def start_experiment(experiment):
# Update experiment status to show that its started
experiment.set_status(ExperimentLifeCycle.SCHEDULED)
project = experiment.project
group = experiment.experiment_group
job_docker_image = None # This will force the spawners to use the default docker image
if experiment.specification.build:
try:
image_name, image_tag = get_image_info(build_job=experiment.build_job)
except ValueError as e:
_logger.warning('Could not start the experiment, %s', e)
experiment.set_status(ExperimentLifeCycle.FAILED,
message='External git repo was note found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start experiment with built image `%s`', job_docker_image)
else:
_logger.info('Start experiment with default image.')
spawner_class = get_spawner_class(experiment.specification.framework)
# Use spawners to start the experiment
spawner = spawner_class(project_name=project.unique_name,
experiment_name=experiment.unique_name,
experiment_group_name=group.unique_name if group else None,
project_uuid=project.uuid.hex,
experiment_group_uuid=group.uuid.hex if group else None,
experiment_uuid=experiment.uuid.hex,
original_name=experiment.original_unique_name,
cloning_strategy=experiment.cloning_strategy,
spec=experiment.specification,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
in_cluster=True,
job_docker_image=job_docker_image,
use_sidecar=True,
sidecar_config=config.get_requested_params(to_str=True))
try:
response = spawner.start_experiment()
except ApiException as e:
_logger.warning('Could not start the experiment, please check your polyaxon spec %s', e)
experiment.set_status(
ExperimentLifeCycle.FAILED,
message='Could not start the experiment, encountered a Kubernetes ApiException.')
return
except Exception as e:
_logger.warning('Could not start the experiment, please check your polyaxon spec %s', e)
experiment.set_status(
ExperimentLifeCycle.FAILED,
message='Could not start the experiment encountered an {} exception.'.format(
e.__class__.__name__
))
return
handle_experiment(experiment=experiment, spawner=spawner, response=response)
def start_job(job):
# Update job status to show that its started
job.set_status(JobLifeCycle.SCHEDULED)
try:
image_name, image_tag = get_image_info(build_job=job.build_job)
except ValueError as e:
_logger.warning('Could not start the notebook, %s', e)
job.set_status(JobLifeCycle.FAILED,
message='External git repo was note found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start notebook with built image `%s`', job_docker_image)
spawner = JobSpawner(
project_name=job.project.unique_name,
project_uuid=job.project.uuid.hex,
job_name=job.unique_name,
job_uuid=job.uuid.hex,
spec=job.specification,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
job_docker_image=job_docker_image,
in_cluster=True,
use_sidecar=True,
sidecar_config=config.get_requested_params(to_str=True))
try:
results = spawner.start_job(resources=job.resources,
node_selectors=job.node_selectors)
except ApiException as e:
_logger.warning(
'Could not start job, please check your polyaxon spec %s', e)
job.set_status(
JobLifeCycle.FAILED,
message=
'Could not start job, encountered a Kubernetes ApiException.')
return
except Exception as e:
_logger.warning(
'Could not start job, please check your polyaxon spec %s', e)
job.set_status(
JobLifeCycle.FAILED,
message='Could not start job encountered an {} exception.'.format(
e.__class__.__name__))
return
job.definition = get_job_definition(results)
job.save()
def start_notebook(notebook):
# Update job status to show that its started
notebook.set_status(JobLifeCycle.SCHEDULED)
try:
image_name, image_tag = get_image_info(build_job=notebook.build_job)
except ValueError as e:
_logger.warning('Could not start the notebook, %s', e)
notebook.set_status(JobLifeCycle.FAILED, message='External git repo was note found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start notebook with built image `%s`', job_docker_image)
spawner = NotebookSpawner(
project_name=notebook.project.unique_name,
project_uuid=notebook.project.uuid.hex,
job_name=notebook.unique_name,
job_uuid=notebook.uuid.hex,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
in_cluster=True)
try:
results = spawner.start_notebook(image=job_docker_image,
resources=notebook.resources,
node_selectors=notebook.node_selectors)
except ApiException as e:
_logger.warning('Could not start notebook, please check your polyaxon spec %s', e)
notebook.set_status(
JobLifeCycle.FAILED,
message='Could not start notebook, encountered a Kubernetes ApiException.')
return
except Exception as e:
_logger.warning('Could not start notebook, please check your polyaxon spec %s', e)
notebook.set_status(
JobLifeCycle.FAILED,
message='Could not start notebook encountered an {} exception.'.format(
e.__class__.__name__
))
return
notebook.definition = get_job_definition(results)
notebook.save()
def start_experiment(experiment):
# Update experiment status to show that its started
experiment.set_status(ExperimentLifeCycle.SCHEDULED)
project = experiment.project
group = experiment.experiment_group
job_docker_image = None # This will force the spawners to use the default docker image
if experiment.specification.build:
try:
image_name, image_tag = get_image_info(
build_job=experiment.build_job)
except (ValueError, AttributeError):
_logger.error('Could not start the experiment.', exc_info=True)
experiment.set_status(ExperimentLifeCycle.FAILED,
message='Image info was not found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start experiment with built image `%s`',
job_docker_image)
else:
_logger.info('Start experiment with default image.')
spawner_class = get_spawner_class(experiment.specification.framework)
token_scope = RedisEphemeralTokens.get_scope(experiment.user.id,
'experiment', experiment.id)
error = {}
try:
# Use spawners to start the experiment
spawner = spawner_class(
project_name=project.unique_name,
experiment_name=experiment.unique_name,
experiment_group_name=group.unique_name if group else None,
project_uuid=project.uuid.hex,
experiment_group_uuid=group.uuid.hex if group else None,
experiment_uuid=experiment.uuid.hex,
persistence_config=experiment.persistence_config,
outputs_refs_experiments=experiment.outputs_refs_experiments,
outputs_refs_jobs=experiment.outputs_refs_jobs,
original_name=experiment.original_unique_name,
cloning_strategy=experiment.cloning_strategy,
spec=experiment.specification,
k8s_config=conf.get('K8S_CONFIG'),
namespace=conf.get('K8S_NAMESPACE'),
in_cluster=True,
job_docker_image=job_docker_image,
use_sidecar=True,
token_scope=token_scope)
response = spawner.start_experiment()
handle_experiment(experiment=experiment,
spawner=spawner,
response=response)
except ApiException as e:
_logger.error(
'Could not start the experiment, please check your polyaxon spec.',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start the experiment, encountered a Kubernetes ApiException.'
}
except VolumeNotFoundError as e:
_logger.error(
'Could not start the experiment, please check your volume definitions.',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start the experiment, '
'encountered a volume definition problem, %s.' % e
}
except Exception as e:
_logger.error(
'Could not start the experiment, please check your polyaxon spec',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start the experiment encountered an {} exception.'.
format(e.__class__.__name__)
}
finally:
if error.get('raised'):
experiment.set_status(ExperimentLifeCycle.FAILED,
message=error.get('message'),
traceback=error.get('traceback'))
def start_notebook(notebook):
# Update job status to show that its started
notebook.set_status(JobLifeCycle.SCHEDULED)
try:
image_name, image_tag = get_image_info(build_job=notebook.build_job)
except (ValueError, AttributeError):
_logger.error('Could not start the notebook.', exc_info=True)
notebook.set_status(JobLifeCycle.FAILED,
message='Image info was not found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start notebook with built image `%s`', job_docker_image)
spawner = NotebookSpawner(project_name=notebook.project.unique_name,
project_uuid=notebook.project.uuid.hex,
job_name=notebook.unique_name,
job_uuid=notebook.uuid.hex,
k8s_config=conf.get('K8S_CONFIG'),
namespace=conf.get('K8S_NAMESPACE'),
job_docker_image=job_docker_image,
in_cluster=True)
error = {}
try:
mount_code_in_notebooks = conf.get('MOUNT_CODE_IN_NOTEBOOKS')
results = spawner.start_notebook(
persistence_outputs=notebook.persistence_outputs,
persistence_data=notebook.persistence_data,
outputs_refs_jobs=notebook.outputs_refs_jobs,
outputs_refs_experiments=notebook.outputs_refs_experiments,
resources=notebook.resources,
secret_refs=notebook.secret_refs,
configmap_refs=notebook.configmap_refs,
node_selector=notebook.node_selector,
affinity=notebook.affinity,
tolerations=notebook.tolerations,
backend=notebook.backend,
mount_code_in_notebooks=mount_code_in_notebooks)
notebook.definition = get_job_definition(results)
notebook.save(update_fields=['definition'])
return
except ApiException:
_logger.error(
'Could not start notebook, please check your polyaxon spec.',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start the job, encountered a Kubernetes ApiException.',
}
except VolumeNotFoundError as e:
_logger.error(
'Could not start the notebook, please check your volume definitions',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start the job, encountered a volume definition problem. %s'
% e,
}
except Exception as e:
_logger.error(
'Could not start notebook, please check your polyaxon spec.',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start notebook encountered an {} exception.'.format(
e.__class__.__name__)
}
finally:
if error.get('raised'):
notebook.set_status(JobLifeCycle.FAILED,
message=error.get('message'),
traceback=error.get('traceback'))
def start_job(job):
# Update job status to show that its started
job.set_status(JobLifeCycle.SCHEDULED)
try:
image_name, image_tag = get_image_info(build_job=job.build_job)
except (ValueError, AttributeError):
_logger.error('Could not start the job.', exc_info=True)
job.set_status(JobLifeCycle.FAILED,
message='Image info was not found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start job with built image `%s`', job_docker_image)
spawner = JobSpawner(
project_name=job.project.unique_name,
project_uuid=job.project.uuid.hex,
job_name=job.unique_name,
job_uuid=job.uuid.hex,
spec=job.specification,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
job_docker_image=job_docker_image,
in_cluster=True,
use_sidecar=True,
sidecar_config=config.get_requested_params(to_str=True))
error = {}
try:
results = spawner.start_job(
persistence_data=job.persistence_data,
persistence_outputs=job.persistence_outputs,
outputs_refs_jobs=job.outputs_refs_jobs,
outputs_refs_experiments=job.outputs_refs_experiments,
resources=job.resources,
node_selector=job.node_selector,
affinity=job.affinity,
tolerations=job.tolerations)
job.definition = get_job_definition(results)
job.save()
return
except ApiException:
_logger.error('Could not start job, please check your polyaxon spec.',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start the job, encountered a Kubernetes ApiException.',
}
except VolumeNotFoundError as e:
_logger.error(
'Could not start the job, please check your volume definitions.',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start the job, encountered a volume definition problem. %s'
% e,
}
except Exception as e:
_logger.error('Could not start job, please check your polyaxon spec.',
exc_info=True)
error = {
'raised':
True,
'traceback':
traceback.format_exc(),
'message':
'Could not start job encountered an {} exception.'.format(
e.__class__.__name__)
}
finally:
if error.get('raised'):
job.set_status(JobLifeCycle.FAILED,
message=error.get('message'),
traceback=error.get('traceback'))
def test_get_tagged_image(self):
tagged_image = get_tagged_image(self.build_job)
image_name = get_image_info(self.build_job)
assert tagged_image == ':'.join(image_name)
def test_get_image_image_info(self):
image_info = get_image_info(self.build_job)
assert image_info[0] == get_image_name(self.build_job)
assert image_info[1] == self.build_job.uuid.hex
def start_experiment(experiment):
# Update experiment status to show that its started
experiment.set_status(ExperimentLifeCycle.SCHEDULED)
project = experiment.project
group = experiment.experiment_group
job_docker_image = None # This will force the spawners to use the default docker image
if experiment.specification.build:
try:
image_name, image_tag = get_image_info(
build_job=experiment.build_job)
except ValueError as e:
_logger.warning('Could not start the experiment, %s', e)
experiment.set_status(ExperimentLifeCycle.FAILED,
message='External git repo was note found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start experiment with built image `%s`',
job_docker_image)
else:
_logger.info('Start experiment with default image.')
spawner_class = get_spawner_class(experiment.specification.framework)
# Use spawners to start the experiment
spawner = spawner_class(
project_name=project.unique_name,
experiment_name=experiment.unique_name,
experiment_group_name=group.unique_name if group else None,
project_uuid=project.uuid.hex,
experiment_group_uuid=group.uuid.hex if group else None,
experiment_uuid=experiment.uuid.hex,
persistence_config=experiment.persistence_config,
outputs_refs_experiments=experiment.outputs_refs_experiments,
outputs_refs_jobs=experiment.outputs_refs_jobs,
original_name=experiment.original_unique_name,
cloning_strategy=experiment.cloning_strategy,
spec=experiment.specification,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
in_cluster=True,
job_docker_image=job_docker_image,
use_sidecar=True,
sidecar_config=config.get_requested_params(to_str=True))
try:
response = spawner.start_experiment()
except ApiException as e:
_logger.warning(
'Could not start the experiment, please check your polyaxon spec %s',
e)
experiment.set_status(
ExperimentLifeCycle.FAILED,
message=
'Could not start the experiment, encountered a Kubernetes ApiException.'
)
return
except VolumeNotFoundError as e:
_logger.warning(
'Could not start the experiment, '
'please check your volume definitions %s', e)
experiment.set_status(ExperimentLifeCycle.FAILED,
message='Could not start the experiment, '
'encountered a volume definition problem. %s' %
e)
return False
except Exception as e:
_logger.warning(
'Could not start the experiment, please check your polyaxon spec %s',
e)
experiment.set_status(
ExperimentLifeCycle.FAILED,
message='Could not start the experiment encountered an {} exception.'
.format(e.__class__.__name__))
return
handle_experiment(experiment=experiment,
spawner=spawner,
response=response)
def start_notebook(notebook):
# Update job status to show that its started
notebook.set_status(JobLifeCycle.SCHEDULED)
try:
image_name, image_tag = get_image_info(build_job=notebook.build_job)
except (ValueError, AttributeError):
_logger.error('Could not start the notebook.', exc_info=True)
notebook.set_status(JobLifeCycle.FAILED,
message='Image info was not found.')
return
job_docker_image = '{}:{}'.format(image_name, image_tag)
_logger.info('Start notebook with built image `%s`', job_docker_image)
spawner = NotebookSpawner(project_name=notebook.project.unique_name,
project_uuid=notebook.project.uuid.hex,
job_name=notebook.unique_name,
job_uuid=notebook.uuid.hex,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
in_cluster=True)
try:
allow_commits = False
if settings.REPOS_CLAIM_NAME or notebook.node_selector:
allow_commits = True
results = spawner.start_notebook(
image=job_docker_image,
persistence_outputs=notebook.persistence_outputs,
persistence_data=notebook.persistence_data,
outputs_refs_jobs=notebook.outputs_refs_jobs,
outputs_refs_experiments=notebook.outputs_refs_experiments,
resources=notebook.resources,
node_selector=notebook.node_selector,
affinity=notebook.affinity,
tolerations=notebook.tolerations,
allow_commits=allow_commits)
except ApiException:
_logger.error(
'Could not start notebook, please check your polyaxon spec.',
exc_info=True)
notebook.set_status(
JobLifeCycle.FAILED,
message=
'Could not start notebook, encountered a Kubernetes ApiException.')
return
except VolumeNotFoundError as e:
_logger.error(
'Could not start the notebook, please check your volume definitions',
exc_info=True)
notebook.set_status(JobLifeCycle.FAILED,
message='Could not start the notebook, '
'encountered a volume definition problem. %s' % e)
return False
except Exception as e:
_logger.error(
'Could not start notebook, please check your polyaxon spec.',
exc_info=True)
notebook.set_status(
JobLifeCycle.FAILED,
message='Could not start notebook encountered an {} exception.'.
format(e.__class__.__name__))
return
notebook.definition = get_job_definition(results)
notebook.save()
def test_get_tagged_image(self):
tagged_image = get_tagged_image(self.build_job)
image_name = get_image_info(self.build_job)
assert tagged_image == ':'.join(image_name)
def test_get_image_image_info(self):
image_info = get_image_info(self.build_job)
assert image_info[0] == get_image_name(self.build_job)
assert image_info[1] == self.build_job.uuid.hex
