def get_env_vars(self):
env_vars = get_service_env_vars(namespace=self.namespace)
for k, v in config.get_requested_params(to_str=True).items():
env_vars.append(get_env_var(name=k, value=v))
# Add private registries secrets keys
for key in config.params_startswith(settings.PRIVATE_REGISTRIES_PREFIX):
env_vars.append(get_from_secret(key, key))
return env_vars
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 job, %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 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))
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 stop_experiment(project_name,
project_uuid,
experiment_name,
experiment_uuid,
specification,
experiment_group_name=None,
experiment_group_uuid=None):
spawner_class = get_spawner_class(specification.framework)
spawner = spawner_class(project_name=project_name,
project_uuid=project_uuid,
experiment_name=experiment_name,
experiment_group_name=experiment_group_name,
experiment_group_uuid=experiment_group_uuid,
experiment_uuid=experiment_uuid,
spec=specification,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
in_cluster=True,
use_sidecar=True,
sidecar_config=config.get_requested_params(to_str=True))
return spawner.stop_experiment()
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 get_env_vars(self):
env_vars = get_service_env_vars(namespace=self.namespace)
for k, v in config.get_requested_params(to_str=True).items():
env_vars.append(get_env_var(name=k, value=v))
return env_vars
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_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)
# 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),
token_scope=token_scope)
error = {}
try:
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'))
