def _add_snapshot(self, snapshot: Snapshot, iter_: int):
# Parts of the metadata will be used in the underlying data model,
# which is be mutable, hence we thaw it here—once.
metadata = pyrsistent.thaw(snapshot.data()[ids.METADATA])
snapshot_tree = Node(
iter_,
{
ids.STATUS: snapshot.data()[ids.STATUS],
SORTED_REALIZATION_IDS: metadata[SORTED_REALIZATION_IDS],
SORTED_JOB_IDS: metadata[SORTED_JOB_IDS],
},
NodeType.ITER,
)
for real_id in snapshot_tree.data[SORTED_REALIZATION_IDS]:
real = snapshot.data()[ids.REALS][real_id]
real_node = Node(
real_id,
{
ids.STATUS:
real[ids.STATUS],
ids.ACTIVE:
real[ids.ACTIVE],
REAL_JOB_STATUS_AGGREGATED:
metadata[REAL_JOB_STATUS_AGGREGATED][real_id],
REAL_STATUS_COLOR:
metadata[REAL_STATUS_COLOR][real_id],
},
NodeType.REAL,
)
snapshot_tree.add_child(real_node)
for step_id, step in real[ids.STEPS].items():
step_node = Node(step_id, {ids.STATUS: step[ids.STATUS]},
NodeType.STEP)
real_node.add_child(step_node)
for job_id in metadata[SORTED_JOB_IDS]:
job = step[ids.JOBS][job_id]
job_dict = dict(job)
job_dict[ids.DATA] = job.data
job_node = Node(job_id, job_dict, NodeType.JOB)
step_node.add_child(job_node)
if iter_ in self.root.children:
self.modelAboutToBeReset.emit()
self.root.children[iter_] = snapshot_tree
snapshot_tree.parent = self.root
self.modelReset.emit()
return
parent = QModelIndex()
next_iter = len(self.root.children)
self.beginInsertRows(parent, next_iter, next_iter)
self.root.add_child(snapshot_tree)
self.root.children[iter_] = snapshot_tree
self.rowsInserted.emit(parent, snapshot_tree.row(),
snapshot_tree.row())
def test_print_progress(self):
out = StringIO()
monitor = Monitor(out=out)
sd = SnapshotDict(status="")
for i in range(0, 100):
status = REALIZATION_STATE_FINISHED if i < 50 else REALIZATION_STATE_WAITING
sd.reals[i] = Realization(status=status, active=True)
monitor._snapshots[0] = Snapshot(sd.dict())
monitor._start_time = datetime.now()
general_event = _UpdateEvent(
phase_name="Test Phase",
current_phase=0,
total_phases=2,
progress=0.5,
indeterminate=False,
iteration=0,
)
monitor._print_progress(general_event)
self.assertEqual(
"""\r
--> Test Phase
1/2 |███████████████ | 50% Running time: 0 seconds
Waiting 50/100
Pending 0/100
Running 0/100
Failed 0/100
Finished 50/100
Unknown 0/100
""",
out.getvalue(),
)
def _create_snapshot(self):
reals = {}
for real in self.get_active_reals():
reals[str(real.get_iens())] = Realization(
active=True,
status=state.REALIZATION_STATE_WAITING,
)
for step in real.get_steps():
reals[str(real.get_iens())].steps[str(step.get_id())] = Step(
status=state.STEP_STATE_UNKNOWN
)
for job in step.get_jobs():
reals[str(real.get_iens())].steps[str(step.get_id())].jobs[
str(job.get_id())
] = Job(
status=state.JOB_STATE_START,
data={},
name=job.get_name(),
)
top = SnapshotDict(
reals=reals,
status=state.ENSEMBLE_STATE_UNKNOWN,
metadata=self.get_metadata(),
)
return Snapshot(top.dict())
def _full_snapshot_event(self,
iter_) -> typing.Optional[FullSnapshotEvent]:
"""Return a FullSnapshotEvent if it was possible to create a snapshot.
Return None if not, indicating that there should be no event."""
run_context = self._model.get_run_context()
detailed_progress = self._model.getDetailedProgress()
if detailed_progress == _THE_EMPTY_DETAILED_PROGRESS:
return None
snapshot_dict = self._create_snapshot_dict(run_context,
detailed_progress, iter_)
if not snapshot_dict:
return None
snapshot = Snapshot(snapshot_dict.dict())
self._set_iter_snapshot(iter_, snapshot)
return FullSnapshotEvent(
phase_name=self._model.getPhaseName(),
current_phase=self._model.currentPhase(),
total_phases=self._model.phaseCount(),
indeterminate=self._model.isIndeterminate(),
progress=self._progress(),
iteration=iter_,
snapshot=snapshot,
)
def track(self):
while True:
event = self._work_queue.get()
if isinstance(event, str):
try:
if event == EvaluatorTracker.DONE:
yield EndEvent(
failed=self._model.hasRunFailed(),
failed_msg=self._model.getFailMessage(),
)
elif event == EvaluatorTracker.CONNECTION_ERROR:
yield EndEvent(
failed=True,
failed_msg="Connection error",
)
except GeneratorExit:
# consumers may exit at this point, make sure the last
# task is marked as done
pass
self._work_queue.task_done()
break
elif event["type"] == ids.EVTYPE_EE_SNAPSHOT:
iter_ = event.data["iter"]
snapshot = Snapshot(event.data)
self._iter_snapshot[iter_] = snapshot
yield FullSnapshotEvent(
phase_name=self._model.getPhaseName(),
current_phase=self._model.currentPhase(),
total_phases=self._model.phaseCount(),
indeterminate=self._model.isIndeterminate(),
progress=self._progress(),
iteration=iter_,
snapshot=snapshot,
)
elif event["type"] == ids.EVTYPE_EE_SNAPSHOT_UPDATE:
iter_ = event.data["iter"]
if iter_ not in self._iter_snapshot:
raise OutOfOrderSnapshotUpdateException(
f"got {ids.EVTYPE_EE_SNAPSHOT_UPDATE} without having stored snapshot for iter {iter_}"
)
partial = PartialSnapshot(
self._iter_snapshot[iter_]).from_cloudevent(event)
self._iter_snapshot[iter_].merge_event(partial)
yield SnapshotUpdateEvent(
phase_name=self._model.getPhaseName(),
current_phase=self._model.currentPhase(),
total_phases=self._model.phaseCount(),
indeterminate=self._model.isIndeterminate(),
progress=self._progress(),
iteration=iter_,
partial_snapshot=partial,
)
self._work_queue.task_done()
def full_snapshot() -> Snapshot:
real = Realization(
status=REALIZATION_STATE_UNKNOWN,
active=True,
steps={
"0":
Step(
status="",
jobs={
"0":
Job(
start_time=dt.now(),
end_time=dt.now(),
name="poly_eval",
status=JOB_STATE_START,
error="error",
stdout="std_out_file",
stderr="std_err_file",
data={
CURRENT_MEMORY_USAGE: "123",
MAX_MEMORY_USAGE: "312",
},
),
"1":
Job(
start_time=dt.now(),
end_time=dt.now(),
name="poly_postval",
status=JOB_STATE_START,
error="error",
stdout="std_out_file",
stderr="std_err_file",
data={
CURRENT_MEMORY_USAGE: "123",
MAX_MEMORY_USAGE: "312",
},
),
},
)
},
)
snapshot = SnapshotDict(
status=ENSEMBLE_STATE_STARTED,
reals={},
)
for i in range(0, 100):
snapshot.reals[str(i)] = copy.deepcopy(real)
return Snapshot(snapshot.dict())
def snapshot_to_tree(snapshot: Snapshot, iter_: int) -> Node:
iter_node = Node(iter_, {ids.STATUS: snapshot.get_status()}, NodeType.ITER)
snapshot_d = SnapshotDict(**snapshot.to_dict())
for real_id in sorted(snapshot_d.reals, key=int):
real = snapshot_d.reals[real_id]
real_node = Node(
real_id,
{
ids.STATUS: real.status,
ids.ACTIVE: real.active
},
NodeType.REAL,
)
iter_node.add_child(real_node)
for step_id, step in real.steps.items():
step_node = Node(step_id, {ids.STATUS: step.status}, NodeType.STEP)
real_node.add_child(step_node)
for job_id in sorted(step.jobs, key=int):
job = step.jobs[job_id]
job_dict = dict(job)
job_dict[ids.DATA] = job.data
job_node = Node(job_id, job_dict, NodeType.JOB)
step_node.add_child(job_node)
return iter_node
def test_legends(self):
monitor = Monitor(out=StringIO())
sd = SnapshotDict(status="")
for i in range(0, 100):
status = REALIZATION_STATE_FINISHED if i < 10 else REALIZATION_STATE_RUNNING
sd.reals[i] = Realization(status=status, active=True)
monitor._snapshots[0] = Snapshot(sd.dict())
legends = monitor._get_legends()
self.assertEqual(
""" Waiting 0/100
Pending 0/100
Running 90/100
Failed 0/100
Finished 10/100
Unknown 0/100
""",
legends,
)
def test_print_progress(self):
out = StringIO()
monitor = Monitor(out=out)
sd = SnapshotDict(status="")
for i in range(0, 100):
status = REALIZATION_STATE_FINISHED if i < 50 else REALIZATION_STATE_WAITING
sd.reals[i] = Realization(status=status, active=True)
monitor._snapshots[0] = Snapshot(sd.dict())
monitor._start_time = datetime.now()
general_event = _UpdateEvent(
phase_name="Test Phase",
current_phase=0,
total_phases=2,
progress=0.5,
indeterminate=False,
iteration=0,
)
monitor._print_progress(general_event)
# For some reason, `tqdm` adds an extra line containing a progress-bar,
# even though this test only calls it once.
# I suspect this has something to do with the way `tqdm` does refresh,
# but do not know how to fix it.
# Seems not be a an issue when used normally.
expected = """ --> Test Phase
| | 0% it
1/2 |##############################5 | 50% Running time: 0 seconds
Waiting 50/100
Pending 0/100
Running 0/100
Failed 0/100
Finished 50/100
Unknown 0/100
"""
assert out.getvalue().replace("\r", "\n") == expected
def create_snapshot(ensemble):
reals = {}
for real in ensemble.get_active_reals():
reals[str(real.get_iens())] = _Realization(
active=True,
start_time=None,
end_time=None,
status="Waiting",
)
for stage in real.get_stages():
reals[str(real.get_iens())].stages[str(
stage.get_id())] = _Stage(
status="Unknown",
start_time=None,
end_time=None,
)
for step in stage.get_steps():
reals[str(real.get_iens())].stages[str(
stage.get_id())].steps[str(step.get_id())] = _Step(
status="Unknown", start_time=None, end_time=None)
for job in step.get_jobs():
reals[str(real.get_iens())].stages[str(
stage.get_id())].steps[str(
step.get_id())].jobs[str(job.get_id())] = _Job(
status="Pending",
data={},
start_time=None,
end_time=None,
name=job.get_name(),
)
top = _SnapshotDict(
reals=reals,
status="Unknown",
forward_model=_ForwardModel(step_definitions={}),
metadata=ensemble.get_metadata(),
)
return Snapshot(top.dict())
def _batch(self, events):
batch: List[CloudEvent] = []
for event in events:
if event["type"] == ids.EVTYPE_EE_SNAPSHOT:
# A new iteration, so ensure any updates for the previous one,
# is emitted.
if batch:
yield self._flush(batch)
batch = []
iter_ = event.data["iter"]
snapshot = Snapshot(event.data)
self._iter_snapshot[iter_] = snapshot
yield FullSnapshotEvent(
phase_name=self._model.getPhaseName(),
current_phase=self._model.currentPhase(),
total_phases=self._model.phaseCount(),
indeterminate=self._model.isIndeterminate(),
progress=self._progress(),
iteration=iter_,
snapshot=snapshot,
)
self._work_queue.task_done()
elif event["type"] == ids.EVTYPE_EE_SNAPSHOT_UPDATE:
iter_ = event.data["iter"]
if iter_ not in self._iter_snapshot:
raise OutOfOrderSnapshotUpdateException(
f"got {ids.EVTYPE_EE_SNAPSHOT_UPDATE} without having stored snapshot for iter {iter_}"
)
batch.append(event)
else:
raise ValueError("got unexpected event type", event["type"])
if batch:
yield self._flush(batch)
def test_monitor_stop(evaluator):
with evaluator.run() as monitor:
for event in monitor.track():
snapshot = Snapshot(event.data)
break
assert snapshot.get_status() == ENSEMBLE_STATE_STARTED
def test_dispatchers_can_connect_and_monitor_can_shut_down_evaluator(
evaluator):
with evaluator.run() as monitor:
events = monitor.track()
host = evaluator._config.host
port = evaluator._config.port
# first snapshot before any event occurs
snapshot_event = next(events)
snapshot = Snapshot(snapshot_event.data)
assert snapshot.get_status() == ENSEMBLE_STATE_STARTED
# two dispatchers connect
with Client(host, port, "/dispatch") as dispatch1, Client(
host, port, "/dispatch") as dispatch2:
# first dispatcher informs that job 0 is running
send_dispatch_event(
dispatch1,
identifiers.EVTYPE_FM_JOB_RUNNING,
"/ert/ee/0/real/0/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("0", "0", "0").status == JOB_STATE_RUNNING
# second dispatcher informs that job 0 is running
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_RUNNING,
"/ert/ee/0/real/1/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("1", "0", "0").status == JOB_STATE_RUNNING
# second dispatcher informs that job 0 is done
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_SUCCESS,
"/ert/ee/0/real/1/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("1", "0", "0").status == JOB_STATE_FINISHED
# second dispatcher informs that job 1 is failed
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_FAILURE,
"/ert/ee/0/real/1/step/0/job/1",
"event_job_1_fail",
{identifiers.ERROR_MSG: "error"},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("1", "0", "1").status == JOB_STATE_FAILURE
# a second monitor connects
with ee_monitor.create(host, port) as monitor2:
events2 = monitor2.track()
snapshot = Snapshot(next(events2).data)
assert snapshot.get_status() == ENSEMBLE_STATE_STARTED
assert snapshot.get_job("0", "0",
"0").status == JOB_STATE_RUNNING
assert snapshot.get_job("1", "0",
"0").status == JOB_STATE_FINISHED
# one monitor requests that server exit
monitor.signal_cancel()
# both monitors should get a terminated event
terminated = next(events)
terminated2 = next(events2)
assert terminated["type"] == identifiers.EVTYPE_EE_TERMINATED
assert terminated2["type"] == identifiers.EVTYPE_EE_TERMINATED
for e in [events, events2]:
for _ in e:
assert False, "got unexpected event from monitor"
def test_dispatchers_can_connect_and_monitor_can_shut_down_evaluator(
evaluator):
monitor = evaluator.run()
events = monitor.track()
host = evaluator._config.host
port = evaluator._config.port
# first snapshot before any event occurs
snapshot_event = next(events)
snapshot = Snapshot(snapshot_event.data)
assert snapshot.get_status() == "Unknown"
# two dispatchers connect
with Client(host, port,
"/dispatch") as dispatch1, Client(host, port,
"/dispatch") as dispatch2:
# first dispatcher informs that job 0 is running
send_dispatch_event(
dispatch1,
identifiers.EVTYPE_FM_JOB_RUNNING,
"/ert/ee/0/real/0/stage/0/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("0", "0", "0", "0")["status"] == "Running"
# second dispatcher informs that job 0 is running
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_RUNNING,
"/ert/ee/0/real/1/stage/0/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("1", "0", "0", "0")["status"] == "Running"
# second dispatcher informs that job 0 is done
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_SUCCESS,
"/ert/ee/0/real/1/stage/0/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("1", "0", "0", "0")["status"] == "Finished"
# a second monitor connects
monitor2 = ee_monitor.create(host, port)
events2 = monitor2.track()
snapshot = Snapshot(next(events2).data)
assert snapshot.get_status() == "Unknown"
assert snapshot.get_job("0", "0", "0", "0")["status"] == "Running"
assert snapshot.get_job("1", "0", "0", "0")["status"] == "Finished"
# one monitor requests that server exit
monitor.signal_cancel()
# both monitors should get a terminated event
terminated = next(events)
terminated2 = next(events2)
assert terminated["type"] == identifiers.EVTYPE_EE_TERMINATED
assert terminated2["type"] == identifiers.EVTYPE_EE_TERMINATED
for e in [events, events2]:
for _ in e:
assert False, "got unexpected event from monitor"
def test_monitor_stop(evaluator):
monitor = evaluator.run()
events = monitor.track()
snapshot = Snapshot(next(events).data)
assert snapshot.get_status() == "Unknown"
def test_dispatchers_can_connect_and_monitor_can_shut_down_evaluator(
evaluator):
with evaluator.run() as monitor:
events = monitor.track()
host = evaluator._config.host
port = evaluator._config.port
token = evaluator._config.token
cert = evaluator._config.cert
url = evaluator._config.url
# first snapshot before any event occurs
snapshot_event = next(events)
snapshot = Snapshot(snapshot_event.data)
assert snapshot.get_status() == ENSEMBLE_STATE_UNKNOWN
# two dispatchers connect
with Client(
url + "/dispatch",
cert=cert,
token=token,
max_retries=1,
timeout_multiplier=1,
) as dispatch1, Client(
url + "/dispatch",
cert=cert,
token=token,
max_retries=1,
timeout_multiplier=1,
) as dispatch2:
# first dispatcher informs that job 0 is running
send_dispatch_event(
dispatch1,
identifiers.EVTYPE_FM_JOB_RUNNING,
f"/ert/ee/{evaluator._ee_id}/real/0/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
# second dispatcher informs that job 0 is running
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_RUNNING,
f"/ert/ee/{evaluator._ee_id}/real/1/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
# second dispatcher informs that job 0 is done
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_SUCCESS,
f"/ert/ee/{evaluator._ee_id}/real/1/step/0/job/0",
"event1",
{"current_memory_usage": 1000},
)
# second dispatcher informs that job 1 is failed
send_dispatch_event(
dispatch2,
identifiers.EVTYPE_FM_JOB_FAILURE,
f"/ert/ee/{evaluator._ee_id}/real/1/step/0/job/1",
"event_job_1_fail",
{identifiers.ERROR_MSG: "error"},
)
snapshot = Snapshot(next(events).data)
assert snapshot.get_job("1", "0", "0").status == JOB_STATE_FINISHED
assert snapshot.get_job("0", "0", "0").status == JOB_STATE_RUNNING
assert snapshot.get_job("1", "0", "1").status == JOB_STATE_FAILURE
# a second monitor connects
with ee_monitor.create(host, port, "wss", cert, token) as monitor2:
events2 = monitor2.track()
full_snapshot_event = next(events2)
assert full_snapshot_event[
"type"] == identifiers.EVTYPE_EE_SNAPSHOT
snapshot = Snapshot(full_snapshot_event.data)
assert snapshot.get_status() == ENSEMBLE_STATE_UNKNOWN
assert snapshot.get_job("0", "0", "0").status == JOB_STATE_RUNNING
assert snapshot.get_job("1", "0", "0").status == JOB_STATE_FINISHED
# one monitor requests that server exit
monitor.signal_cancel()
# both monitors should get a terminated event
terminated = next(events)
terminated2 = next(events2)
assert terminated["type"] == identifiers.EVTYPE_EE_TERMINATED
assert terminated2["type"] == identifiers.EVTYPE_EE_TERMINATED
for e in [events, events2]:
for undexpected_event in e:
assert (
False
), f"got unexpected event {undexpected_event} from monitor"
