def _create_job(self,
task_type,
task_idx,
add_service,
command=None,
args=None,
env_vars=None,
resources=None,
node_selector=None,
restart_policy='Never'):
job_name = self.pod_manager.get_job_name(task_type=task_type,
task_idx=task_idx)
sidecar_args = get_sidecar_args(pod_id=job_name)
labels = self.pod_manager.get_labels(task_type=task_type,
task_idx=task_idx)
volumes, volume_mounts = get_pod_volumes(
persistence_outputs=self.persistence_config.outputs,
persistence_data=self.persistence_config.data)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=self.outputs_refs_jobs,
persistence_outputs=self.persistence_config.outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=self.outputs_refs_experiments,
persistence_outputs=self.persistence_config.outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
pod = self.pod_manager.get_pod(
task_type=task_type,
task_idx=task_idx,
volume_mounts=volume_mounts,
volumes=volumes,
labels=labels,
env_vars=env_vars,
command=command,
args=args,
sidecar_args=sidecar_args,
persistence_outputs=self.persistence_config.outputs,
persistence_data=self.persistence_config.data,
outputs_refs_jobs=self.outputs_refs_jobs,
outputs_refs_experiments=self.outputs_refs_experiments,
resources=resources,
node_selector=node_selector,
restart_policy=restart_policy)
pod_resp, _ = self.create_or_update_pod(name=job_name, data=pod)
results = {'pod': pod_resp.to_dict()}
if add_service:
service = services.get_service(namespace=self.namespace,
name=job_name,
labels=labels,
ports=self.pod_manager.ports,
target_ports=self.pod_manager.ports)
service_resp, _ = self.create_or_update_service(name=job_name,
data=service)
results['service'] = service_resp.to_dict()
return results
def _create_job(self,
task_type,
task_idx,
add_service,
command=None,
args=None,
sidecar_args_fn=None,
env_vars=None,
resources=None,
node_selector=None,
restart_policy='Never'):
job_name = self.pod_manager.get_job_name(task_type=task_type,
task_idx=task_idx)
sidecar_args = sidecar_args_fn(pod_id=job_name)
labels = self.pod_manager.get_labels(task_type=task_type,
task_idx=task_idx)
volumes, volume_mounts = get_pod_volumes()
pod = self.pod_manager.get_pod(task_type=task_type,
task_idx=task_idx,
volume_mounts=volume_mounts,
volumes=volumes,
env_vars=env_vars,
command=command,
args=args,
sidecar_args=sidecar_args,
resources=resources,
node_selector=node_selector,
restart_policy=restart_policy)
pod_resp, _ = self.create_or_update_pod(name=job_name, data=pod)
service = services.get_service(namespace=self.namespace,
name=job_name,
labels=labels,
ports=self.pod_manager.ports,
target_ports=self.pod_manager.ports)
results = {'pod': pod_resp.to_dict()}
if add_service:
service_resp, _ = self.create_or_update_service(name=job_name,
data=service)
results['service'] = service_resp.to_dict()
return results
def _create_job(self,
task_type,
task_idx,
add_service,
command=None,
args=None,
env_vars=None,
resources=None,
node_selector=None,
restart_policy='Never'):
job_name = self.pod_manager.get_job_name(task_type=task_type, task_idx=task_idx)
sidecar_args = get_sidecar_args(pod_id=job_name)
labels = self.pod_manager.get_labels(task_type=task_type, task_idx=task_idx)
volumes, volume_mounts = get_pod_volumes()
pod = self.pod_manager.get_pod(task_type=task_type,
task_idx=task_idx,
volume_mounts=volume_mounts,
volumes=volumes,
env_vars=env_vars,
command=command,
args=args,
sidecar_args=sidecar_args,
resources=resources,
node_selector=node_selector,
restart_policy=restart_policy)
pod_resp, _ = self.create_or_update_pod(name=job_name, data=pod)
service = services.get_service(namespace=self.namespace,
name=job_name,
labels=labels,
ports=self.pod_manager.ports,
target_ports=self.pod_manager.ports)
results = {'pod': pod_resp.to_dict()}
if add_service:
service_resp, _ = self.create_or_update_service(name=job_name, data=service)
results['service'] = service_resp.to_dict()
return results
def start_notebook(self,
persistence_outputs=None,
persistence_data=None,
outputs_refs_jobs=None,
outputs_refs_experiments=None,
resources=None,
secret_refs=None,
configmap_refs=None,
node_selector=None,
affinity=None,
tolerations=None,
backend=None,
mount_code_in_notebooks=False):
ports = [self.request_notebook_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_volumes(persistence_outputs=persistence_outputs,
persistence_data=persistence_data)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_jobs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_experiments,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
shm_volumes, shm_volume_mounts = get_shm_volumes()
volumes += shm_volumes
volume_mounts += shm_volume_mounts
context_volumes, context_mounts = get_auth_context_volumes()
volumes += context_volumes
volume_mounts += context_mounts
if mount_code_in_notebooks:
code_volume, code_volume_mount = self.get_notebook_code_volume()
volumes.append(code_volume)
volume_mounts.append(code_volume_mount)
secret_refs = validate_secret_refs(secret_refs)
configmap_refs = validate_configmap_refs(configmap_refs)
resource_name = self.resource_manager.get_resource_name()
args = self.get_notebook_args(deployment_name=resource_name,
ports=ports,
mount_code_in_notebooks=mount_code_in_notebooks,
backend=backend)
command = ["/bin/sh", "-c"]
deployment = self.resource_manager.get_deployment(
resource_name=resource_name,
volume_mounts=volume_mounts,
volumes=volumes,
labels=self.resource_manager.labels,
env_vars=None,
command=command,
args=args,
init_env_vars=self.get_init_env_vars(),
persistence_outputs=persistence_outputs,
persistence_data=persistence_data,
outputs_refs_jobs=outputs_refs_jobs,
outputs_refs_experiments=outputs_refs_experiments,
secret_refs=secret_refs,
configmap_refs=configmap_refs,
resources=resources,
ephemeral_token=None,
node_selector=node_selector,
affinity=affinity,
tolerations=tolerations,
ports=target_ports,
init_context_mounts=context_mounts,
restart_policy='Never')
dep_resp, _ = self.create_or_update_deployment(name=resource_name,
data=deployment)
service = services.get_service(
namespace=self.namespace,
name=resource_name,
labels=self.resource_manager.get_labels(),
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=resource_name, data=service)
results = {'deployment': dep_resp.to_dict(), 'service': service_resp.to_dict()}
if self._use_ingress():
annotations = json.loads(conf.get('K8S_INGRESS_ANNOTATIONS'))
paths = [{
'path': '/notebooks/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': resource_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(namespace=self.namespace,
name=resource_name,
labels=self.resource_manager.get_labels(),
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=resource_name, data=ingress)
return results
def start_notebook(self,
image,
persistence_outputs=None,
persistence_data=None,
outputs_refs_jobs=None,
outputs_refs_experiments=None,
resources=None,
secret_refs=None,
configmap_refs=None,
node_selector=None,
affinity=None,
tolerations=None,
allow_commits=False):
ports = [self.request_notebook_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_volumes(
persistence_outputs=persistence_outputs,
persistence_data=persistence_data)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_jobs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_experiments,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
shm_volumes, shm_volume_mounts = get_shm_volumes()
volumes += shm_volumes
volume_mounts += shm_volume_mounts
env_vars = get_job_env_vars(
persistence_outputs=persistence_outputs,
outputs_path=get_notebook_job_outputs_path(
persistence_outputs=persistence_outputs,
notebook_job=self.job_name),
persistence_data=persistence_data,
outputs_refs_jobs=outputs_refs_jobs,
outputs_refs_experiments=outputs_refs_experiments)
secret_refs = validate_secret_refs(secret_refs)
configmap_refs = validate_configmap_refs(configmap_refs)
env_from = get_pod_env_from(secret_refs=secret_refs,
configmap_refs=configmap_refs)
code_volume, code_volume_mount = self.get_notebook_code_volume()
volumes.append(code_volume)
volume_mounts.append(code_volume_mount)
deployment_name = JOB_NAME_FORMAT.format(name=NOTEBOOK_JOB_NAME,
job_uuid=self.job_uuid)
node_selector = get_node_selector(
node_selector=node_selector,
default_node_selector=settings.NODE_SELECTOR_EXPERIMENTS)
affinity = get_affinity(affinity=affinity,
default_affinity=settings.AFFINITY_EXPERIMENTS)
tolerations = get_tolerations(
tolerations=tolerations,
default_tolerations=settings.TOLERATIONS_EXPERIMENTS)
deployment = deployments.get_deployment(
namespace=self.namespace,
app=settings.APP_LABELS_NOTEBOOK,
name=NOTEBOOK_JOB_NAME,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
volume_mounts=volume_mounts,
volumes=volumes,
image=image,
command=["/bin/sh", "-c"],
args=self.get_notebook_args(deployment_name=deployment_name,
ports=ports,
allow_commits=allow_commits),
ports=target_ports,
container_name=settings.CONTAINER_NAME_PLUGIN_JOB,
env_vars=env_vars,
env_from=env_from,
resources=resources,
node_selector=node_selector,
affinity=affinity,
tolerations=tolerations,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_RUNNER,
service_account_name=settings.K8S_SERVICE_ACCOUNT_EXPERIMENTS)
deployment_labels = deployments.get_labels(
app=settings.APP_LABELS_NOTEBOOK,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_RUNNER)
dep_resp, _ = self.create_or_update_deployment(name=deployment_name,
data=deployment)
service = services.get_service(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=deployment_name,
data=service)
results = {
'deployment': dep_resp.to_dict(),
'service': service_resp.to_dict()
}
if self._use_ingress():
annotations = json.loads(settings.K8S_INGRESS_ANNOTATIONS)
paths = [{
'path':
'/notebook/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': deployment_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=deployment_name, data=ingress)
return results
def start_tensorboard(self,
image,
outputs_path,
resources=None,
node_selectors=None):
ports = [self.request_tensorboard_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_volumes()
deployment = deployments.get_deployment(
namespace=self.namespace,
app=settings.APP_LABELS_TENSORBOARD,
name=self.TENSORBOARD_JOB_NAME,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
volume_mounts=volume_mounts,
volumes=volumes,
image=image,
command=["/bin/sh", "-c"],
args=[
"tensorboard --logdir={} --port={}".format(
outputs_path, self.PORT)
],
ports=target_ports,
container_name=settings.CONTAINER_NAME_PLUGIN_JOB,
resources=resources,
node_selector=node_selectors,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
deployment_name = constants.JOB_NAME.format(
name=self.TENSORBOARD_JOB_NAME, job_uuid=self.job_uuid)
deployment_labels = deployments.get_labels(
app=settings.APP_LABELS_TENSORBOARD,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
dep_resp, _ = self.create_or_update_deployment(name=deployment_name,
data=deployment)
service = services.get_service(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=deployment_name,
data=service)
results = {
'deployment': dep_resp.to_dict(),
'service': service_resp.to_dict()
}
if self._use_ingress():
annotations = json.loads(settings.K8S_INGRESS_ANNOTATIONS)
paths = [{
'path':
'/tensorboard/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': deployment_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=deployment_name, data=ingress)
return results
def _create_job(self,
task_type,
task_idx,
add_service,
command=None,
args=None,
env_vars=None,
resources=None,
annotations=None,
node_selector=None,
affinity=None,
tolerations=None,
max_restarts=None):
ephemeral_token = None
if self.token_scope:
ephemeral_token = RedisEphemeralTokens.generate_header_token(
scope=self.token_scope)
resource_name = self.resource_manager.get_resource_name(
task_type=task_type, task_idx=task_idx)
job_uuid = self.get_job_uuids(task_type=task_type, task_idx=task_idx)
reconcile_url = get_experiment_reconcile_url(self.experiment_name,
job_uuid)
labels = self.get_labels(task_type=task_type,
task_idx=task_idx,
job_uuid=job_uuid)
# Set and validate volumes
volumes, volume_mounts = get_pod_volumes(
persistence_outputs=self.persistence_config.outputs,
persistence_data=self.persistence_config.data)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=self.outputs_refs_jobs,
persistence_outputs=self.persistence_config.outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=self.outputs_refs_experiments,
persistence_outputs=self.persistence_config.outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
shm_volumes, shm_volume_mounts = get_shm_volumes()
volumes += shm_volumes
volume_mounts += shm_volume_mounts
context_volumes, context_mounts = get_auth_context_volumes()
volumes += context_volumes
volume_mounts += context_mounts
pod = self.resource_manager.get_task_pod(
task_type=task_type,
task_idx=task_idx,
volume_mounts=volume_mounts,
volumes=volumes,
labels=labels,
env_vars=env_vars,
command=command,
args=args,
ports=self.ports,
init_env_vars=self.get_init_env_vars(),
persistence_outputs=self.persistence_config.outputs,
persistence_data=self.persistence_config.data,
outputs_refs_jobs=self.outputs_refs_jobs,
outputs_refs_experiments=self.outputs_refs_experiments,
secret_refs=self.spec.secret_refs,
config_map_refs=self.spec.config_map_refs,
resources=resources,
ephemeral_token=ephemeral_token,
node_selector=node_selector,
affinity=affinity,
tolerations=tolerations,
init_context_mounts=context_mounts,
reconcile_url=reconcile_url,
max_restarts=max_restarts,
restart_policy=get_pod_restart_policy(max_restarts))
pod_resp, _ = self.create_or_update_pod(name=resource_name,
body=pod,
reraise=True)
results = {'pod': pod_resp.to_dict()}
if add_service:
service = services.get_service(namespace=self.namespace,
name=resource_name,
labels=labels,
ports=self.ports,
target_ports=self.ports)
service_resp, _ = self.create_or_update_service(name=resource_name,
body=service,
reraise=True)
results['service'] = service_resp.to_dict()
return results
def start_notebook(self, image, resources=None, node_selectors=None):
ports = [self.request_notebook_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_volumes()
code_volume, code_volume_mount = self.get_notebook_code_volume()
volumes.append(code_volume)
volume_mounts.append(code_volume_mount)
deployment_name = constants.JOB_NAME.format(name=self.NOTEBOOK_JOB_NAME,
job_uuid=self.job_uuid)
notebook_token = self.get_notebook_token()
notebook_url = self._get_proxy_url(
namespace=self.namespace,
job_name=self.NOTEBOOK_JOB_NAME,
deployment_name=deployment_name,
port=ports[0])
notebook_dir = get_project_repos_path(self.project_name)
notebook_dir = '{}/{}'.format(notebook_dir, notebook_dir.split('/')[-1])
deployment = deployments.get_deployment(
namespace=self.namespace,
app=settings.APP_LABELS_NOTEBOOK,
name=self.NOTEBOOK_JOB_NAME,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
volume_mounts=volume_mounts,
volumes=volumes,
image=image,
command=["/bin/sh", "-c"],
args=[
"jupyter notebook "
"--no-browser "
"--port={port} "
"--ip=0.0.0.0 "
"--allow-root "
"--NotebookApp.token={token} "
"--NotebookApp.trust_xheaders=True "
"--NotebookApp.base_url={base_url} "
"--NotebookApp.notebook_dir={notebook_dir} ".format(
port=self.PORT,
token=notebook_token,
base_url=notebook_url,
notebook_dir=notebook_dir)],
ports=target_ports,
container_name=settings.CONTAINER_NAME_PLUGIN_JOB,
resources=resources,
node_selector=node_selectors,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
deployment_labels = deployments.get_labels(app=settings.APP_LABELS_NOTEBOOK,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
dep_resp, _ = self.create_or_update_deployment(name=deployment_name, data=deployment)
service = services.get_service(
namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=deployment_name, data=service)
results = {'deployment': dep_resp.to_dict(), 'service': service_resp.to_dict()}
if self._use_ingress():
annotations = json.loads(settings.K8S_INGRESS_ANNOTATIONS)
paths = [{
'path': '/notebook/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': deployment_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=deployment_name, data=ingress)
return results
def _create_job(self,
task_type,
task_idx,
add_service,
command=None,
args=None,
env_vars=None,
resources=None,
node_selector=None,
affinity=None,
tolerations=None,
restart_policy='Never'):
ephemeral_token = RedisEphemeralTokens.generate_header_token(
scope=self.token_scope)
job_name = self.pod_manager.get_job_name(task_type=task_type,
task_idx=task_idx)
sidecar_args = get_sidecar_args(pod_id=job_name)
labels = self.pod_manager.get_labels(task_type=task_type,
task_idx=task_idx)
# Set and validate volumes
volumes, volume_mounts = get_pod_volumes(
persistence_outputs=self.persistence_config.outputs,
persistence_data=self.persistence_config.data)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=self.outputs_refs_jobs,
persistence_outputs=self.persistence_config.outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=self.outputs_refs_experiments,
persistence_outputs=self.persistence_config.outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
shm_volumes, shm_volume_mounts = get_shm_volumes()
volumes += shm_volumes
volume_mounts += shm_volume_mounts
# Validate secret and configmap refs
secret_refs = validate_secret_refs(self.spec.secret_refs)
configmap_refs = validate_configmap_refs(self.spec.configmap_refs)
pod = self.pod_manager.get_pod(
task_type=task_type,
task_idx=task_idx,
volume_mounts=volume_mounts,
volumes=volumes,
labels=labels,
env_vars=env_vars,
command=command,
args=args,
sidecar_args=sidecar_args,
persistence_outputs=self.persistence_config.outputs,
persistence_data=self.persistence_config.data,
outputs_refs_jobs=self.outputs_refs_jobs,
outputs_refs_experiments=self.outputs_refs_experiments,
secret_refs=secret_refs,
configmap_refs=configmap_refs,
resources=resources,
ephemeral_token=ephemeral_token,
node_selector=node_selector,
affinity=affinity,
tolerations=tolerations,
restart_policy=restart_policy)
pod_resp, _ = self.create_or_update_pod(name=job_name, data=pod)
results = {'pod': pod_resp.to_dict()}
if add_service:
service = services.get_service(namespace=self.namespace,
name=job_name,
labels=labels,
ports=self.pod_manager.ports,
target_ports=self.pod_manager.ports)
service_resp, _ = self.create_or_update_service(name=job_name,
data=service)
results['service'] = service_resp.to_dict()
return results
def start_tensorboard(self,
outputs_path,
persistence_outputs,
outputs_specs=None,
outputs_refs_jobs=None,
outputs_refs_experiments=None,
resources=None,
node_selector=None,
affinity=None,
tolerations=None):
ports = [self.request_tensorboard_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_outputs_volume(persistence_outputs)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_jobs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_specs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_experiments,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
# Add volumes for persistence outputs secrets
stores_secrets = get_stores_secrets(specs=outputs_specs)
self.validate_stores_secrets_keys(stores_secrets=stores_secrets)
secrets_volumes, secrets_volume_mounts = self.get_stores_secrets_volumes(
stores_secrets=stores_secrets)
volumes += secrets_volumes
volume_mounts += secrets_volume_mounts
resource_name = self.resource_manager.get_resource_name()
# Get persistence outputs secrets auth commands
command_args = self.get_stores_secrets_command_args(
stores_secrets=stores_secrets)
command_args.append("tensorboard --logdir={} --port={}".format(
outputs_path, self.PORT))
args = [' && '.join(command_args)]
command = ["/bin/sh", "-c"]
deployment = self.resource_manager.get_deployment(
resource_name=resource_name,
volume_mounts=volume_mounts,
volumes=volumes,
labels=self.resource_manager.labels,
env_vars=None,
command=command,
args=args,
persistence_outputs=persistence_outputs,
outputs_refs_jobs=outputs_refs_jobs,
outputs_refs_experiments=outputs_refs_experiments,
resources=resources,
ephemeral_token=None,
node_selector=node_selector,
affinity=affinity,
tolerations=tolerations,
ports=target_ports,
restart_policy='Never')
dep_resp, _ = self.create_or_update_deployment(name=resource_name,
data=deployment)
service = services.get_service(
namespace=self.namespace,
name=resource_name,
labels=self.resource_manager.get_labels(),
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=resource_name,
data=service)
results = {
'deployment': dep_resp.to_dict(),
'service': service_resp.to_dict()
}
if self._use_ingress():
annotations = json.loads(conf.get('K8S_INGRESS_ANNOTATIONS'))
paths = [{
'path':
'/tensorboards/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': resource_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(
namespace=self.namespace,
name=resource_name,
labels=self.resource_manager.get_labels(),
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=resource_name, data=ingress)
return results
def start_tensorboard(self,
image,
outputs_path,
persistence_outputs,
outputs_specs=None,
outputs_refs_jobs=None,
outputs_refs_experiments=None,
resources=None,
node_selector=None,
affinity=None,
tolerations=None):
ports = [self.request_tensorboard_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_outputs_volume(persistence_outputs)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_jobs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_specs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_experiments,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
# Add volumes for persistence outputs secrets
stores_secrets = get_stores_secrets(specs=outputs_specs)
self.validate_stores_secrets_keys(stores_secrets=stores_secrets)
secrets_volumes, secrets_volume_mounts = self.get_stores_secrets_volumes(
stores_secrets=stores_secrets)
volumes += secrets_volumes
volume_mounts += secrets_volume_mounts
# Get persistence outputs secrets auth commands
command_args = self.get_stores_secrets_command_args(
stores_secrets=stores_secrets)
command_args.append("tensorboard --logdir={} --port={}".format(
outputs_path, self.PORT))
node_selector = get_node_selector(
node_selector=node_selector,
default_node_selector=settings.NODE_SELECTOR_TENSORBOARDS)
affinity = get_affinity(
affinity=affinity, default_affinity=settings.AFFINITY_TENSORBOARDS)
tolerations = get_tolerations(
tolerations=tolerations,
default_tolerations=settings.TOLERATIONS_TENSORBOARDS)
deployment = deployments.get_deployment(
namespace=self.namespace,
app=settings.APP_LABELS_TENSORBOARD,
name=TENSORBOARD_JOB_NAME,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
volume_mounts=volume_mounts,
volumes=volumes,
image=image,
command=["/bin/sh", "-c"],
args=[' && '.join(command_args)],
ports=target_ports,
container_name=settings.CONTAINER_NAME_PLUGIN_JOB,
resources=resources,
node_selector=node_selector,
affinity=affinity,
tolerations=tolerations,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_RUNNER)
deployment_name = JOB_NAME_FORMAT.format(name=TENSORBOARD_JOB_NAME,
job_uuid=self.job_uuid)
deployment_labels = deployments.get_labels(
app=settings.APP_LABELS_TENSORBOARD,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_RUNNER)
dep_resp, _ = self.create_or_update_deployment(name=deployment_name,
data=deployment)
service = services.get_service(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=deployment_name,
data=service)
results = {
'deployment': dep_resp.to_dict(),
'service': service_resp.to_dict()
}
if self._use_ingress():
annotations = json.loads(settings.K8S_INGRESS_ANNOTATIONS)
paths = [{
'path':
'/tensorboard/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': deployment_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=deployment_name, data=ingress)
return results
def start_tensorboard(self,
outputs_path,
persistence_outputs,
outputs_specs=None,
outputs_refs_jobs=None,
outputs_refs_experiments=None,
resources=None,
labels=None,
annotations=None,
node_selector=None,
affinity=None,
tolerations=None,
max_restarts=None,
reconcile_url=None):
ports = [self.request_tensorboard_port()]
target_ports = [self.port]
volumes, volume_mounts = get_pod_outputs_volume(persistence_outputs)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_jobs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_specs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_experiments,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
# Add volumes for persistence outputs secrets
stores_secrets = get_stores_secrets(specs=outputs_specs)
self.validate_stores_secrets_keys(stores_secrets=stores_secrets)
secrets_volumes, secrets_volume_mounts = self.get_stores_secrets_volumes(
stores_secrets=stores_secrets)
volumes += secrets_volumes
volume_mounts += secrets_volume_mounts
resource_name = self.resource_manager.get_resource_name()
tensorboard_url = self._get_proxy_url(namespace=self.namespace,
job_name=TENSORBOARD_JOB_NAME,
deployment_name=resource_name)
# Get persistence outputs secrets auth commands
command_args = self.get_stores_secrets_command_args(
stores_secrets=stores_secrets)
command_args.append("tensorboard "
"--logdir={log_dir} "
"--port={port} "
"--path_prefix={path_prefix}".format(
log_dir=outputs_path,
port=self.port,
path_prefix=tensorboard_url))
args = [' && '.join(command_args)]
command = ["/bin/sh", "-c"]
labels = get_labels(default_labels=self.resource_manager.labels,
labels=labels)
deployment = self.resource_manager.get_deployment(
resource_name=resource_name,
volume_mounts=volume_mounts,
volumes=volumes,
labels=labels,
env_vars=None,
command=command,
args=args,
persistence_outputs=persistence_outputs,
outputs_refs_jobs=outputs_refs_jobs,
outputs_refs_experiments=outputs_refs_experiments,
resources=resources,
annotations=annotations,
ephemeral_token=None,
node_selector=node_selector,
affinity=affinity,
tolerations=tolerations,
ports=target_ports,
reconcile_url=reconcile_url,
max_restarts=max_restarts,
restart_policy=get_deployment_restart_policy(max_restarts))
dep_resp, _ = self.create_or_update_deployment(name=resource_name,
body=deployment,
reraise=True)
service = services.get_service(
namespace=self.namespace,
name=resource_name,
labels=self.resource_manager.get_labels(),
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=resource_name,
body=service,
reraise=True)
results = {
'deployment': dep_resp.to_dict(),
'service': service_resp.to_dict()
}
if self._use_ingress():
annotations = json.loads(conf.get(K8S_INGRESS_ANNOTATIONS))
paths = [{
'path':
'/tensorboards/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': resource_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(
namespace=self.namespace,
name=resource_name,
labels=self.resource_manager.get_labels(),
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=resource_name,
body=ingress,
reraise=True)
return results
def start_tensorboard(self, image, resources=None, node_selectors=None):
ports = [self.request_tensorboard_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_volumes()
outputs_path = get_project_outputs_path(project_name=self.project_name)
deployment = deployments.get_deployment(
namespace=self.namespace,
app=settings.APP_LABELS_TENSORBOARD,
name=self.TENSORBOARD_JOB_NAME,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
volume_mounts=volume_mounts,
volumes=volumes,
image=image,
command=["/bin/sh", "-c"],
args=["tensorboard --logdir={} --port={}".format(outputs_path, self.PORT)],
ports=target_ports,
container_name=settings.CONTAINER_NAME_PLUGIN_JOB,
resources=resources,
node_selector=node_selectors,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
deployment_name = constants.JOB_NAME.format(name=self.TENSORBOARD_JOB_NAME,
job_uuid=self.job_uuid)
deployment_labels = deployments.get_labels(app=settings.APP_LABELS_TENSORBOARD,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
dep_resp, _ = self.create_or_update_deployment(name=deployment_name, data=deployment)
service = services.get_service(
namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=deployment_name, data=service)
results = {'deployment': dep_resp.to_dict(), 'service': service_resp.to_dict()}
if self._use_ingress():
annotations = json.loads(settings.K8S_INGRESS_ANNOTATIONS)
paths = [{
'path': '/tensorboard/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': deployment_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=deployment_name, data=ingress)
return results
def start_notebook(self,
image,
persistence_outputs=None,
persistence_data=None,
outputs_refs_jobs=None,
outputs_refs_experiments=None,
resources=None,
node_selectors=None,
allow_commits=False):
ports = [self.request_notebook_port()]
target_ports = [self.PORT]
volumes, volume_mounts = get_pod_volumes(
persistence_outputs=persistence_outputs,
persistence_data=persistence_data)
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_jobs,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
refs_volumes, refs_volume_mounts = get_pod_refs_outputs_volumes(
outputs_refs=outputs_refs_experiments,
persistence_outputs=persistence_outputs)
volumes += refs_volumes
volume_mounts += refs_volume_mounts
env_vars = get_job_env_vars(
outputs_path=get_notebook_job_outputs_path(
persistence_outputs=persistence_outputs,
notebook_job=self.job_name),
data_paths=get_data_paths(persistence_data),
outputs_refs_jobs=outputs_refs_jobs,
outputs_refs_experiments=outputs_refs_experiments)
code_volume, code_volume_mount = self.get_notebook_code_volume()
volumes.append(code_volume)
volume_mounts.append(code_volume_mount)
deployment_name = constants.JOB_NAME.format(
name=self.NOTEBOOK_JOB_NAME, job_uuid=self.job_uuid)
deployment = deployments.get_deployment(
namespace=self.namespace,
app=settings.APP_LABELS_NOTEBOOK,
name=self.NOTEBOOK_JOB_NAME,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
volume_mounts=volume_mounts,
volumes=volumes,
image=image,
command=["/bin/sh", "-c"],
args=self.get_notebook_args(deployment_name=deployment_name,
ports=ports,
allow_commits=allow_commits),
ports=target_ports,
container_name=settings.CONTAINER_NAME_PLUGIN_JOB,
env_vars=env_vars,
resources=resources,
node_selector=node_selectors,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
deployment_labels = deployments.get_labels(
app=settings.APP_LABELS_NOTEBOOK,
project_name=self.project_name,
project_uuid=self.project_uuid,
job_name=self.job_name,
job_uuid=self.job_uuid,
role=settings.ROLE_LABELS_DASHBOARD,
type=settings.TYPE_LABELS_EXPERIMENT)
dep_resp, _ = self.create_or_update_deployment(name=deployment_name,
data=deployment)
service = services.get_service(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
ports=ports,
target_ports=target_ports,
service_type=self._get_service_type())
service_resp, _ = self.create_or_update_service(name=deployment_name,
data=service)
results = {
'deployment': dep_resp.to_dict(),
'service': service_resp.to_dict()
}
if self._use_ingress():
annotations = json.loads(settings.K8S_INGRESS_ANNOTATIONS)
paths = [{
'path':
'/notebook/{}'.format(self.project_name.replace('.', '/')),
'backend': {
'serviceName': deployment_name,
'servicePort': ports[0]
}
}]
ingress = ingresses.get_ingress(namespace=self.namespace,
name=deployment_name,
labels=deployment_labels,
annotations=annotations,
paths=paths)
self.create_or_update_ingress(name=deployment_name, data=ingress)
return results
