def get_by_url(ce: CEClient, image_url: str):
image_url = ImageUrl(image_url)
if image_url.is_family:
image_data = ce.get_image_from_family(
family_name=image_url.name, project_id=image_url.project_id)
else:
res = ce.list_images(image_name=image_url.name,
project_id=image_url.project_id)
image_data = res[0] if res else None
if not image_data:
return None
return Image(image_data)
def get_by_name(ce: CEClient, machine_name: str):
"""Returns an instance by its stack name."""
res = ce.list_instances(machine_name)
if not res:
return None
return Instance(ce, res[0])
def get_by_name(ce: CEClient, disk_name: str):
"""Returns a disk by its name."""
res = ce.list_disks(disk_name)
if not res:
return None
return Disk(ce, res[0])
def get_by_name(ce: CEClient, image_name: str):
"""Returns an image by its name."""
res = ce.list_images(image_name)
if not res:
return None
return Image(res[0])
def get_by_name(ce: CEClient, snapshot_name: str):
"""Returns a snapshot by its name."""
res = ce.list_snapshots(snapshot_name)
if not res:
return None
return Snapshot(res[0])
def __init__(self, machine_name: str, project_id: str, zone: str):
self._dm = DMClient(project_id, zone)
self._ce = CEClient(project_id, zone)
self._rtc = RtcClient(project_id, zone)
self._machine_name = machine_name
self._stack_name = 'spotty-instance-' + machine_name
# resource names
self._INSTANCE_RESOURCE_NAME = machine_name
self._DOCKER_WAITER_RESOURCE_NAME = machine_name + '-docker-waiter'
self._DOCKER_STATUS_CONFIG_RESOURCE_NAME = machine_name + '-docker-status'
def check_gpu_configuration(ce: CEClient, gpu_parameters: dict):
if not gpu_parameters:
return
# check GPU type
accelerator_types = ce.get_accelerator_types()
gpu_type = gpu_parameters['type']
if gpu_type not in accelerator_types:
if accelerator_types:
error_msg = 'GPU type "%s" is not supported in the "%s" zone.\nAvailable GPU types are: %s.' \
% (gpu_type, ce.zone, ', '.join(accelerator_types.keys()))
else:
error_msg = 'The "%s" zone doesn\'t support any GPU accelerators.' % ce.zone
raise ValueError(error_msg)
# check the number of GPUs is not exceed the maximum
max_cards_per_instance = accelerator_types[gpu_parameters['type']]
if gpu_parameters['count'] > max_cards_per_instance:
raise ValueError(
'Maximum allowed number of cards per instance for the "%s" type is %d.'
% (gpu_parameters['type'], max_cards_per_instance))
def create_disks(ce: CEClient,
volumes: List[AbstractInstanceVolume],
output: AbstractOutputWriter,
dry_run: bool = False):
disks_to_create = []
# do some checks and prepare disk parameters
for i, volume in enumerate(volumes):
if isinstance(volume, DiskVolume):
# check if the disk already exists
disk = Disk.get_by_name(ce, volume.disk_name)
if disk:
# check if the volume is available
if not disk.is_available():
raise ValueError(
'Disk "%s" is not available (status: %s).' %
(volume.disk_name, disk.status))
# check size of the volume
if volume.size and (volume.size != disk.size):
raise ValueError(
'Specified size for the "%s" volume (%dGB) doesn\'t match the size of the '
'existing disk (%dGB).' %
(volume.name, volume.size, disk.size))
output.write('- disk "%s" will be attached' % disk.name)
else:
# check if the snapshot exists
snapshot = Snapshot.get_by_name(ce, volume.disk_name)
if snapshot:
# disk will be restored from the snapshot
# check size of the volume
if volume.size and (volume.size < snapshot.size):
raise ValueError(
'Specified size for the "%s" volume (%dGB) is less than size of the '
'snapshot (%dGB).' %
(volume.name, volume.size, snapshot.size))
output.write(
'- disk "%s" will be restored from the snapshot' %
volume.disk_name)
disks_to_create.append(
(volume.disk_name, volume.size, snapshot.self_link))
else:
# empty volume will be created, check that the size is specified
if not volume.size:
raise ValueError('Size for the new disk is required.')
if volume.size < 10:
raise ValueError(
'Size of a disk cannot be less than 10GB.')
disks_to_create.append(
(volume.disk_name, volume.size, None))
# create disks
for disk_name, disk_size, snapshot_link in disks_to_create:
if not dry_run:
ce.create_disk(disk_name, disk_size, snapshot_link)
output.write('- disk "%s" was created' % disk_name)
def __init__(self, image_name: str, project_id: str, zone: str):
self._dm = DMClient(project_id, zone)
self._ce = CEClient(project_id, zone)
self._image_name = image_name
self._stack_name = 'spotty-image-%s' % image_name
def __init__(self, instance_config: InstanceConfig):
super().__init__(instance_config)
self._project_name = instance_config.project_config.project_name
self._credentials = GcpCredentials()
self._ce = CEClient(self._credentials.project_id, instance_config.zone)
class InstanceDeployment(AbstractInstanceDeployment):
instance_config: InstanceConfig
def __init__(self, instance_config: InstanceConfig):
super().__init__(instance_config)
self._project_name = instance_config.project_config.project_name
self._credentials = GcpCredentials()
self._ce = CEClient(self._credentials.project_id, instance_config.zone)
@property
def stack_manager(self) -> InstanceStackManager:
return InstanceStackManager(self.instance_config.machine_name,
self._credentials.project_id,
self.instance_config.zone)
@property
def ssh_key_manager(self) -> SshKeyManager:
return SshKeyManager(self._project_name, self.instance_config.zone)
def get_instance(self) -> Instance:
return Instance.get_by_name(self._ce,
self.instance_config.machine_name)
def deploy(self,
container_commands: DockerCommands,
bucket_name: str,
data_transfer: DataTransfer,
output: AbstractOutputWriter,
dry_run: bool = False):
# check machine type
if not self._ce.get_machine_types(self.instance_config.machine_type):
raise ValueError(
'"%s" machine type is not available in the "%s" zone.' %
(self.instance_config.machine_type, self.instance_config.zone))
# check GPU configuration
check_gpu_configuration(self._ce, self.instance_config.gpu)
# remove the stack it it exists to make all the disks available
stack_manager = self.stack_manager
stack_manager.delete_stack(output=output)
# sync the project with the S3 bucket
if bucket_name is not None:
output.write('Syncing the project with the bucket...')
data_transfer.upload_local_to_bucket(bucket_name, dry_run=dry_run)
# create volumes
if self.instance_config.volumes:
# create disks
output.write('\nCreating disks...')
with output.prefix(' '):
create_disks(self._ce,
self.instance_config.volumes,
output=output,
dry_run=dry_run)
output.write('')
# prepare Deployment Manager template
output.write('Preparing the deployment template...')
with output.prefix(' '):
# get an image
image_link = get_image(self._ce, self.instance_config.image_uri,
self.instance_config.image_name).self_link
# get or create an SSH key
public_key_value = self.ssh_key_manager.get_public_key_value()
# prepare the deployment template
sync_project_cmd = data_transfer.get_download_bucket_to_instance_command(
bucket_name=bucket_name)
template = prepare_instance_template(
instance_config=self.instance_config,
docker_commands=container_commands,
image_link=image_link,
bucket_name=bucket_name,
sync_project_cmd=sync_project_cmd,
public_key_value=public_key_value,
service_account_email=self._credentials.service_account_email,
output=output,
)
output.write('')
# print information about the volumes
output.write('Volumes:\n%s\n' % render_volumes_info_table(
self.instance_config.volume_mounts, self.instance_config.volumes))
# create stack
if not dry_run:
stack_manager.create_stack(template, output=output)
def delete(self, output: AbstractOutputWriter):
self.stack_manager.delete_stack(output)
def __init__(self, project_name: str, instance_config: InstanceConfig):
self._project_name = project_name
self._instance_config = instance_config
self._credentials = GcpCredentials()
self._ce = CEClient(self._credentials.project_id, instance_config.zone)
