this
# requirement if we make our error checking more
# consistent.
if operator.xor(
"gpu" in parameters["partition"].lower(),
"gpu" in op.lower(),
):
continue
tmp_out = io.TextIOWrapper(io.BytesIO(),
sys.stdout.encoding)
with open(os.devnull, "w") as devnull:
with redirect_stderr(devnull):
with redirect_stdout(tmp_out):
fp.submit(
jobs=[job],
names=[op],
pretend=True,
force=True,
**parameters,
)
tmp_out.seek(0)
msg = f"---------- Submission of operation {op} for job {job}."
generated.extend([msg] + tmp_out.read().splitlines())
with open(job.fn(f"script_{op}.sh")) as file:
reference.extend([msg] + file.read().splitlines())
assert "\n".join(generated) == "\n".join(reference)