def test_ens_compose_restore(simulation_compiled, job_restart, scheduler,
tmpdir):
sim = simulation_compiled
ens = EnsembleSimulation()
ens.add(job_restart)
ens.add([sim, sim])
# Do not keep the members in memory
ens_disk = copy.deepcopy(ens)
compose_dir = pathlib.Path(tmpdir).joinpath('ensemble_disk')
os.mkdir(str(compose_dir))
os.chdir(str(compose_dir))
ens_disk.compose()
ens_disk_run_success = ens_disk.run()
assert ens_disk_run_success == 0
# Keep the members in memory
compose_dir = pathlib.Path(tmpdir).joinpath('ensemble_memory')
os.mkdir(str(compose_dir))
os.chdir(str(compose_dir))
ens.compose(rm_members_from_memory=False)
ens_run_success = ens.run()
assert ens_run_success == 0
# Check that the members are all now simply pathlib objects
assert all([type(mm) is str for mm in ens_disk.members])
ens_disk.restore_members()
# Since the ens_disk has data from the run. Collect data from the run:
ens.collect(output=False)
# The members are not restored, the simultaion sub objects are:
ens.restore_members()
# These will never be the same for two different runs.
for mem, dsk in zip(ens.members, ens_disk.members):
del mem.jobs[0].job_end_time, mem.jobs[0].job_start_time
del dsk.jobs[0].job_end_time, dsk.jobs[0].job_start_time
from pprint import pprint
assert deepdiff.DeepDiff(ens, ens_disk) == {}
def test_ens_parallel_compose(simulation_compiled, job, scheduler, tmpdir):
sim = simulation_compiled
ens = EnsembleSimulation()
ens.add(job)
ens.add(scheduler)
with pytest.raises(Exception) as e_info:
ens.compose()
ens.add([sim, sim])
# Make a copy where we keep the members in memory for checking.
ens_check_members = copy.deepcopy(ens)
compose_dir = pathlib.Path(tmpdir).joinpath('ensemble_compose')
os.mkdir(str(compose_dir))
os.chdir(str(compose_dir))
ens.compose()
# TODO(JLM): test the run here
# ens_run_success = ens.run()
# assert ens_run_success
ens.pickle(
str(pathlib.Path(tmpdir) / 'ensemble_compose/WrfHydroEnsSim.pkl'))
# The ensemble-in-memory version for checking the members.
compose_dir = pathlib.Path(tmpdir).joinpath(
'ensemble_compose_check_members')
os.mkdir(str(compose_dir))
os.chdir(str(compose_dir))
ens_check_members.compose(rm_members_from_memory=False)
# The job gets heavily modified on compose.
answer = {
'_entry_cmd': 'bogus entry cmd',
'_exe_cmd': './wrf_hydro.exe',
'_exit_cmd': 'bogus exit cmd',
'_hrldas_namelist': {
'noahlsm_offline': {
'btr_option':
1,
'canopy_stomatal_resistance_option':
1,
'hrldas_setup_file':
'./NWM/DOMAIN/wrfinput_d01.nc',
'indir':
'./FORCING',
'restart_filename_requested':
'./NWM/RESTART/RESTART.2011082600_DOMAIN1'
},
'wrf_hydro_offline': {
'forc_typ': 1
}
},
'_hrldas_times': {
'noahlsm_offline': {
'khour': 282480,
'restart_frequency_hours': 1,
'output_timestep': 3600,
'restart_filename_requested': None,
'start_day': 14,
'start_hour': 0,
'start_min': 0,
'start_month': 10,
'start_year': 1984
}
},
'_hydro_namelist': {
'hydro_nlist': {
'aggfactrt': 4,
'channel_option': 2,
'chanobs_domain': 0,
'chanrtswcrt': 1,
'chrtout_domain': 1,
'geo_static_flnm': './NWM/DOMAIN/geo_em.d01.nc',
'restart_file':
'./NWM/RESTART/HYDRO_RST.2011-08-26_00:00_DOMAIN1',
'udmp_opt': 1
},
'nudging_nlist': {
'maxagepairsbiaspersist':
3,
'minnumpairsbiaspersist':
1,
'nudginglastobsfile':
'./NWM/RESTART/nudgingLastObs.2011-08-26_00:00:00.nc'
}
},
'_hydro_times': {
'hydro_nlist': {
'restart_file': None,
'rst_dt': 60,
'out_dt': 60
},
'nudging_nlist': {
'nudginglastobsfile': None
}
},
'_job_end_time': None,
'_job_start_time': None,
'_job_submission_time': None,
'_model_end_time': pandas.Timestamp('2017-01-04 00:00:00'),
'_model_start_time': pandas.Timestamp('1984-10-14 00:00:00'),
'exit_status': None,
'job_id': 'test_job_1',
'restart_freq_hr': 1,
'output_freq_hr': 1,
'restart': False
}
# For the ensemble where the compse retains the members...
# This fails:
# deepdiff.DeepDiff(answer, ens.members[0].jobs[0].__dict__)
# Instead, iterate on keys to "declass":
for kk in ens_check_members.members[0].jobs[0].__dict__.keys():
assert ens_check_members.members[0].jobs[0].__dict__[kk] == answer[kk]
# Check the scheduler too
assert ens_check_members.members[
0].scheduler.__dict__ == scheduler.__dict__
# For the ensemble where the compse removes the members...
# Check that the members are all now simply pathlib objects
assert all([type(mm) is str for mm in ens.members])
# The tmpdir gets nuked after the test... ?
# Test the member pickle size in terms of load speed.
# Note that the deletion of the model, domain, and output objects are
# done for the members regardless of not removing the members
# from memory (currently).
os.chdir(str(pathlib.Path(tmpdir) / 'ensemble_compose/member_000'))
time_taken = timeit.timeit(
setup='import pickle',
stmt='pickle.load(open("WrfHydroSim.pkl","rb"))',
number=10000)
# If your system is busy, this could take longer... and spuriously fail the test.
# Notes(JLM): OSX spinning disk is < .5, cheyenne scratch is < .8
assert time_taken < .8
# Test the ensemble pickle size in terms of load speed.
os.chdir(str(pathlib.Path(tmpdir) / 'ensemble_compose/'))
time_taken = timeit.timeit(
setup='import pickle',
stmt='pickle.load(open("WrfHydroEnsSim.pkl","rb"))',
number=10000)
# If your system is busy, this could take longer...
# Notes(JLM): .7 seems to work on OSX spinning disk and chyenne scratch.
assert time_taken < .7
