def validate_model_metadata(data: Dict) -> Dict:
"""Validate metadata.
Parameters
----------
data
User defined model metadata (json)
Returns
-------
Validated model metadata (json)
Raises
------
SeldonInvalidMetadataError if data cannot be properly validated
Notes
-----
Read data from json and create ModelMetadataValidator and MetadataTensorValidator objects.
This function reads data in safe manner from json: validation and exceptions
will happen in ModelMetadataValidator and MetadataTensorValidator classes.
SeldonInvalidMetadataError has been chosen for exception as validation mostly depend on having
a correct type for specific components.
"""
if MODEL_IMAGE is not None:
image_name, image_version = split_image_tag(MODEL_IMAGE)
else:
image_name, image_version = "", ""
name = data.get("name", image_name)
versions = data.get("versions", [image_version])
platform = data.get("platform", "")
try:
inputs = [
MetadataTensorValidator(x.get("name"), x.get("datatype"),
x.get("shape"))
for x in data.get("inputs", [])
]
outputs = [
MetadataTensorValidator(x.get("name"), x.get("datatype"),
x.get("shape"))
for x in data.get("outputs", [])
]
except (AttributeError, TypeError):
raise SeldonInvalidMetadataError(
"Model metadata inputs and outputs must be sequence of dictionaries."
)
meta = ModelMetadataValidator(
name=name,
versions=versions,
platform=platform,
inputs=inputs,
outputs=outputs,
)
return meta.to_dict()
def test_split_image_tag():
image = "seldonio/sklearn-iris:0.1"
expected_name = "seldonio/sklearn-iris"
expected_version = "0.1"
name, version = split_image_tag(image)
assert name == expected_name
assert version == expected_version
def test_split_image_tag_with_colon():
image = "localhost:32000/sklearn-iris:0.1"
expected_name = "localhost:32000/sklearn-iris"
expected_version = "0.1"
name, version = split_image_tag(image)
assert name == expected_name
assert version == expected_version
def validate_model_metadata(data: Dict) -> Dict:
"""Validate metadata.
Parameters
----------
data
User defined model metadata (json)
Returns
-------
Validated model metadata (json)
Raises
------
SeldonInvalidMetadataError if data cannot be properly validated
Notes
-----
Read data from json and validate against v1 or v2 metadata schema.
SeldonInvalidMetadataError exception will be raised if validation fails.
"""
if MODEL_IMAGE is not None:
image_name, image_version = split_image_tag(MODEL_IMAGE)
else:
image_name, image_version = "", ""
default_meta = {
"apiVersion": "v2",
"name": image_name,
"versions": [image_version],
"platform": "",
"inputs": [],
"outputs": [],
}
data = {**default_meta, **data}
v = data.get("apiVersion", "v2")
if v == "v1":
schema = V1_SCHEMA
elif v == "v2":
schema = V2_SCHEMA
else:
raise SeldonInvalidMetadataError(f"Unknown metadata schema: {v}")
try:
validate(data, schema)
except ValidationError as e:
raise SeldonInvalidMetadataError(e)
logger.debug(
f"Successfully validated metadata:\n{json.dumps(data, indent=2)}")
return data