def verify_completed_experiment_metadata(
experiment_id: int, num_expected_trials: Optional[int], has_zeroth_step: bool = False
) -> None:
# If `expected_trials` is None, the expected number of trials is
# non-deterministic.
if num_expected_trials is not None:
assert num_trials(experiment_id) == num_expected_trials
assert num_completed_trials(experiment_id) == num_expected_trials
# Check that every trial and step is COMPLETED.
trials = experiment_trials(experiment_id)
assert len(trials) > 0
for trial in trials:
if trial["state"] != "COMPLETED":
report_failed_trial(trial["id"], trial["state"])
pytest.fail(f"Trial {trial['id']} was not COMPLETED but {trial['state']}")
assert len(trial["steps"]) > 0
# Check that steps appear in increasing order of step ID.
# Step IDs should start at 0 or 1 and have no gaps.
step_ids = [s["id"] for s in trial["steps"]]
assert step_ids == sorted(step_ids)
if has_zeroth_step:
assert step_ids == list(range(0, len(step_ids)))
else:
assert step_ids == list(range(1, len(step_ids) + 1))
for step in trial["steps"]:
assert step["state"] == "COMPLETED"
if step["validation"]:
validation = step["validation"]
assert validation["state"] == "COMPLETED"
if step["checkpoint"]:
checkpoint = step["checkpoint"]
assert checkpoint["state"] in {"COMPLETED", "DELETED"}
# The last step of every trial should have a checkpoint.
for trial in trials:
last_step = trial["steps"][-1]
assert last_step["checkpoint"]
# When the experiment completes, all slots should now be free. This
# requires terminating the experiment's last container, which might
# take some time.
max_secs_to_free_slots = 30
for _ in range(max_secs_to_free_slots):
if cluster.num_free_slots() == cluster.num_slots():
break
time.sleep(1)
else:
raise AssertionError("Slots failed to free after experiment {}".format(experiment_id))
# Run a series of CLI tests on the finished experiment, to sanity check
# that basic CLI commands don't raise errors.
run_describe_cli_tests(experiment_id)
run_list_cli_tests(experiment_id)
def verify_completed_experiment_metadata(
experiment_id: int, num_expected_trials: Optional[int]) -> None:
# If `expected_trials` is None, the expected number of trials is
# non-deterministic.
if num_expected_trials is not None:
assert num_trials(experiment_id) == num_expected_trials
assert num_completed_trials(experiment_id) == num_expected_trials
# Check that every trial and step is COMPLETED.
trials = experiment_trials(experiment_id)
assert len(trials) > 0
for trial in trials:
assert trial.state == "COMPLETED"
assert len(trial.steps) > 0
# Check that steps appear in increasing order of step ID.
# Step IDs should start at 1 and have no gaps.
step_ids = [s.id for s in trial.steps]
assert step_ids == sorted(step_ids)
assert step_ids == list(range(1, len(step_ids) + 1))
for step in trial.steps:
assert step.state == "COMPLETED"
if step.validation:
assert step.validation.state == "COMPLETED"
if step.checkpoint:
assert step.checkpoint.state in {"COMPLETED", "DELETED"}
# The last step of every trial should have a checkpoint.
for trial in trials:
assert trial.steps[-1].checkpoint
# When the experiment completes, all slots should now be free. This
# requires terminating the experiment's last container, which might
# take some time.
max_secs_to_free_slots = 30
for _ in range(max_secs_to_free_slots):
if cluster.num_free_slots() == cluster.num_slots():
break
time.sleep(1)
else:
raise AssertionError(
"Slots failed to free after experiment {}".format(experiment_id))
# Run a series of CLI tests on the finished experiment, to sanity check
# that basic CLI commands don't raise errors.
run_describe_cli_tests(experiment_id)
run_list_cli_tests(experiment_id)
