def download_file_content_from_url(url: str,
bearer_token: str = None) -> bytes:
request_headers = dict()
if bearer_token and "?token=" not in url:
request_headers.update({"Authorization": f"Bearer {bearer_token}"})
try:
raw_url = url.replace("/blob/", "/") \
.replace("/github.ibm.com/", "/raw.github.ibm.com/") \
.replace("/github.com/", "/raw.githubusercontent.com/")
response = requests.get(raw_url,
allow_redirects=True,
headers=request_headers)
if response.ok:
file_content = response.content
return file_content
except Exception as e:
raise ApiError(f"Could not download file '{url}'. \n{str(e)}", 422)
raise ApiError(
f"Could not download file '{url}'. Reason: {response.reason}",
response.status_code)
def _download_notebook(url: str, enterprise_github_api_token: str) -> dict:
request_headers = dict()
if "ibm.com" in url and "?token=" not in url:
if not enterprise_github_api_token and not ghe_api_token:
raise ApiError(
f"Must provide API token to access notebooks on Enterprise GitHub: {url}",
422)
else:
request_headers.update({
'Authorization':
f'token {enterprise_github_api_token or ghe_api_token}'
})
try:
raw_url = url.replace("/github.ibm.com/", "/raw.github.ibm.com/")\
.replace("/github.com/", "/raw.githubusercontent.com/")\
.replace("/blob/", "/")
response = requests.get(raw_url,
allow_redirects=True,
headers=request_headers)
if response.ok:
notebook_dict = response.json()
return notebook_dict
except Exception as e:
raise ApiError(f"Could not download notebook file '{url}'. \n{str(e)}",
422)
raise ApiError(
f"Could not download notebook file '{url}'. Reason: {response.reason}",
response.status_code)
def _update_bucket_policy(bucket_name: str, prefix: str):
client = _get_minio_client()
try:
bucket_policy = json.loads(client.get_bucket_policy(bucket_name))
except NoSuchBucketPolicy:
bucket_policy = dict(_bucket_policy_template)
getobject_stmts = [s for s in bucket_policy["Statement"] if s.get("Sid") == _bucket_policy_sid] or \
[s for s in bucket_policy["Statement"] if "s3:GetObject" in s["Action"]]
if not getobject_stmts:
bucket_policy["Statement"].append(_bucket_policy_stmt)
getobject_stmts = bucket_policy["Statement"][-1]
resources = getobject_stmts[-1]["Resource"]
new_resource = f"arn:aws:s3:::{bucket_name}/{prefix}"
if new_resource not in resources and not any(
[r.strip("*") in new_resource for r in resources]):
resources.append(new_resource)
new_policy_str = json.dumps(bucket_policy)
try:
client.set_bucket_policy(bucket_name, new_policy_str)
except ResponseError as e:
if e.code == 'XMinioPolicyNesting':
raise ApiError(
f"{e.message.split('.')[0]}."
f" New policy: '{new_policy_str}'."
f" Existing policy: '{client.get_bucket_policy(bucket_name)}'")
def run_pipeline_in_experiment(api_pipeline: ApiPipeline, parameters: dict = None, run_name: str = None,
wait_for_status: bool = False):
try:
client = KfpClient(_pipeline_service_url)
experiment = client.create_experiment('PIPELINE_RUNS')
run_result = client.run_pipeline(experiment_id=experiment.id,
job_name=run_name or api_pipeline.name,
params=parameters,
pipeline_id=api_pipeline.id)
run_id = run_result.id
if wait_for_status:
run_details = wait_for_run_status(client, run_id, 10)
run_status = json.loads(run_details.pipeline_runtime.workflow_manifest)["status"]
if run_status \
and run_status.get("phase", "").lower() in ["failed", "error"] \
and run_status.get("message"):
raise RuntimeError(f"Run {run_id} failed with error: {run_status['message']}")
return run_id
except Exception as e:
print(f"Exception trying to run pipeline {api_pipeline.id} '{api_pipeline.name}'"
f" with parameters {parameters}:"
f" %s\n" % e)
raise ApiError(message=f"{e.body}\nKFP URL: {_pipeline_service_url}", http_status_code=e.status or 422)
return None
def upload_pipeline_to_kfp(uploadfile: str, name: str = None) -> ApiPipeline:
config = UploadClientConfig()
config.host = _pipeline_service_url
api_client = UploadApiClient(configuration=config)
api_instance = PipelineUploadServiceApi(api_client=api_client)
try:
kfp_pipeline: KfpPipeline = api_instance.upload_pipeline(uploadfile=uploadfile, name=name)
api_pipeline: ApiPipeline = ApiPipeline.from_dict(kfp_pipeline.to_dict())
api_pipeline.status = kfp_pipeline.error
return api_pipeline
except PipelineUploadApiException as e:
kfp_host = api_instance.api_client.configuration.host
print(f"Error calling PipelineServiceApi ({kfp_host}) -> upload_pipeline(name='{name}'): {e}")
error_body = json.loads(e.body) or {"error_message": str(e)}
error_msg = error_body["error_message"]
status_code = 409 if "already exist. Please specify a new name" in error_msg else e.status
raise ApiError(error_msg, status_code)
return None
def _validate_schema(table_name: str, swagger_class):
# swagger_object.swagger_types dictionary maintains insertion order since Python 3.6
# but does not show defaults, use inspection to get argument list from constructor
sig = inspect.signature(swagger_class.__init__)
swagger_columns_w_type = []
for _, p in sig.parameters.items():
if p.name == "self":
continue
col_name = _convert_attr_name_to_col_name(p.name)
col_type = _get_column_type(p.name, p.annotation, swagger_class)
swagger_columns_w_type.append((col_name, col_type))
query = f"SELECT COLUMN_NAME, SUBSTR(COLUMN_TYPE,1,64) as COLUMN_TYPE " \
f" FROM INFORMATION_SCHEMA.COLUMNS " \
f" WHERE TABLE_SCHEMA = '{_database}' AND TABLE_NAME = '{table_name}'"
cnx = _get_connection()
cursor = cnx.cursor(buffered=True)
table_columns_w_type = []
try:
cursor.execute(query)
for column_name, column_type in cursor:
table_columns_w_type.append((column_name, column_type))
except Error as err:
print(err.msg)
raise err
finally:
cursor.close()
cnx.close()
if table_columns_w_type and set(table_columns_w_type) != set(
swagger_columns_w_type):
if isinstance(swagger_class, Model):
swagger_class = type(swagger_class)
cols_found = "\n - ".join(
[f"'{n}' {t}" for n, t in table_columns_w_type])
cols_expect = "\n - ".join(
[f"'{n}' {t}" for n, t in swagger_columns_w_type])
err_msg = f"The MySQL table '{_database}.{table_name}' does not match Swagger" \
f" class '{swagger_class.__name__}'.\n" \
f" Found table with columns:\n" \
f" - {cols_found}.\n" \
f" Expected table with columns:\n" \
f" - {cols_expect}.\n" \
f" Delete and recreate the table by calling the API endpoint 'DELETE /{table_name}/*'"
raise ApiError(err_msg)
return len(table_columns_w_type) > 0
def upload_pipeline_to_kfp(uploadfile: str,
name: str = None,
description: str = None) -> ApiPipeline:
kfp_client = KfpClient()
try:
kfp_pipeline: KfpPipeline = kfp_client.upload_pipeline(
pipeline_package_path=uploadfile,
pipeline_name=name,
description=description)
api_pipeline: ApiPipeline = ApiPipeline.from_dict(
kfp_pipeline.to_dict())
api_pipeline.status = kfp_pipeline.error
return api_pipeline
except PipelineApiException as e:
kfp_host = _pipeline_service_url
print(
f"Error calling PipelineServiceApi ({kfp_host}) -> upload_pipeline(name='{name}'): {e}"
)
error_body = json.loads(e.body) or {"error_message": str(e)}
error_msg = error_body["error_message"]
status_code = 409 if "already exist. Please specify a new name" in error_msg else e.status
raise ApiError(error_msg, status_code)
return None
def _convert_value_to_mysql(value,
python_type: type,
mysql_type_override: str = None,
quote_str=False):
# turn child attributes of type swagger._base_model.Model into dicts
def to_dict(v):
return v.to_dict() if hasattr(v, "to_dict") else v
if type(
python_type
) == typing._GenericAlias: # or str(python_type).startswith("typing."):
python_type = eval(python_type._name.lower())
if value and not issubclass(type(value), python_type) \
and not (isinstance(value, dict) and issubclass(python_type, Model)):
err_msg = f"The type '{type(value)}' does not match expected target type '{python_type}' for value '{value}'"
raise ApiError(err_msg, 422)
if not value:
if python_type == bool:
return False
elif python_type in [int, float]:
return 0
elif python_type == dict and mysql_type_override == "json":
return "{}"
elif python_type in [str, dict, list] or issubclass(
python_type, Model):
return ""
else:
return None
if hasattr(value, "to_dict"):
mysql_value = json.dumps(value.to_dict())
elif python_type == list: # or isinstance(value, list):
mysql_value = json.dumps(list(map(to_dict, value)))
elif python_type == dict or issubclass(python_type, Model) and isinstance(
value, dict):
mysql_value = json.dumps(
dict(map(lambda item: (item[0], to_dict(item[1])), value.items())))
elif python_type == datetime: # or isinstance(value, datetime):
mysql_value = int(value.timestamp())
else:
mysql_value = value
# DON'T quote strings when using MySQL queries with parameters
if mysql_value and quote_str and type(mysql_value) == str:
mysql_value_escaped = mysql_value.replace("'", r"\'")
mysql_value = f"'{mysql_value_escaped}'"
return mysql_value
def store_data(swagger_object: Model) -> str:
table_name = _get_table_name(swagger_object)
_verify_or_create_table(table_name, swagger_object)
swagger_fields = swagger_object.to_dict().keys()
# TODO: remove generate_id() calls in controller_impl methods, do it here
if "id" in swagger_fields and not swagger_object.id:
swagger_object.id = generate_id(swagger_object.name if "name" in
swagger_fields else None)
# TODO: remove creating a new data in controller_impl methods, do it here
if "created_at" in swagger_fields and not swagger_object.created_at:
swagger_object.created_at = datetime.now()
column_names = [_convert_attr_name_to_col_name(f) for f in swagger_fields]
column_values = [
_convert_value_to_mysql(getattr(swagger_object, f),
swagger_object.swagger_types[f])
for f in swagger_fields
]
column_names_str = ", ".join(column_names)
values_list_str = ('%s,' * len(column_values)).rstrip(',')
insert_stmt = (f"INSERT INTO {table_name} "
f"({column_names_str}) "
f"VALUES ({values_list_str})")
cnx = _get_connection()
cnx.autocommit = True
cursor = cnx.cursor()
try:
cursor.execute(insert_stmt, tuple(column_values))
cnx.commit()
except IntegrityError as e:
cnx.rollback()
raise ApiError(e.msg, 409)
except Error as err:
cnx.rollback()
print(err.msg)
print(insert_stmt)
print(column_values)
raise err
finally:
cursor.close()
cnx.close()
return swagger_object.id
def create_secret(secret_name: str, secret_contents: dict):
try:
command = [
'kubectl', '-n', _namespace, 'create', 'secret', 'generic',
secret_name
]
for key, value in secret_contents.items():
if type(value) == dict:
raise ApiError(f"Secret values must not be of type 'dict'")
if type(value) == list:
value = ",".join([str(v) for v in value])
if type(value) == str and " " in value:
value = f"\"{value}\""
command.append(f"--from-literal={key}={value or ''}")
output = subprocess.run(command,
capture_output=True,
check=True,
timeout=10)
pprint(output.stdout.decode())
except Exception as e:
if output and output.stderr:
pprint(output.stderr.decode())
raise ApiError(f"Error trying to create secret '{secret_name}': {e}")
def run_custom_pipeline_in_experiment(custom_pipeline: ApiPipelineCustom,
run_name: str,
parameters: dict,
wait_for_status: bool = False):
try:
source_code = generate_custom_pipeline_run_script(custom_pipeline,
parameters,
run_name,
hide_secrets=False)
except Exception as e:
# TODO: remove this debug logging for development only
print(
f"Error trying to generate code for custom pipeline run '{run_name or custom_pipeline.name}': {e}"
)
print(custom_pipeline)
print(parameters)
raise e
try:
run_id = run_code_in_experiment(source_code, wait_for_status)
except SyntaxError as e:
print(
f"SyntaxError trying to run pipeline DSL '{run_name or custom_pipeline.name}': {e}"
)
print(source_code)
print("Custom pipeline payload:")
print(custom_pipeline)
raise ApiError(
f"SyntaxError trying to run pipeline DSL: {e.msg}\n"
f"{source_code}", 500)
except Exception as e:
# TODO: remove this debug logging for development only
print(
f"Error trying to run custom pipeline code '{run_name or custom_pipeline.name}': {e}"
)
print(custom_pipeline)
print(source_code)
raise e
# TODO: remove this debug logging for development only
print("Custom pipeline payload:")
print(custom_pipeline)
print("Pipeline DSL:")
print(source_code)
return run_id
def get_yaml_file_content_from_uploadfile(uploadfile: FileStorage):
file_name = uploadfile.filename
file_ext = file_name.lower().split(".")[-1]
if file_ext in ["tgz", "gz"]:
yaml_file_content = extract_yaml_from_tarfile(uploadfile)
elif file_ext in ["yaml", "yml"]:
yaml_file_content = uploadfile.stream.read()
else:
raise ApiError(
f"File extension not supported: '{file_ext}', uploadfile: '{file_name}'."
f"Supported file extensions: .tar.gz, .gz, .yaml, .yml", 501)
return yaml_file_content
def delete_secret(secret_name):
if secret_name == "*":
return delete_all_secrets()
output = None
try:
delete_command = [
'kubectl', 'delete', '-n', _namespace, 'secret', secret_name
]
output = subprocess.run(delete_command,
capture_output=True,
check=True,
timeout=10)
print(f"Credential {secret_name} was deleted")
except Exception as e:
if output and output.stderr:
pprint(output.stderr.decode())
raise ApiError(f"Error trying to delete secret '{secret_name}': {e}")
def delete_kfp_pipeline(pipeline_id: str):
config = PipelineClientConfig()
config.host = _pipeline_service_url
api_client = PipelineApiClient(configuration=config)
api_instance = PipelineServiceApi(api_client=api_client)
try:
api_instance.delete_pipeline(pipeline_id)
except AttributeError as e:
# ignore KFP AttributeError. It is a bug in the Swagger-generated client code for Kubeflow Pipelines
if not str(e) == "module 'kfp_pipeline.models' has no attribute 'ERRORUNKNOWN'":
raise e
except PipelineApiException as e:
kfp_host = api_instance.api_client.configuration.host
print(f"Exception when calling PipelineServiceApi ({kfp_host}) -> delete_pipeline: %s\n" % e)
raise ApiError(message=f"{e.body}\nKFP URL: {kfp_host}", http_status_code=e.status or 422)
def get_secret(secret_name, decode=False) -> dict:
output = None
try:
get_command = [
'kubectl', '-n', _namespace, '-o', 'json', 'get', 'secret',
secret_name
]
output = subprocess.run(get_command,
capture_output=True,
check=True,
timeout=10)
secret_data = json.loads(output.stdout.decode()) or {}
secret = secret_data.get("data")
if decode:
for k, v in secret.items():
secret[k] = b64decode(v).decode()
return secret
except Exception as e:
if output and output.stderr:
pprint(output.stderr.decode())
raise ApiError(f"Error trying to retrieve secret '{secret_name}': {e}")
def list_secrets(name_prefix=secret_name_prefix, decode=False) -> [dict]:
output = None
try:
list_command = [
'kubectl', '-n', _namespace, '-o', 'json', 'get', 'secrets'
]
output = subprocess.run(list_command,
capture_output=True,
check=True,
timeout=10)
secrets_data = json.loads(output.stdout.decode()) or {}
mlx_secrets = [
d for d in secrets_data.get("items") or []
if d["metadata"]["name"].startswith(name_prefix)
]
if decode:
for s in mlx_secrets:
for k, v in s["data"].items():
s["data"][k] = b64decode(v).decode()
return mlx_secrets
except Exception as e:
if output and output.stderr:
pprint(output.stderr.decode())
raise ApiError(f"Error trying to list secrets '{name_prefix}...': {e}")
def generate_dataset_run_script(dataset: ApiDataset, dataset_template_url, run_parameters=dict(),
run_name: str = None, fail_on_missing_prereqs=False):
name = f"{dataset.name} ({generate_id(length=4)})"
description = dataset.description.strip().replace("'", "\\'")
# TODO: some of the parameters, template URLs should move out of here
# dataset_parameters = dataset.parameters
# TODO: ApiParameters should not be defined here
dataset_parameters = [ApiParameter(name="action", default="create"),
ApiParameter(name="namespace", default=_namespace)]
pipeline_method_args = generate_pipeline_method_args(dataset_parameters)
parameter_names = ",".join([p.name for p in dataset_parameters])
# TODO: the action parameter is required by DLF-to-PVC op, so it should not be dynamically generated here
parameter_dict = {
"action": "create",
"namespace": run_parameters.get("namespace", _namespace)
}
# see component name at https://github.com/machine-learning-exchange/mlx/blob/main/components/component-samples/dax-to-dlf/component.yaml#L1
dax_to_dlf_component_id = generate_id(name="Generate Dataset Metadata")
# see component name at https://github.com/machine-learning-exchange/mlx/blob/main/components/component-samples/dlf/component.yaml#L1
dlf_to_pvc_component_id = generate_id(name="Create Dataset Volume")
dax_to_dlf_component_url = get_object_url(bucket_name="mlpipeline",
prefix=f"components/{dax_to_dlf_component_id}/",
file_extensions=[".yaml"])
dlf_to_pvc_component_url = get_object_url(bucket_name="mlpipeline",
prefix=f"components/{dlf_to_pvc_component_id}/",
file_extensions=[".yaml"])
if fail_on_missing_prereqs:
if not dax_to_dlf_component_url:
raise ApiError(f"Missing required component '{dax_to_dlf_component_id}'")
if not dlf_to_pvc_component_url:
raise ApiError(f"Missing required component '{dlf_to_pvc_component_id}'")
namespace = run_parameters.get("namespace", _namespace)
pipeline_server = "" if "POD_NAMESPACE" in os.environ else f"'{_pipeline_service_url}'"
run_name = (run_name or "").replace("'", "\"") or dataset.name
substitutions = dict(locals())
template_file = f"run_dataset.TEMPLATE.py"
with open(join(CODE_TEMPLATE_DIR, template_file), 'r') as f:
template_raw = f.read()
template_rendered = Template(template_raw).substitute(substitutions)
run_script = autopep8.fix_code(template_rendered, options={"aggressive": 2})
return run_script
def generate_custom_pipeline_function_body(custom_pipeline: ApiPipelineCustom, hide_secrets=True):
function_body = """
from kfp import components
"""
component_template_raw = """
${comp_name} = components.load_component_from_url('${template_url}')
${op_name} = ${comp_name}(${component_args})
"""
op_dependency_template_raw = """
${op_name}.after(${required_op_name})
"""
for task in custom_pipeline.dag.tasks:
parameters = []
if task.artifact_type == "notebook":
component_s3_prefix = f"components/jupyter/"
notebook_url = get_object_url(bucket_name="mlpipeline",
prefix=f"notebooks/{task.artifact_id}/",
file_extensions=[".ipynb"])
if not notebook_url:
raise ApiError(f"Could not find notebook '{task.artifact_id}'")
task_parameters = list(task.arguments.parameters) if task.arguments and task.arguments.parameters else []
for p in task_parameters:
if type(p.value) == str and p.value.startswith("{{inputs.parameters."):
raise ApiError("Referencing '{{inputs.parameters.*}}' is not supported for notebook parameter"
f" values: {task.to_dict()}", 422)
notebook_parameters = {p.name: p.value or p.default for p in task_parameters}
notebook_parameters_str = json.dumps(notebook_parameters) if notebook_parameters else ""
jupyter_component_parameters = {
"notebook_url": notebook_url,
"notebook_params": notebook_parameters_str,
"api_token": "",
"endpoint_url": "",
"bucket_name": "",
"object_name": "",
"access_key": "",
"secret_access_key": ""
}
if not hide_secrets:
output_folder = f"notebooks/{task.artifact_id}/runs/{datetime.now().strftime('%Y%m%d-%H%M%S')}"
notebook_file_name = notebook_url.split("/")[-1]
output_file_name = notebook_file_name.replace(r'.ipynb', '_out.ipynb')
output_file_path = f"{output_folder}/{output_file_name}"
output_bucket = "mlpipeline"
jupyter_component_parameters.update({
"endpoint_url": "minio-service:9000", # f"{minio_host}:{minio_port}",
"bucket_name": output_bucket,
"object_name": output_file_path,
"access_key": minio_access_key,
"secret_access_key": minio_secret_key
})
for name, value in jupyter_component_parameters.items():
parameters.append(f"{name} = '{value}'")
elif task.artifact_type == "component":
component_s3_prefix = f"components/{task.artifact_id}/"
# replace parameter values that reference pipeline input parameters {{inputs.parameters.parameter_name}}
task_parameters = list(task.arguments.parameters) if task.arguments and task.arguments.parameters else []
missing_parameter_values = [p.name for p in task_parameters
if not p.value and not p.default and p.description \
and p.description.title().startswith("Required")]
if missing_parameter_values:
raise ApiError(f"Missing required task parameters {missing_parameter_values}", 422)
for p in task_parameters:
if type(p.value) == str and p.value.startswith("{{inputs.parameters."):
match = re.match(r"{{inputs.parameters.(?P<pipeline_parameter_name>\w+)}}", p.value)
if not match:
raise ApiError(f"Cannot match pipeline input.parameter '{p.value}'", 422)
pipeline_param_ref = match.groupdict().get("pipeline_parameter_name")
parameters.append(f"{p.name} = {pipeline_param_ref}")
else:
arg = generate_method_arg_from_parameter(p)
parameters.append(arg)
else:
raise ApiError(f"Unknown or unsupported artifact_type '{task.artifact_type}':\n'{task}'", 422)
comp_name = "comp_" + re.sub(r"\W+", "_", task.name, flags=re.ASCII).lower()
op_name = "op_" + re.sub(r"\W+", "_", task.name, flags=re.ASCII).lower()
template_url = get_object_url(bucket_name="mlpipeline",
prefix=component_s3_prefix,
file_extensions=[".yaml", ".yml"])
if not template_url:
raise ApiError(f"Could not find component template '{component_s3_prefix}'")
substitutions = {
"comp_name": comp_name,
"op_name": op_name,
"template_url": template_url,
"component_args": ", ".join(parameters)
}
template_rendered = Template(component_template_raw).substitute(substitutions)
function_body += template_rendered
for task in custom_pipeline.dag.tasks:
for required_task_name in task.dependencies or []:
substitutions = {
"op_name": "op_" + re.sub(r"\W+", "_", task.name, flags=re.ASCII).lower(),
"required_op_name": "op_" + re.sub(r"\W+", "_", required_task_name, flags=re.ASCII).lower()
}
template_rendered = Template(op_dependency_template_raw).substitute(substitutions)
function_body += template_rendered
return function_body
