def test__issue_213(self):
""" Test if we can wait for different pilot states.
"""
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
c = radical.pilot.Context('ssh')
c.user_id = self.test_ssh_uid
c.user_key = self.test_ssh_key
session.add_context(c)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = self.test_resource
cpd.cores = self.test_cores
cpd.runtime = self.test_runtime
cpd.sandbox = self.test_workdir
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
pilot.wait(state=radical.pilot.PMGR_ACTIVE, timeout=self.test_timeout*60)
assert pilot.state == radical.pilot.PMGR_ACTIVE
# the pilot should finish after it has reached run_time
pilot.wait(timeout=self.test_timeout*60)
assert pilot.state == radical.pilot.DONE
session.close()
def test__issue_213(self):
""" Test if we can wait for different pilot states.
"""
session = radical.pilot.Session(database_url=DBURL,
database_name=DBNAME)
c = radical.pilot.Context('ssh')
c.user_id = self.test_ssh_uid
c.user_key = self.test_ssh_key
session.add_context(c)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = self.test_resource
cpd.cores = self.test_cores
cpd.runtime = self.test_runtime
cpd.sandbox = self.test_workdir
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
pilot.wait(state=radical.pilot.PMGR_ACTIVE,
timeout=self.test_timeout * 60)
assert pilot.state == radical.pilot.PMGR_ACTIVE
# the pilot should finish after it has reached run_time
pilot.wait(timeout=self.test_timeout * 60)
assert pilot.state == radical.pilot.DONE
session.close()
def test__add_resource_config_2(self):
""" Test if we can wait for different pilot states.
"""
session = radical.pilot.Session(database_url=DBURL)
rc = radical.pilot.ResourceConfig("mylocalhost")
rc.task_launch_method = "LOCAL"
rc.mpi_launch_method = "MPIRUN"
rc.job_manager_endpoint = "fork://localhost"
rc.filesystem_endpoint = "file://localhost/"
rc.bootstrapper = "default_bootstrapper.sh"
rc.pilot_agent = "radical-pilot-agent-multicore.py"
pm = radical.pilot.PilotManager(session=session)
session.add_resource_config(rc)
pd = radical.pilot.ComputePilotDescription()
pd.resource = "mylocalhost"
pd.cores = 1
pd.runtime = 1 # minutes
pd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
pd.cleanup = True
pilot = pm.submit_pilots(pd)
pilot.wait(timeout=5 * 60)
pilot.cancel()
session.close()
def test__add_resource_config_2(self):
""" Test if we can wait for different pilot states.
"""
session = radical.pilot.Session(database_url=DBURL)
rc = radical.pilot.ResourceConfig("mylocalhost")
rc.task_launch_method = "LOCAL"
rc.mpi_launch_method = "MPIRUN"
rc.job_manager_endpoint = "fork://localhost"
rc.filesystem_endpoint = "file://localhost/"
rc.bootstrapper = "default_bootstrapper.sh"
rc.pilot_agent = "radical-pilot-agent-multicore.py"
pm = radical.pilot.PilotManager(session=session)
session.add_resource_config(rc)
pd = radical.pilot.ComputePilotDescription()
pd.resource = "mylocalhost"
pd.cores = 1
pd.runtime = 1 # minutes
pd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
pd.cleanup = True
pilot = pm.submit_pilots(pd)
pilot.wait(timeout=5*60)
pilot.cancel()
session.close()
def test__pilot_errors(self):
""" Test if pilot errors are raised properly.
"""
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 1
cpd.sandbox = "/non-/existing/directory..."
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.FAILED, "State is '%s' instead of 'Failed'." % pilot.state
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 100000000000 # This should fail - at least in 2014 ;-)
cpd.runtime = 1
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.FAILED, ("state should be %s and not %s" % (radical.pilot.FAILED, pilot.state))
session.close()
def test__pilot_cancel(self):
""" Test if we can cancel a pilot.
"""
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 1
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
assert pilot.start_time is None
assert pilot.stop_time is None
pilot.wait(state=[radical.pilot.PMGR_ACTIVE, radical.pilot.FAILED], timeout=5*60)
assert pilot.submission_time is not None
assert pilot.state == radical.pilot.PMGR_ACTIVE
assert pilot.start_time is not None
# the pilot should finish after it has reached run_time
pilot.cancel()
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.CANCELED
assert pilot.stop_time is not None
session.close()
def test__pilot_wait(self):
""" Test if we can wait for different pilot states.
"""
session = radical.pilot.Session(database_url=DBURL)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 1
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
assert pilot.start_time is None
assert pilot.stop_time is None
pilot.wait(state=[radical.pilot.PMGR_ACTIVE, radical.pilot.FAILED], timeout=5*60)
assert pilot.submission_time is not None
assert pilot.state == radical.pilot.PMGR_ACTIVE
assert pilot.start_time is not None
assert pilot.log is not None
assert pilot.sandbox == "file://localhost%s/pilot-%s/" % (cpd.sandbox, pilot.uid)
# the pilot should finish after it has reached run_time
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.DONE
assert pilot.stop_time is not None
session.close()
def test__pilot_errors(self):
""" Test if pilot errors are raised properly.
"""
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 1
cpd.sandbox = "/non-/existing/directory..."
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.FAILED, "State is '%s' instead of 'Failed'." % pilot.state
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 100000000000 # This should fail - at least in 2014 ;-)
cpd.runtime = 1
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.FAILED, ("state should be %s and not %s" % (radical.pilot.FAILED, pilot.state))
session.close()
def test__pilot_wait(self):
""" Test if we can wait for different pilot states.
"""
session = radical.pilot.Session(database_url=DBURL)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 1
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
assert pilot.start_time is None
assert pilot.stop_time is None
pilot.wait(state=[radical.pilot.ACTIVE, radical.pilot.FAILED], timeout=5*60)
assert pilot.submission_time is not None
assert pilot.state == radical.pilot.ACTIVE
assert pilot.start_time is not None
assert pilot.log is not None
assert pilot.sandbox == "file://localhost%s/pilot-%s/" % (cpd.sandbox, pilot.uid)
# the pilot should finish after it has reached run_time
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.DONE
assert pilot.stop_time is not None
session.close()
def test__pilot_cancel(self):
""" Test if we can cancel a pilot.
"""
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 1
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
assert pilot.start_time is None
assert pilot.stop_time is None
pilot.wait(state=[radical.pilot.ACTIVE, radical.pilot.FAILED], timeout=5*60)
assert pilot.submission_time is not None
assert pilot.state == radical.pilot.ACTIVE
assert pilot.start_time is not None
# the pilot should finish after it has reached run_time
pilot.cancel()
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.CANCELED
assert pilot.stop_time is not None
session.close()
def test__issue_169_part_1(self):
""" https://github.com/radical-cybertools/radical.pilot/issues/169
"""
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 1
cpd.sandbox = "/non-/existing/directory..."
cpd.cleanup = True
for i in range(0, 8):
pilot = pm.submit_pilots(pilot_descriptions=cpd)
pilot.wait(state=radical.pilot.FAILED, timeout=5*60)
assert pilot.state == radical.pilot.FAILED, "State is %s instead of 'Failed'." % pilot.state
session.close()
def test__issue_18_part_1(self):
""" https://github.com/radical-cybertools/radical.pilot/issues/18
"""
import saga
session = radical.pilot.Session(database_url=DBURL,
database_name=DBNAME)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 5
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
pilot.wait(state=radical.pilot.PMGR_ACTIVE, timeout=5 * 60)
# Now we extract the saga job id from the logs and KILL THE PROCESS
saga_id = None
for log_entry in pilot.log:
if "SAGA job submitted with job id" in log_entry:
saga_id = log_entry.replace("SAGA job submitted with job id ",
"")
if saga_id is None:
assert False, "Couldn't find SAGA Job ID in logs."
# KILL THE AGENT PROCESS.
s = saga.job.Service("fork://localhost")
j = s.get_job(saga_id)
j.cancel()
pilot.wait(timeout=60)
assert pilot.state == radical.pilot.FAILED
session.close()
def test__remote_pilot_cancel(self):
""" Test if we can cancel a pilot.
"""
session = radical.pilot.Session(database_url=DBURL,
database_name=DBNAME)
c = radical.pilot.Context('ssh')
c.user_id = self.test_ssh_uid
c.user_key = self.test_ssh_key
session.add_context(c)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = self.test_resource
cpd.cores = self.test_cores
cpd.runtime = 2
cpd.sandbox = self.test_workdir
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
#assert cu.start_time is None
#assert cu.start_time is None
pilot.wait(state=radical.pilot.ACTIVE, timeout=5 * 60)
assert pilot.state == radical.pilot.ACTIVE, "Expected state 'ACTIVE' but got %s" % pilot.state
assert pilot.submission_time is not None
assert pilot.start_time is not None
# the pilot should finish after it has reached run_time
pilot.cancel()
pilot.wait(timeout=5 * 60)
assert pilot.state == radical.pilot.CANCELED
assert pilot.stop_time is not None
session.close()
def test__issue_18_part_1(self):
""" https://github.com/radical-cybertools/radical.pilot/issues/18
"""
import saga
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = "local.localhost"
cpd.cores = 1
cpd.runtime = 5
cpd.sandbox = "/tmp/radical.pilot.sandbox.unittests"
cpd.cleanup = True
pilot = pm.submit_pilots(pilot_descriptions=cpd)
pilot.wait(state=radical.pilot.ACTIVE, timeout=5*60)
# Now we extract the saga job id from the logs and KILL THE PROCESS
saga_id = None
for log_entry in pilot.log:
if "SAGA job submitted with job id" in log_entry:
saga_id = log_entry.replace("SAGA job submitted with job id ", "")
if saga_id is None:
assert False, "Couldn't find SAGA Job ID in logs."
# KILL THE AGENT PROCESS.
s = saga.job.Service("fork://localhost")
j = s.get_job(saga_id)
j.cancel()
pilot.wait(timeout=60)
assert pilot.state == radical.pilot.FAILED
session.close()
def test__remote_pilot_cancel(self):
""" Test if we can cancel a pilot.
"""
session = radical.pilot.Session(database_url=DBURL, database_name=DBNAME)
c = radical.pilot.Context('ssh')
c.user_id = self.test_ssh_uid
c.user_key = self.test_ssh_key
session.add_context(c)
pm = radical.pilot.PilotManager(session=session)
cpd = radical.pilot.ComputePilotDescription()
cpd.resource = self.test_resource
cpd.cores = self.test_cores
cpd.runtime = 2
cpd.sandbox = self.test_workdir
pilot = pm.submit_pilots(pilot_descriptions=cpd)
assert pilot is not None
#assert cu.start_time is None
#assert cu.start_time is None
pilot.wait(state=radical.pilot.ACTIVE, timeout=5*60)
assert pilot.state == radical.pilot.ACTIVE, "Expected state 'ACTIVE' but got %s" % pilot.state
assert pilot.submission_time is not None
assert pilot.start_time is not None
# the pilot should finish after it has reached run_time
pilot.cancel()
pilot.wait(timeout=5*60)
assert pilot.state == radical.pilot.CANCELED
assert pilot.stop_time is not None
session.close()