def _retrieve_logs(self):
from foundations_contrib.global_state import config_manager
from foundations_core_cli.job_submission.config import load
from foundations_internal.change_directory import ChangeDirectory
import os
arguments = self._cli.arguments()
env_name = arguments.scheduler_config
job_id = arguments.job_id
current_directory = os.getcwd()
with ChangeDirectory(current_directory):
load(arguments.scheduler_config or "scheduler")
job_deployment_class = config_manager["deployment_implementation"][
"deployment_type"
]
job_deployment = job_deployment_class(job_id, None, None)
job_status = job_deployment.get_job_status()
if job_status is None:
self._cli._fail_with_message(
"Error: Job `{}` does not exist for environment `{}`".format(
job_id, env_name
)
)
elif job_status == "queued":
self._cli._fail_with_message(
"Error: Job `{}` is queued and has not produced any logs".format(job_id)
)
else:
logs = job_deployment.get_job_logs()
print(logs)
def test_loads_config_into_config_manager_when_config_present(self):
from foundations_local_docker_scheduler_plugin.config.scheduler import translate
self._set_up_config()
load(self.config_name)
self.mock_config_listing.update_config_manager_with_config.assert_called_with(
self.config_name, translate)
def _stop(self):
from foundations_contrib.global_state import config_manager
from foundations_core_cli.job_submission.config import load
from foundations_internal.change_directory import ChangeDirectory
import os
arguments = self._cli.arguments()
env_name = arguments.scheduler_config
job_id = arguments.job_id
current_directory = os.getcwd()
with ChangeDirectory(current_directory):
load(arguments.scheduler_config or "scheduler")
job_deployment_class = config_manager["deployment_implementation"][
"deployment_type"
]
job_deployment = job_deployment_class(job_id, None, None)
try:
job_status = job_deployment.get_job_status()
if job_status is None:
self._cli._fail_with_message(
"Error: Job `{}` does not exist for environment `{}`".format(
job_id, env_name
)
)
elif job_status == "queued":
self._cli._fail_with_message(
"Error: Job `{}` is queued and cannot be stopped".format(job_id)
)
elif job_status == "completed":
self._cli._fail_with_message(
"Error: Job `{}` is completed and cannot be stopped".format(job_id)
)
else:
if job_deployment.stop_running_job():
print("Stopped running job {}".format(job_id))
else:
print("Error stopping job {}".format(job_id))
except AttributeError:
print("The specified scheduler does not support this functionality")
def _delete_job(self):
from foundations_contrib.global_state import config_manager
from foundations_core_cli.job_submission.config import load
from foundations_internal.change_directory import ChangeDirectory
import os
arguments = self._cli.arguments()
env_name = arguments.scheduler_config
job_id = arguments.job_id
current_directory = os.getcwd()
with ChangeDirectory(current_directory):
load(arguments.scheduler_config or "scheduler")
job_deployment_class = config_manager["deployment_implementation"][
"deployment_type"
]
job_deployment = job_deployment_class(job_id, None, None)
job_status = job_deployment.get_job_status()
if job_status is None:
self._cli._fail_with_message(
"Error: Job `{}` does not exist for environment `{}`".format(
job_id, env_name
)
)
elif job_status in ("queued", "running", "pending"):
self._cli._fail_with_message(
"Error: Job `{}` has status `{}` and cannot be deleted".format(
job_id, job_status
)
)
else:
if job_deployment.cancel_jobs([job_id])[job_id]:
print(f"Job {job_id} successfully deleted")
else:
print(
f"Could not completely delete job {job_id}. Please make sure that the job bundle exists under ~/.foundations/job_data/"
)
def _clear_queue(self):
from foundations_contrib.global_state import config_manager
from foundations_core_cli.job_submission.config import load
from foundations_internal.change_directory import ChangeDirectory
import os
arguments = self._cli.arguments()
current_directory = os.getcwd()
with ChangeDirectory(current_directory):
load(arguments.scheduler_config or "scheduler")
job_deployment_class = config_manager["deployment_implementation"][
"deployment_type"
]
try:
num_jobs_dequeued = job_deployment_class.clear_queue()
print("Removed {} job(s) from queue".format(num_jobs_dequeued))
except AttributeError:
print("The specified scheduler does not support this functionality")
def _retrieve_artifacts(self):
from foundations_contrib.global_state import config_manager
from foundations_core_cli.job_submission.config import load
from foundations_internal.change_directory import ChangeDirectory
import os
arguments = self._cli.arguments()
env_name = arguments.scheduler_config
job_id = arguments.job_id
current_directory = os.getcwd()
if arguments.save_dir is None:
arguments.save_dir = os.path.join(current_directory, str(job_id))
with ChangeDirectory(current_directory):
load(arguments.scheduler_config or "scheduler")
job_deployment_class = config_manager["deployment_implementation"][
"deployment_type"
]
job_deployment = job_deployment_class(job_id, None, None)
job_status = job_deployment.get_job_status()
if job_status is None:
self._cli._fail_with_message(
"Error: Job `{}` does not exist for environment `{}`".format(
job_id, env_name
)
)
else:
if job_deployment.get_job_archive():
print(f"Successfully retrieved Job {job_id} from archive store")
else:
print(f"Error: Could not download Job {job_id}")
def submit(arguments):
from foundations_core_cli.job_submission.config import load
from foundations_core_cli.job_submission.deployment import deploy
from foundations_core_cli.job_submission.logs import stream_job_logs
from foundations_internal.change_directory import ChangeDirectory
from foundations_contrib.global_state import config_manager, log_manager
from foundations_contrib.set_job_resources import set_job_resources
from jsonschema import validate
import os
import os.path
current_directory = os.getcwd()
with ChangeDirectory(arguments.job_directory or current_directory):
load(arguments.scheduler_config or 'scheduler')
job_config = {}
if os.path.exists('job.config.yaml'):
with open('job.config.yaml') as file:
job_config = yaml.load(file.read(), Loader=yaml.FullLoader)
# validate(instance=job_config, schema=_job_schema)
job_resource_args = {}
if 'log_level' in job_config:
config_manager['log_level'] = job_config['log_level']
if 'worker' in job_config:
config_manager['worker_container_overrides'].update(
job_config['worker'])
if 'num_gpus' in job_config:
job_resource_args['num_gpus'] = job_config['num_gpus']
if 'ram' in job_config:
job_resource_args['ram'] = job_config['ram']
logger = log_manager.get_logger(__name__)
if arguments.command:
config_manager['worker_container_overrides'][
'args'] = arguments.command
if not os.path.exists(arguments.command[0]):
logger.warning(
f"Hey, seems like your command '{arguments.command[0]}' is not an existing file in your current directory. If you are using Atlas's advanced custom docker image functionality and know what you are doing, you can ignore this message."
)
else:
logger.warning('No command was specified.')
if arguments.num_gpus is not None:
job_resource_args['num_gpus'] = arguments.num_gpus
if arguments.ram is not None:
job_resource_args['ram'] = arguments.ram
set_job_resources(**job_resource_args)
from foundations.global_state import current_foundations_context
try:
cur_job_id = current_foundations_context().pipeline_context(
).file_name
except ValueError:
cur_job_id = None
deployment = deploy(
arguments.project_name or job_config.get('project_name'),
arguments.entrypoint or job_config.get('entrypoint'),
arguments.params or job_config.get('params'))
if arguments.stream_job_logs:
try:
stream_job_logs(deployment)
except KeyboardInterrupt:
pass
if cur_job_id is not None:
current_foundations_context().pipeline_context(
).file_name = cur_job_id
return deployment
def test_does_not_print_error_when_config_present(self):
self._set_up_config()
load(self.config_name)
self.print_mock.assert_not_called()
def test_prints_warning_message_when_config_missing(self):
load(self.config_name)
self.print_mock.assert_called_with(
f"Could not find submission configuration with name: `{self.config_name}`"
)
def test_exits_when_config_missing(self):
load(self.config_name)
self.exit_mock.assert_called_with(1)