def test_local_manager_wait4all(tmpdir):
cores = 4
# switch on debugging (by default in api.log file)
m = LocalManager(['--wd', str(tmpdir), '--nodes',
str(cores)], {'wdir': str(tmpdir)})
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.wait4all()
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'))))
m.finish()
# m.stopManager()
m.cleanup()
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()
