def stop_job(project_name, project_uuid, job_name, job_uuid):
spawner = JobSpawner(project_name=project_name,
project_uuid=project_uuid,
job_name=job_name,
job_uuid=job_uuid,
k8s_config=conf.get(K8S_CONFIG),
namespace=conf.get(K8S_NAMESPACE),
in_cluster=True)
return spawner.stop_job()
def stop_job(project_name, project_uuid, job_name, job_uuid, specification):
spawner = JobSpawner(project_name=project_name,
project_uuid=project_uuid,
job_name=job_name,
job_uuid=job_uuid,
spec=specification,
k8s_config=settings.K8S_CONFIG,
namespace=settings.K8S_NAMESPACE,
in_cluster=True)
return spawner.stop_job()
def stop_job(job, update_status=False):
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,
in_cluster=True)
spawner.stop_job()
if update_status:
# Update experiment status to show that its stopped
job.set_status(status=JobLifeCycle.STOPPED, message='Job was stopped')
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_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 start_job(job):
# Update job status to show that its started
job.set_status(JobLifeCycle.SCHEDULED)
try:
registry_spec = get_registry_context(build_backend=None)
except ContainerRegistryError:
job.set_status(
JobLifeCycle.FAILED,
message=
'Could not start the job, please check your registry configuration.'
)
return
try:
image_name, image_tag = get_image_info(
build_job=job.build_job, registry_host=registry_spec.host)
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,
k8s_config=conf.get(K8S_CONFIG),
namespace=conf.get(K8S_NAMESPACE),
job_docker_image=job_docker_image,
in_cluster=True,
use_sidecar=True,
log_level=job.specification.log_level)
error = {}
try:
results = spawner.start_job(
container_cmd_callback=job.specification.run.get_container_cmd,
persistence_data=job.persistence_data,
persistence_outputs=job.persistence_outputs,
outputs_refs_jobs=job.outputs_refs_jobs,
outputs_refs_experiments=job.outputs_refs_experiments,
secret_refs=job.secret_refs,
config_map_refs=job.config_map_refs,
resources=job.resources,
labels=job.labels,
annotations=job.annotations,
node_selector=job.node_selector,
affinity=job.affinity,
tolerations=job.tolerations,
max_restarts=get_max_restart(job.max_restarts,
conf.get(MAX_RESTARTS_JOBS)),
reconcile_url=get_job_reconcile_url(job.unique_name))
job.definition = get_job_definition(results)
job.save(update_fields=['definition'])
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 StoreNotFoundError 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'))
