def stream(state: State, name: str, data: str, method_verb: InferenceVerb):
"""
Perform stream inference task on launched prediction instance.
"""
method_verb = InferenceVerb(method_verb)
try:
namespace = get_kubectl_current_context_namespace()
# TODO: check if kind field of inference instance Run is correct
inference_instance = Run.get(name=name, namespace=namespace)
if not inference_instance:
handle_error(user_msg=Texts.INSTANCE_NOT_EXISTS_ERROR_MSG.format(
name=name))
exit(1)
if not inference_instance.state == RunStatus.RUNNING:
handle_error(user_msg=Texts.INSTANCE_NOT_RUNNING_ERROR_MSG.format(
name=name, running_code=RunStatus.RUNNING.value))
exit(1)
inference_instance_url = get_inference_instance_url(
inference_instance=inference_instance)
stream_url = f'{inference_instance_url}:{method_verb.value}'
except Exception:
handle_error(logger,
Texts.INSTANCE_GET_FAIL_ERROR_MSG.format(name=name),
Texts.INSTANCE_GET_FAIL_ERROR_MSG.format(name=name),
add_verbosity_msg=state.verbosity == 0)
exit(1)
try:
with open(data, 'r', encoding='utf-8') as data_file:
stream_data = json.load(data_file)
except (json.JSONDecodeError, IOError):
handle_error(logger, Texts.JSON_LOAD_ERROR_MSG.format(data=data),
Texts.JSON_LOAD_ERROR_MSG.format(data=data))
exit(1)
try:
api_key = get_api_key()
headers = {
'Authorization': api_key,
'Accept': 'application/json',
'Content-Type': 'application/json'
}
with spinner(text=Texts.WAITING_FOR_RESPONSE_MSG):
stream_response = requests.post(
stream_url,
data=json.dumps(stream_data), # nosec - request to k8s cluster
verify=False,
headers=headers)
stream_response.raise_for_status()
click.echo(stream_response.text)
except Exception as e:
error_msg = Texts.INFERENCE_OTHER_ERROR_MSG.format(exception=e)
if hasattr(e, 'response'):
error_msg += Texts.INFERENCE_ERROR_RESPONSE_MSG.format(
response_text=e.response.text) # type: ignore
handle_error(logger, error_msg, error_msg)
exit(1)
def view(context, state: State, experiment_name: str, tensorboard: bool,
username: str):
"""
Displays details of an experiment.
"""
try:
if username:
namespace = username
else:
namespace = get_kubectl_current_context_namespace()
run = Run.get(name=experiment_name, namespace=namespace)
if not run:
handle_error(user_msg=Texts.EXPERIMENT_NOT_FOUND_ERROR_MSG.format(
experiment_name=experiment_name))
exit(2)
click.echo(
tabulate([run.cli_representation],
headers=EXPERIMENTS_LIST_HEADERS,
tablefmt="orgtbl"))
click.echo(Texts.PODS_PARTICIPATING_LIST_HEADER)
pods = get_namespaced_pods(label_selector="runName=" + experiment_name,
namespace=namespace)
tabular_output = []
containers_resources = []
pending_pods = []
for pod in pods:
status_string = ""
if pod.status.conditions:
for cond in pod.status.conditions:
msg = "\n" if not cond.reason else "\n reason: " + \
wrap_text(cond.reason, width=POD_CONDITIONS_MAX_WIDTH)
msg = msg + ", \n message: " + wrap_text(cond.message, width=POD_CONDITIONS_MAX_WIDTH) \
if cond.message else msg
status_string += wrap_text(
cond.type + ": " + cond.status,
width=POD_CONDITIONS_MAX_WIDTH) + msg + "\n"
else:
pod_events = get_pod_events(namespace=namespace,
name=pod.metadata.name)
for event in pod_events:
msg = "\n" if not event.reason else "\n reason: " + \
wrap_text(event.reason, width=POD_CONDITIONS_MAX_WIDTH)
msg = msg + ", \n message: " + wrap_text(event.message, width=POD_CONDITIONS_MAX_WIDTH) \
if event.message else msg
status_string += msg + "\n"
if pod.status.phase.upper() == PodStatus.PENDING.value:
pending_pods.append(pod.metadata.name)
container_statuses = defaultdict(lambda: None)
if pod.status.container_statuses:
for container_status in pod.status.container_statuses:
container_statuses[
container_status.name] = container_status.state
container_details = []
for container in pod.spec.containers:
container_description = Texts.CONTAINER_DETAILS_MSG.format(
name=container.name,
status=container_status_to_msg(
container_statuses[container.name]),
volumes=container_volume_mounts_to_msg(
container.volume_mounts, spaces=2),
resources=container_resources_to_msg(container.resources,
spaces=4))
container_details.append(container_description)
containers_resources.append(container.resources)
container_details = ''.join(container_details)
tabular_output.append([
pod.metadata.name,
wrap_text(pod.metadata.uid, width=UID_MAX_WIDTH, spaces=0),
status_string, container_details
])
click.echo(
tabulate(tabular_output,
Texts.PODS_TABLE_HEADERS,
tablefmt="orgtbl"))
try:
cpu_requests_sum = sum_cpu_resources([
container_resource.requests["cpu"]
for container_resource in containers_resources
if container_resource.requests
and container_resource.requests.get("cpu")
])
mem_requests_sum = sum_mem_resources([
container_resource.requests["memory"]
for container_resource in containers_resources
if container_resource.requests
and container_resource.requests.get("memory")
])
cpu_limits_sum = sum_cpu_resources([
container_resource.limits["cpu"]
for container_resource in containers_resources
if container_resource.limits
and container_resource.limits.get("cpu")
])
mem_limits_sum = sum_mem_resources([
container_resource.limits["memory"]
for container_resource in containers_resources
if container_resource.limits
and container_resource.limits.get("memory")
])
except ValueError as exception:
handle_error(
logger,
Texts.RESOURCES_SUM_PARSING_ERROR_MSG.format(
error_msg=str(exception)),
Texts.RESOURCES_SUM_PARSING_ERROR_MSG.format(
error_msg=str(exception)))
click.echo(Texts.RESOURCES_SUM_LIST_HEADER)
click.echo(
tabulate(list(
zip(Texts.RESOURCES_SUM_TABLE_ROWS_HEADERS, [
cpu_requests_sum, mem_requests_sum, cpu_limits_sum,
mem_limits_sum
])),
Texts.RESOURCES_SUM_TABLE_HEADERS,
tablefmt="orgtbl"))
if tensorboard:
click.echo()
context.invoke(tensorboard_command,
experiment_name=[experiment_name])
if pending_pods:
click.echo()
try:
cpu = False
memory = False
for pod in pending_pods:
events_list = get_pod_events(namespace=namespace, name=pod)
for event in events_list:
if "insufficient cpu" in event.message.lower():
cpu = True
elif "insufficient memory" in event.message.lower():
memory = True
if cpu and memory:
break
if cpu and memory:
break
if not cpu and not memory:
exit(0)
if cpu and memory:
resources = "number of cpus and amount of memory"
elif cpu:
resources = "number of cpus"
else:
resources = "amount of memory"
click.echo(
Texts.INSUFFICIENT_RESOURCES_MESSAGE.format(
resources=resources))
click.echo()
top_cpu_users, top_mem_users = get_highest_usage()
click.echo(
Texts.TOP_CPU_CONSUMERS.format(consumers=", ".join([
res.user_name for res in
top_cpu_users[0:3 if len(top_cpu_users
) > 2 else len(top_cpu_users)]
])))
click.echo(
Texts.TOP_MEMORY_CONSUMERS.format(consumers=", ".join([
res.user_name for res in
top_mem_users[0:3 if len(top_mem_users
) > 2 else len(top_mem_users)]
])))
except Exception:
click.echo(Texts.PROBLEMS_WHILE_GATHERING_USAGE_DATA)
logger.exception(
Texts.PROBLEMS_WHILE_GATHERING_USAGE_DATA_LOGS)
except Exception:
handle_error(logger, Texts.VIEW_OTHER_ERROR_MSG,
Texts.VIEW_OTHER_ERROR_MSG)
exit(1)
def test_get_run_not_found(mock_k8s_api_client: CustomObjectsApi):
mock_k8s_api_client.get_namespaced_custom_object.side_effect = ApiException(status=404)
run = Run.get(name=RUN_NAME, namespace=NAMESPACE)
assert run is None
def test_get_run_failure(mock_k8s_api_client: CustomObjectsApi):
mock_k8s_api_client.get_namespaced_custom_object.side_effect = ApiException(status=500)
with pytest.raises(ApiException):
Run.get(name=RUN_NAME, namespace=NAMESPACE)
def test_get_run_from_namespace(mock_k8s_api_client: CustomObjectsApi):
mock_k8s_api_client.get_namespaced_custom_object.return_value = GET_RUN_RESPONSE_RAW
run = Run.get(name=RUN_NAME, namespace=NAMESPACE)
assert run is not None and type(run) is Run
