def create_task(**kwargs):
"""Creates a new task in our database/object storage.
Args:
**kwargs: arbitrary keyword arguments.
Returns:
The task info.
"""
name = kwargs.get('name', None)
description = kwargs.get('description', None)
tags = kwargs.get('tags', None)
image = kwargs.get('image', None)
commands = kwargs.get('commands', None)
arguments = kwargs.get('arguments', None)
experiment_notebook = kwargs.get('experiment_notebook', None)
deployment_notebook = kwargs.get('deployment_notebook', None)
is_default = kwargs.get('is_default', None)
copy_from = kwargs.get('copy_from', None)
if not isinstance(name, str):
raise BadRequest("name is required")
if copy_from and (experiment_notebook or deployment_notebook):
raise BadRequest("Either provide notebooks or a task to copy from")
if tags is None or len(tags) == 0:
tags = ["DEFAULT"]
if any(tag not in VALID_TAGS for tag in tags):
valid_str = ",".join(VALID_TAGS)
raise BadRequest(f"Invalid tag. Choose any of {valid_str}")
# check if image is a valid docker image
if image:
pattern = re.compile('[a-z0-9.-]+([/]{1}[a-z0-9.-]+)+([:]{1}[a-z0-9.-]+){0,1}$')
if pattern.match(image) is None:
raise BadRequest("invalid docker image name")
check_comp_name = db_session.query(Task).filter_by(name=name).first()
if check_comp_name:
raise BadRequest("a task with that name already exists")
# creates a task with specified name,
# but copies notebooks from a source task
if copy_from:
return copy_task(name, description, tags, copy_from)
task_id = str(uuid_alpha())
# loads a sample notebook if none was sent
if experiment_notebook is None and "DATASETS" not in tags:
experiment_notebook = EXPERIMENT_NOTEBOOK
if deployment_notebook is None and "DATASETS" not in tags:
deployment_notebook = DEPLOYMENT_NOTEBOOK
# The new task must have its own task_id, experiment_id and operator_id.
# Notice these values are ignored when a notebook is run in a pipeline.
# They are only used by JupyterLab interface.
init_notebook_metadata(task_id, deployment_notebook, experiment_notebook)
# saves new notebooks to object storage
if "DATASETS" not in tags:
obj_name = f"{PREFIX}/{task_id}/Experiment.ipynb"
experiment_notebook_path = f"minio://{BUCKET_NAME}/{obj_name}"
put_object(obj_name, dumps(experiment_notebook).encode())
obj_name = f"{PREFIX}/{task_id}/Deployment.ipynb"
deployment_notebook_path = f"minio://{BUCKET_NAME}/{obj_name}"
put_object(obj_name, dumps(deployment_notebook).encode())
# create deployment notebook and experiment_notebook on jupyter
create_jupyter_files(task_name=name,
deployment_notebook=dumps(deployment_notebook).encode(),
experiment_notebook=dumps(experiment_notebook).encode())
else:
experiment_notebook_path = None
deployment_notebook_path = None
if commands is None or len(commands) == 0:
commands = DEFAULT_COMMANDS
if arguments is None or len(arguments) == 0:
arguments = DEFAULT_ARGUMENTS
# saves task info to the database
task = Task(uuid=task_id,
name=name,
description=description,
tags=tags,
image=image,
commands=commands,
arguments=arguments,
experiment_notebook_path=experiment_notebook_path,
deployment_notebook_path=deployment_notebook_path,
is_default=is_default)
db_session.add(task)
db_session.commit()
return task.as_dict()
def copy_task(name, description, tags, copy_from):
"""Makes a copy of a task in our database/object storage.
Args:
name (str): the task name.
description (str): the task description.
tags (list): the task tags list.
copy_from (str): the task_id from which the notebooks are copied.
Returns:
The task info.
"""
task = Task.query.get(copy_from)
if task is None:
raise BadRequest("Source task does not exist")
task_id = uuid_alpha()
image = task.image
commands = task.commands
arguments = task.arguments
# reads source notebooks from object storage
source_name = f"{PREFIX}/{copy_from}/Deployment.ipynb"
deployment_notebook = loads(get_object(source_name))
source_name = f"{PREFIX}/{copy_from}/Experiment.ipynb"
experiment_notebook = loads(get_object(source_name))
# Even though we are creating 'copies', the new task must have
# its own task_id, experiment_id and operator_id.
# We don't want to mix models and metrics of different tasks.
# Notice these values are ignored when a notebook is run in a pipeline.
# They are only used by JupyterLab interface.
init_notebook_metadata(task_id, deployment_notebook, experiment_notebook)
# saves new notebooks to object storage
destination_name = f"{PREFIX}/{task_id}/Deployment.ipynb"
deployment_notebook_path = f"minio://{BUCKET_NAME}/{destination_name}"
put_object(destination_name, dumps(deployment_notebook).encode())
destination_name = f"{PREFIX}/{task_id}/Experiment.ipynb"
experiment_notebook_path = f"minio://{BUCKET_NAME}/{destination_name}"
put_object(destination_name, dumps(experiment_notebook).encode())
# create deployment notebook and eperiment notebook on jupyter
create_jupyter_files(task_name=name,
deployment_notebook=dumps(deployment_notebook).encode(),
experiment_notebook=dumps(experiment_notebook).encode())
# saves task info to the database
task = Task(uuid=task_id,
name=name,
description=description,
tags=tags,
image=image,
commands=commands,
arguments=arguments,
deployment_notebook_path=deployment_notebook_path,
experiment_notebook_path=experiment_notebook_path,
is_default=False)
db_session.add(task)
db_session.commit()
return task.as_dict()
