def _upload_config(self):
"""Upload source config and save config version."""
self._check_bucket_existence()
try:
# Upload config with default values and sections
if self.config:
result = self.bucket.upload_config(
config=ClusterSchema(cluster_name=self.name).dump(
deepcopy(self.config)),
config_name=PCLUSTER_S3_ARTIFACTS_DICT.get("config_name"),
)
self.config.config_version = result.get("VersionId")
# Upload original config
result = self.bucket.upload_config(
config=self.source_config_text,
config_name=PCLUSTER_S3_ARTIFACTS_DICT.get(
"source_config_name"),
format=S3FileFormat.TEXT,
)
# original config version will be stored in CloudFormation Parameters
self.config.original_config_version = result.get("VersionId")
except Exception as e:
raise _cluster_error_mapper(
e,
f"Unable to upload cluster config to the S3 bucket {self.bucket.name} due to exception: {e}"
)
def _generate_artifact_dir(self):
"""
Generate artifact directory in S3 bucket.
cluster artifact dir is generated before cfn stack creation and only generate once.
artifact_directory: e.g. parallelcluster/{version}/clusters/{cluster_name}-jfr4odbeonwb1w5k
"""
service_directory = generate_random_name_with_prefix(self.name)
self.__s3_artifact_dir = "/".join([
PCLUSTER_S3_ARTIFACTS_DICT.get("root_directory"),
get_installed_version(),
PCLUSTER_S3_ARTIFACTS_DICT.get("root_cluster_directory"),
service_directory,
])
def _persist_stack_resources(self, keys):
"""Set the resources in template identified by keys to have a DeletionPolicy of 'Retain'."""
template = self._get_stack_template()
for key in keys:
template["Resources"][key]["DeletionPolicy"] = "Retain"
try:
self.bucket.upload_cfn_template(
template, PCLUSTER_S3_ARTIFACTS_DICT.get("template_name"))
self._update_stack_template(
self.bucket.get_cfn_template_url(
PCLUSTER_S3_ARTIFACTS_DICT.get("template_name")))
except AWSClientError as e:
raise _cluster_error_mapper(
e,
f"Unable to persist logs on cluster deletion, failed with error: {e}."
)
def _upload_artifacts(self):
"""
Upload cluster specific resources and cluster template.
All dirs contained in resource dir will be uploaded as zip files to
{bucket_name}/parallelcluster/{version}/clusters/{cluster_name}/{resource_dir}/artifacts.zip.
All files contained in root dir will be uploaded to
{bucket_name}/parallelcluster/{version}/clusters/{cluster_name}/{resource_dir}/artifact.
"""
self._check_bucket_existence()
try:
resources = pkg_resources.resource_filename(
__name__, "../resources/custom_resources")
self.bucket.upload_resources(
resource_dir=resources,
custom_artifacts_name=PCLUSTER_S3_ARTIFACTS_DICT.get(
"custom_artifacts_name"))
if self.config.scheduler_resources:
self.bucket.upload_resources(
resource_dir=self.config.scheduler_resources,
custom_artifacts_name=PCLUSTER_S3_ARTIFACTS_DICT.get(
"scheduler_resources_name"),
)
# Upload template
if self.template_body:
self.bucket.upload_cfn_template(
self.template_body,
PCLUSTER_S3_ARTIFACTS_DICT.get("template_name"))
if isinstance(self.config.scheduling,
(SlurmScheduling, SchedulerPluginScheduling)):
# upload instance types data
self.bucket.upload_config(
self.config.get_instance_types_data(),
PCLUSTER_S3_ARTIFACTS_DICT.get("instance_types_data_name"),
format=S3FileFormat.JSON,
)
if isinstance(self.config.scheduling, SchedulerPluginScheduling):
self._render_and_upload_scheduler_plugin_template()
except BadRequestClusterActionError:
raise
except Exception as e:
message = f"Unable to upload cluster resources to the S3 bucket {self.bucket.name} due to exception: {e}"
LOGGER.error(message)
raise _cluster_error_mapper(e, message)
def create(
self,
disable_rollback: bool = False,
validator_suppressors: Set[ValidatorSuppressor] = None,
validation_failure_level: FailureLevel = FailureLevel.ERROR,
) -> Tuple[Optional[str], List]:
"""
Create cluster.
raises ClusterActionError: in case of generic error
raises ConfigValidationError: if configuration is invalid
"""
creation_result = None
artifact_dir_generated = False
try:
suppressed_validation_failures = self.validate_create_request(
validator_suppressors, validation_failure_level)
self._add_version_tag()
self._generate_artifact_dir()
artifact_dir_generated = True
self._upload_config()
# Create template if not provided by the user
if not (self.config.dev_settings
and self.config.dev_settings.cluster_template):
self.template_body = CDKTemplateBuilder(
).build_cluster_template(cluster_config=self.config,
bucket=self.bucket,
stack_name=self.stack_name)
# upload cluster artifacts and generated template
self._upload_artifacts()
LOGGER.info("Creating stack named: %s", self.stack_name)
creation_result = AWSApi.instance().cfn.create_stack_from_url(
stack_name=self.stack_name,
template_url=self.bucket.get_cfn_template_url(
template_name=PCLUSTER_S3_ARTIFACTS_DICT.get(
"template_name")),
disable_rollback=disable_rollback,
tags=self._get_cfn_tags(),
)
return creation_result.get(
"StackId"), suppressed_validation_failures
except ConfigValidationError as e:
raise e
except Exception as e:
if not creation_result and artifact_dir_generated:
# Cleanup S3 artifacts if stack is not created yet
self.bucket.delete_s3_artifacts()
raise _cluster_error_mapper(e, str(e))
def _get_cluster_config(self):
"""Retrieve cluster config content."""
config_version = self.stack.original_config_version
try:
return self.bucket.get_config(
version_id=config_version,
config_name=PCLUSTER_S3_ARTIFACTS_DICT.get(
"source_config_name"))
except Exception as e:
raise _cluster_error_mapper(
e,
f"Unable to load configuration from bucket '{self.bucket.name}/{self.s3_artifacts_dir}'.\n{e}"
)
def presigned_config_url(self) -> str:
"""Return a pre-signed Url to download the config from the S3 bucket."""
return self.bucket.get_config_presigned_url(config_name=PCLUSTER_S3_ARTIFACTS_DICT.get("image_config_name"))
def update(
self,
target_source_config: str,
validator_suppressors: Set[ValidatorSuppressor] = None,
validation_failure_level: FailureLevel = FailureLevel.ERROR,
force: bool = False,
):
"""
Update cluster.
raises ClusterActionError: in case of generic error
raises ConfigValidationError: if configuration is invalid
raises ClusterUpdateError: if update is not allowed
"""
try:
target_config, changes, ignored_validation_failures = self.validate_update_request(
target_source_config, validator_suppressors,
validation_failure_level, force)
self.config = target_config
self.__source_config_text = target_source_config
self._add_version_tag()
self._upload_config()
# Create template if not provided by the user
if not (self.config.dev_settings
and self.config.dev_settings.cluster_template):
self.template_body = CDKTemplateBuilder(
).build_cluster_template(
cluster_config=self.config,
bucket=self.bucket,
stack_name=self.stack_name,
log_group_name=self.stack.log_group_name,
)
# upload cluster artifacts and generated template
self._upload_artifacts()
LOGGER.info("Updating stack named: %s", self.stack_name)
AWSApi.instance().cfn.update_stack_from_url(
stack_name=self.stack_name,
template_url=self.bucket.get_cfn_template_url(
template_name=PCLUSTER_S3_ARTIFACTS_DICT.get(
"template_name")),
tags=self._get_cfn_tags(),
)
self.__stack = ClusterStack(AWSApi.instance().cfn.describe_stack(
self.stack_name))
LOGGER.debug("StackId: %s", self.stack.id)
LOGGER.info("Status: %s", self.stack.status)
return changes, ignored_validation_failures
except ClusterActionError as e:
# It can be a ConfigValidationError or ClusterUpdateError
raise e
except Exception as e:
LOGGER.critical(e)
raise _cluster_error_mapper(e, f"Cluster update failed.\n{e}")
def config_presigned_url(self) -> str:
"""Return a pre-signed Url to download the config from the S3 bucket."""
return self.bucket.get_config_presigned_url(
config_name=PCLUSTER_S3_ARTIFACTS_DICT.get("source_config_name"),
version_id=self.stack.original_config_version,
)