def list_clusters(region=None, next_token=None, cluster_status=None):
"""
Retrieve the list of existing clusters managed by the API. Deleted clusters are not listed by default.
:param region: List clusters deployed to a given AWS Region.
:type region: str
:param next_token: Token to use for paginated requests.
:type next_token: str
:param cluster_status: Filter by cluster status. (Defaults to all clusters.)
:type cluster_status: list | bytes
:rtype: ListClustersResponseContent
"""
stacks, next_token = AWSApi.instance().cfn.list_pcluster_stacks(next_token=next_token)
stacks = [ClusterStack(stack) for stack in stacks]
clusters = []
for stack in stacks:
current_cluster_status = cloud_formation_status_to_cluster_status(stack.status)
if not cluster_status or current_cluster_status in cluster_status:
cluster_info = ClusterInfoSummary(
cluster_name=stack.cluster_name,
cloudformation_stack_status=stack.status,
cloudformation_stack_arn=stack.id,
region=os.environ.get("AWS_DEFAULT_REGION"),
version=stack.version,
cluster_status=current_cluster_status,
)
clusters.append(cluster_info)
return ListClustersResponseContent(clusters=clusters, next_token=next_token)
def delete_cluster(cluster_name, region=None):
"""
Initiate the deletion of a cluster.
:param cluster_name: Name of the cluster
:type cluster_name: str
:param region: AWS Region that the operation corresponds to.
:type region: str
:rtype: DeleteClusterResponseContent
"""
try:
cluster = Cluster(cluster_name)
if not check_cluster_version(cluster):
raise BadRequestException(
f"Cluster '{cluster_name}' belongs to an incompatible ParallelCluster major version."
)
if not cluster.status == CloudFormationStackStatus.DELETE_IN_PROGRESS:
# TODO: remove keep_logs logic from delete
cluster.delete(keep_logs=False)
return DeleteClusterResponseContent(
cluster=ClusterInfoSummary(
cluster_name=cluster_name,
cloudformation_stack_status=CloudFormationStackStatus.DELETE_IN_PROGRESS,
cloudformation_stack_arn=cluster.stack.id,
region=os.environ.get("AWS_DEFAULT_REGION"),
version=cluster.stack.version,
cluster_status=cloud_formation_status_to_cluster_status(CloudFormationStackStatus.DELETE_IN_PROGRESS),
)
)
except StackNotFoundError:
raise NotFoundException(
f"Cluster '{cluster_name}' does not exist or belongs to an incompatible ParallelCluster major version. "
"In case you have running instances belonging to a deleted cluster please use the DeleteClusterInstances "
"API."
)
def create_cluster(
create_cluster_request_content: Dict,
region: str = None,
suppress_validators: List[str] = None,
validation_failure_level: str = None,
dryrun: bool = None,
rollback_on_failure: bool = None,
) -> CreateClusterResponseContent:
"""
Create a managed cluster in a given region.
:param create_cluster_request_content:
:type create_cluster_request_content: dict | bytes
:param region: AWS Region that the operation corresponds to.
:type region: str
:param suppress_validators: Identifies one or more config validators to suppress.
Format: (ALL|type:[A-Za-z0-9]+)
:param validation_failure_level: Min validation level that will cause the cluster creation to fail.
(Defaults to 'ERROR'.)
:param dryrun: Only perform request validation without creating any resource. May be used to validate the cluster
configuration. (Defaults to 'false'.)
:type dryrun: bool
:param rollback_on_failure: When set it automatically initiates a cluster stack rollback on failures.
(Defaults to 'true'.)
:type rollback_on_failure: bool
"""
# Set defaults
rollback_on_failure = rollback_on_failure in {True, None}
validation_failure_level = validation_failure_level or ValidationLevel.ERROR
dryrun = dryrun is True
create_cluster_request_content = CreateClusterRequestContent.from_dict(create_cluster_request_content)
cluster_config = create_cluster_request_content.cluster_configuration
if not cluster_config:
LOGGER.error("Failed: configuration is required and cannot be empty")
raise BadRequestException("configuration is required and cannot be empty")
try:
cluster = Cluster(create_cluster_request_content.cluster_name, cluster_config)
if dryrun:
ignored_validation_failures = cluster.validate_create_request(
get_validator_suppressors(suppress_validators), FailureLevel[validation_failure_level]
)
validation_messages = validation_results_to_config_validation_errors(ignored_validation_failures)
raise DryrunOperationException(validation_messages=validation_messages or None)
stack_id, ignored_validation_failures = cluster.create(
disable_rollback=not rollback_on_failure,
validator_suppressors=get_validator_suppressors(suppress_validators),
validation_failure_level=FailureLevel[validation_failure_level],
)
return CreateClusterResponseContent(
ClusterInfoSummary(
cluster_name=create_cluster_request_content.cluster_name,
cloudformation_stack_status=CloudFormationStackStatus.CREATE_IN_PROGRESS,
cloudformation_stack_arn=stack_id,
region=os.environ.get("AWS_DEFAULT_REGION"),
version=get_installed_version(),
cluster_status=cloud_formation_status_to_cluster_status(CloudFormationStackStatus.CREATE_IN_PROGRESS),
),
validation_messages=validation_results_to_config_validation_errors(ignored_validation_failures) or None,
)
except ConfigValidationError as e:
config_validation_messages = validation_results_to_config_validation_errors(e.validation_failures) or None
raise CreateClusterBadRequestException(
CreateClusterBadRequestExceptionResponseContent(
configuration_validation_errors=config_validation_messages, message=str(e)
)
)
def update_cluster(
update_cluster_request_content: Dict,
cluster_name,
suppress_validators=None,
validation_failure_level=None,
region=None,
dryrun=None,
force_update=None,
):
"""
Update a cluster managed in a given region.
:param update_cluster_request_content:
:param cluster_name: Name of the cluster
:type cluster_name: str
:param suppress_validators: Identifies one or more config validators to suppress.
Format: (ALL|type:[A-Za-z0-9]+)
:type suppress_validators: List[str]
:param validation_failure_level: Min validation level that will cause the update to fail.
(Defaults to 'error'.)
:type validation_failure_level: dict | bytes
:param region: AWS Region that the operation corresponds to.
:type region: str
:param dryrun: Only perform request validation without creating any resource.
May be used to validate the cluster configuration and update requirements. Response code: 200
:type dryrun: bool
:param force_update: Force update by ignoring the update validation errors.
(Defaults to 'false'.)
:type force_update: bool
:rtype: UpdateClusterResponseContent
"""
# Set defaults
validation_failure_level = validation_failure_level or ValidationLevel.ERROR
dryrun = dryrun is True
force_update = force_update is True
update_cluster_request_content = UpdateClusterRequestContent.from_dict(update_cluster_request_content)
cluster_config = update_cluster_request_content.cluster_configuration
if not cluster_config:
LOGGER.error("Failed: configuration is required and cannot be empty")
raise BadRequestException("configuration is required and cannot be empty")
try:
cluster = Cluster(cluster_name)
if not check_cluster_version(cluster, exact_match=True):
raise BadRequestException(
f"the update can be performed only with the same ParallelCluster version ({cluster.stack.version}) "
"used to create the cluster."
)
if dryrun:
_, changes, ignored_validation_failures = cluster.validate_update_request(
target_source_config=cluster_config,
force=force_update,
validator_suppressors=get_validator_suppressors(suppress_validators),
validation_failure_level=FailureLevel[validation_failure_level],
)
change_set, _ = _analyze_changes(changes)
validation_messages = validation_results_to_config_validation_errors(ignored_validation_failures)
raise DryrunOperationException(change_set=change_set, validation_messages=validation_messages or None)
changes, ignored_validation_failures = cluster.update(
target_source_config=cluster_config,
validator_suppressors=get_validator_suppressors(suppress_validators),
validation_failure_level=FailureLevel[validation_failure_level],
force=force_update,
)
change_set, _ = _analyze_changes(changes)
return UpdateClusterResponseContent(
cluster=ClusterInfoSummary(
cluster_name=cluster_name,
cloudformation_stack_status=CloudFormationStackStatus.UPDATE_IN_PROGRESS,
cloudformation_stack_arn=cluster.stack.id,
region=os.environ.get("AWS_DEFAULT_REGION"),
version=cluster.stack.version,
cluster_status=cloud_formation_status_to_cluster_status(CloudFormationStackStatus.UPDATE_IN_PROGRESS),
),
validation_messages=validation_results_to_config_validation_errors(ignored_validation_failures) or None,
change_set=change_set,
)
except ConfigValidationError as e:
config_validation_messages = validation_results_to_config_validation_errors(e.validation_failures) or None
raise UpdateClusterBadRequestException(
UpdateClusterBadRequestExceptionResponseContent(
configuration_validation_errors=config_validation_messages, message=str(e)
)
)
except ClusterUpdateError as e:
raise _handle_cluster_update_error(e)
except (NotFoundClusterActionError, StackNotFoundError):
raise NotFoundException(
f"Cluster '{cluster_name}' does not exist or belongs to an incompatible ParallelCluster major version."
)