def get_os_client_ex(spark, region):
from oci_core import dfapp_get_os_client, get_delegation_token, get_os_client
if USE_INSTANCE_PRINCIPLE:
delegation_token = get_delegation_token(spark)
os_client = dfapp_get_os_client(region, delegation_token)
else:
with tempfile.NamedTemporaryFile(mode='w+t', delete=False) as key_f:
key_f.write(OCI_KEY)
_oci_config = dict(OCI_CONFIG)
_oci_config['key_file'] = key_f.name
os_client = get_os_client(None, config=_oci_config)
return os_client
def run(self,
deployment_location,
options={},
args={},
handlers=None,
on_job_submitted=None,
cli_mode=False):
# options fields
# num_executors : number
# driver_shape : string
# executor_shape : string
# lib_url_duration : number (repre the number of minutes)
# on_job_submitted : callback, on_job_submitted(run_id, vendor_info={'oci_run_id': 'xxxyyy'})
o = urlparse(deployment_location)
if o.scheme != 'oci':
raise SparkETLLaunchFailure("deployment_location must be in OCI")
run_dir = self.config.get('run_dir') or self.config.get('run_base_dir')
run_id = str(uuid.uuid4())
namespace = o.netloc.split('@')[1]
bucket = o.netloc.split('@')[0]
root_path = o.path[1:] # remove the leading "/"
# let's get the deployment.json
os_client = get_os_client(self.region, self.config.get("oci_config"))
deployment = os_download_json(
os_client, namespace, bucket,
os.path.join(root_path, "deployment.json"))
# let's upload the args
client_channel = ClientChannel(self.region,
self.config.get("oci_config"), run_dir,
run_id)
client_channel.write_json("args.json", args)
o = urlparse(run_dir)
namespace = o.netloc.split('@')[1]
bucket = o.netloc.split('@')[0]
root_path = o.path[1:] # remove the leading "/"
os_upload_json(os_client, args, namespace, bucket,
f"{root_path}/{run_id}/args.json")
df_client = get_df_client(self.region, self.config.get("oci_config"))
crd_argv = {
'compartment_id':
deployment['compartment_id'],
'application_id':
deployment['application_id'],
'display_name':
options["display_name"],
'arguments': [
"--deployment-location",
deployment_location,
"--run-id",
run_id,
"--run-dir",
os.path.join(run_dir, run_id),
"--app-region",
self.region,
],
}
for key in ['num_executors', 'driver_shape', 'executor_shape']:
if key in options:
crd_argv[key] = options[key]
create_run_details = oci.data_flow.models.CreateRunDetails(**crd_argv)
r = df_client.create_run(create_run_details=create_run_details)
check_response(
r, lambda: SparkETLLaunchFailure(
"dataflow failed to run the application"))
run = r.data
oci_run_id = run.id
print(f"Job launched, run_id = {run_id}, oci_run_id = {run.id}")
if on_job_submitted is not None:
on_job_submitted(run_id, vendor_info={'oci_run_id': run.id})
cli_entered = False
while True:
time.sleep(10)
r = df_client.get_run(run_id=run.id)
check_response(
r, lambda: SparkETLGetStatusFailure(
"dataflow failed to get run status"))
run = r.data
print(f"Status: {run.lifecycle_state}")
if run.lifecycle_state in ('FAILED', 'SUCCEEDED', 'CANCELED'):
break
handle_server_ask(client_channel, handlers)
if cli_mode and not cli_entered and run.lifecycle_state == 'IN_PROGRESS':
cli_entered = True
cli_handler = CLIHandler(client_channel, None, handlers)
cli_handler.loop()
if run.lifecycle_state in ('FAILED', 'CANCELED'):
raise Exception(f"Job failed with status: {run.lifecycle_state}")
return client_channel.read_json('result.json')
def delete_json(self, name):
os_client = get_os_client(self.region, self.oci_config)
object_name = os.path.join(self.root_path, self.run_id, name)
os_delete_object(os_client, self.namespace, self.bucket, object_name)
def write_json(self, name, payload):
os_client = get_os_client(self.region, self.oci_config)
object_name = os.path.join(self.root_path, self.run_id, name)
os_upload_json(os_client, payload, self.namespace, self.bucket,
object_name)
def read_json(self, name):
os_client = get_os_client(self.region, self.oci_config)
object_name = os.path.join(self.root_path, self.run_id, name)
result = os_download_json(os_client, self.namespace, self.bucket,
object_name)
return result
def deploy(self, build_dir, destination_location):
o = urlparse(destination_location)
if o.scheme != 'oci':
raise SparkETLDeploymentFailure("destination_location must be in OCI")
namespace = o.netloc.split('@')[1]
bucket = o.netloc.split('@')[0]
root_path = o.path[1:] # remove the leading "/"
build = Build(build_dir)
print("Uploading files:")
# Data flow want to call python lib python.zip
os_client = get_os_client(self.region, config=self.config.get("oci_config"))
for artifact in build.artifacts:
os_upload(
os_client,
f"{build_dir}/{artifact}",
namespace,
bucket,
f"{root_path}/{build.version}/{artifact}"
)
# let's upload the job loader
job_loader_filename = get_job_loader(self.config.get("oci_config"))
os_upload(
os_client,
job_loader_filename,
namespace,
bucket,
f"{root_path}/{build.version}/job_loader.py"
)
application = self.create_application(build.manifest, destination_location)
app_info = {
"application_id": application.id,
"compartment_id": application.compartment_id
}
os_upload_json(
os_client, app_info,
namespace, bucket, f"{root_path}/{build.version}/deployment.json"
)
oci_config = self.config.get("oci_config")
if oci_config is not None:
os_upload(
os_client,
_save_json_temp(oci_config),
namespace,
bucket,
"oci_config.json"
)
os_upload(
os_client,
oci_config['key_file'],
namespace,
bucket,
"oci_api_key.pem",
)
