def test_realtime_api(
client: cx.Client, api: str, timeout: int = None, api_config_name: str = "cortex.yaml"
):
api_dir = TEST_APIS_DIR / api
with open(str(api_dir / api_config_name)) as f:
api_specs = yaml.safe_load(f)
expectations = None
expectations_file = api_dir / "expectations.yaml"
if expectations_file.exists():
expectations = parse_expectations(str(expectations_file))
api_name = api_specs[0]["name"]
for api_spec in api_specs:
client.create_api(api_spec=api_spec, project_dir=api_dir)
try:
assert apis_ready(
client=client, api_names=[api_name], timeout=timeout
), f"apis {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
response = request_prediction(client, api_name, payload)
assert (
response.status_code == HTTPStatus.OK
), f"status code: got {response.status_code}, expected {HTTPStatus.OK}"
if expectations and "response" in expectations:
assert_response_expectations(response, expectations["response"])
finally:
delete_apis(client, [api_name])
def test_realtime_api(
printer: Callable,
client: cx.Client,
api: str,
timeout: int = None,
api_config_name: str = "cortex_cpu.yaml",
node_groups: List[str] = [],
extra_path: str = "",
method: str = "POST",
):
api_dir = TEST_APIS_DIR / api
with open(str(api_dir / api_config_name)) as f:
api_specs = yaml.safe_load(f)
assert len(api_specs) == 1
if len(node_groups) > 0:
api_specs[0]["node_groups"] = node_groups
expectations = None
expectations_file = api_dir / "expectations.yaml"
if expectations_file.exists():
expectations = parse_expectations(str(expectations_file))
api_name = api_specs[0]["name"]
for api_spec in api_specs:
client.deploy(api_spec=api_spec)
try:
assert apis_ready(client=client, api_names=[api_name],
timeout=timeout), f"apis {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
if method == "POST":
response = post_request(client, api_name, payload, extra_path)
else:
response = get_request(client, api_name, payload, extra_path)
assert (
response.status_code == HTTPStatus.OK
), f"status code: got {response.status_code}, expected {HTTPStatus.OK}"
if expectations and "response" in expectations:
assert_response_expectations(response, expectations["response"])
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
td.Thread(target=lambda: stream_api_logs(client, api_name),
daemon=True).start()
time.sleep(5)
finally:
raise
finally:
delete_apis(client, [api_name])
def test_realtime_api(
client: cx.Client, api: str, timeout: int = None, api_config_name: str = "cortex.yaml"
):
api_dir = TEST_APIS_DIR / api
with open(str(api_dir / api_config_name)) as f:
api_specs = yaml.safe_load(f)
expectations = None
expectations_file = api_dir / "expectations.yaml"
if expectations_file.exists():
expectations = parse_expectations(str(expectations_file))
api_name = api_specs[0]["name"]
for api_spec in api_specs:
client.create_api(api_spec=api_spec, project_dir=api_dir)
try:
assert apis_ready(
client=client, api_names=[api_name], timeout=timeout
), f"apis {api_name} not ready"
if not expectations or "grpc" not in expectations:
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
response = request_prediction(client, api_name, payload)
assert (
response.status_code == HTTPStatus.OK
), f"status code: got {response.status_code}, expected {HTTPStatus.OK}"
if expectations and "response" in expectations:
assert_response_expectations(response, expectations["response"])
if expectations and "grpc" in expectations:
stub, input_sample, output_values, output_type, is_output_stream = generate_grpc(
client, api_name, api_dir, expectations["grpc"]
)
if is_output_stream:
for response, output_val in zip(stub.Predict(input_sample), output_values):
assert (
type(response) == output_type
), f"didn't receive response of type {str(output_type)}, but received {str(type(response))}"
assert response == output_val, f"received {response} instead of {output_val}"
else:
response = stub.Predict(input_sample)
assert (
type(stub.Predict(input_sample)) == output_type
), f"didn't receive response of type {str(output_type)}, but received {str(type(response))}"
assert (
response == output_values[0]
), f"received {response} instead of {output_values[0]}"
finally:
delete_apis(client, [api_name])
def test_async_api(
client: cx.Client,
api: str,
deploy_timeout: int = None,
poll_retries: int = 5,
poll_sleep_seconds: int = 1,
api_config_name: str = "cortex.yaml",
):
api_dir = TEST_APIS_DIR / api
with open(str(api_dir / api_config_name)) as f:
api_specs = yaml.safe_load(f)
expectations = None
expectations_file = api_dir / "expectations.yaml"
if expectations_file.exists():
expectations = parse_expectations(str(expectations_file))
assert len(api_specs) == 1
api_name = api_specs[0]["name"]
client.create_api(api_spec=api_specs[0], project_dir=api_dir)
try:
assert apis_ready(
client=client, api_names=[api_name], timeout=deploy_timeout
), f"apis {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
response = request_prediction(client, api_name, payload)
assert (
response.status_code == HTTPStatus.OK
), f"workload submission status code: got {response.status_code}, expected {HTTPStatus.OK}"
response_json = response.json()
assert "id" in response_json
request_id = response_json["id"]
result_response = None
for i in range(poll_retries + 1):
result_response = retrieve_async_result(
cliett=client, api_name=api_name, request_id=request_id
)
if result_response.status_code == HTTPStatus.OK:
break
time.sleep(poll_sleep_seconds)
assert (
result_response.status_code == HTTPStatus.OK
), f"result retrieval status code: got {result_response.status_code}, expected {HTTPStatus.OK}"
result_response_json = result_response.json()
# validate keys are in the result json response
assert (
"id" in result_response_json
), f"id key was not present in result response (response: {result_response_json})"
assert (
"status" in result_response_json
), f"status key was not present in result response (response: {result_response_json})"
assert (
"result" in result_response_json
), f"result key was not present in result response (response: {result_response_json})"
assert (
"timestamp" in result_response_json
), f"timestamp key was not present in result response (response: {result_response_json})"
# validate result json response has valid values
assert (
result_response_json["id"] == request_id
), f"result 'id' and request 'id' mismatch ({result_response_json['id']} != {request_id})"
assert (
result_response_json["status"] == "completed"
), f"async workload did not complete (response: {result_response_json})"
assert result_response_json["timestamp"] != "", "result 'timestamp' value was empty"
assert result_response_json["result"] != "", "result 'result' value was empty"
# assert result expectations
if expectations:
assert_json_expectations(result_response_json["result"], expectations["response"])
finally:
delete_apis(client, [api_name])
def test_long_running_realtime(
printer: Callable,
client: cx.Client,
api: str,
long_running_config: Dict[str, Union[int, float]],
deploy_timeout: int = None,
api_config_name: str = "cortex.yaml",
):
api_dir = TEST_APIS_DIR / api
with open(str(api_dir / api_config_name)) as f:
api_specs = yaml.safe_load(f)
time_to_run = long_running_config["time_to_run"]
status_code_timeout = long_running_config["status_code_timeout"]
expectations = None
expectations_file = api_dir / "expectations.yaml"
if expectations_file.exists():
expectations = parse_expectations(str(expectations_file))
api_name = api_specs[0]["name"]
for api_spec in api_specs:
client.create_api(api_spec=api_spec, project_dir=api_dir)
try:
assert apis_ready(client=client,
api_names=[api_name],
timeout=deploy_timeout), f"apis {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
counter = 0
start_time = time.time()
while time.time() - start_time <= time_to_run:
response = request_prediction(client, api_name, payload)
assert (
response.status_code == HTTPStatus.OK
), f"status code: got {response.status_code}, expected {HTTPStatus.OK}"
if expectations and "response" in expectations:
assert_response_expectations(response,
expectations["response"])
counter += 1
printer("verifying number of processed requests using the client")
assert api_requests(
client, api_name, counter, timeout=status_code_timeout
), f"the number of 2xx response codes for api {api_name} doesn't match the expected number {counter}"
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
td.Thread(target=lambda: client.stream_api_logs(api_name),
daemon=True).start()
time.sleep(5)
except:
pass
raise
finally:
delete_apis(client, [api_name])
def test_long_running_realtime(
printer: Callable,
client: cx.Client,
api: str,
long_running_config: Dict[str, Union[int, float]],
deploy_timeout: int = None,
api_config_name: str = "cortex_cpu.yaml",
node_groups: List[str] = [],
):
api_dir = TEST_APIS_DIR / api
with open(str(api_dir / api_config_name)) as f:
api_specs = yaml.safe_load(f)
assert len(api_specs) == 1
time_to_run = long_running_config["time_to_run"]
if len(node_groups) > 0:
api_specs[0]["node_groups"] = node_groups
expectations = None
expectations_file = api_dir / "expectations.yaml"
if expectations_file.exists():
expectations = parse_expectations(str(expectations_file))
api_name = api_specs[0]["name"]
for api_spec in api_specs:
client.deploy(api_spec=api_spec)
try:
assert apis_ready(client=client,
api_names=[api_name],
timeout=deploy_timeout), f"apis {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
counter = 0
start_time = time.time()
while time.time() - start_time <= time_to_run:
response = post_request(client, api_name, payload)
assert (
response.status_code == HTTPStatus.OK
), f"status code: got {response.status_code}, expected {HTTPStatus.OK}"
if expectations and "response" in expectations:
assert_response_expectations(response,
expectations["response"])
counter += 1
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
td.Thread(target=lambda: stream_api_logs(client, api_name),
daemon=True).start()
time.sleep(5)
finally:
raise
finally:
delete_apis(client, [api_name])
def test_async_api(
printer: Callable,
client: cx.Client,
api: str,
deploy_timeout: int = None,
poll_retries: int = 5,
poll_sleep_seconds: int = 1,
api_config_name: str = "cortex_cpu.yaml",
node_groups: List[str] = [],
):
api_dir = TEST_APIS_DIR / api
with open(str(api_dir / api_config_name)) as f:
api_specs = yaml.safe_load(f)
assert len(api_specs) == 1
if len(node_groups) > 0:
api_specs[0]["node_groups"] = node_groups
expectations = None
expectations_file = api_dir / "expectations.yaml"
if expectations_file.exists():
expectations = parse_expectations(str(expectations_file))
api_name = api_specs[0]["name"]
client.deploy(api_spec=api_specs[0])
try:
assert apis_ready(client=client,
api_names=[api_name],
timeout=deploy_timeout), f"apis {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
response = post_request(client, api_name, payload)
assert (
response.status_code == HTTPStatus.OK
), f"workload submission status code: got {response.status_code}, expected {HTTPStatus.OK}"
response_json = response.json()
assert "id" in response_json
request_id = response_json["id"]
result_response = None
for _ in range(poll_retries + 1):
result_response = retrieve_async_result(client=client,
api_name=api_name,
request_id=request_id)
if result_response.status_code != HTTPStatus.OK:
time.sleep(poll_sleep_seconds)
continue
result_response_json = result_response.json()
assert (
"id" in result_response_json
), f"id key was not present in result response (response: {result_response_json})"
assert (
"status" in result_response_json
), f"status key was not present in result response (response: {result_response_json})"
if result_response_json["status"] != "completed":
time.sleep(poll_sleep_seconds)
continue
break
assert (
result_response.status_code == HTTPStatus.OK
), f"result retrieval status code: got {result_response.status_code}, expected {HTTPStatus.OK}"
# validate keys are in the result json response
assert (
"result" in result_response_json
), f"result key was not present in result response (response: {result_response_json})"
assert (
"timestamp" in result_response_json
), f"timestamp key was not present in result response (response: {result_response_json})"
# validate result json response has valid values
assert (
result_response_json["id"] == request_id
), f"result 'id' and request 'id' mismatch ({result_response_json['id']} != {request_id})"
assert (
result_response_json["status"] == "completed"
), f"async workload did not complete (response: {result_response_json})"
assert result_response_json[
"timestamp"] != "", "result 'timestamp' value was empty"
assert result_response_json[
"result"] != "", "result 'result' value was empty"
# assert result expectations
if expectations:
assert_json_expectations(result_response_json["result"],
expectations["response"])
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
printer(json.dumps(result_response_json, indent=2))
td.Thread(
target=lambda: stream_api_logs(client, api_name),
daemon=True,
).start()
time.sleep(5)
except:
pass
raise
finally:
delete_apis(client, [api_name])