def test_task_api(
printer: Callable,
client: cx.Client,
api: str,
deploy_timeout: int = None,
job_timeout: int = None,
retry_attempts: int = 0,
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)
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 endpoint_ready(
client=client, api_name=api_name,
timeout=deploy_timeout), f"api {api_name} not ready"
response = None
for _ in range(retry_attempts + 1):
response = request_task(
client,
api_name,
)
if response.status_code == HTTPStatus.OK:
break
time.sleep(1)
job_spec = response.json()
assert job_done(
client=client,
api_name=api_name,
job_id=job_spec["job_id"],
timeout=job_timeout,
), f"task job did not succeed (api_name: {api_name}, job_id: {job_spec['job_id']})"
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
job_status = client.get_job(api_name, job_spec["job_id"])
printer(json.dumps(job_status, indent=2))
td.Thread(target=lambda: client.stream_job_logs(
api_name, job_spec["job_id"]),
daemon=True).start()
time.sleep(5)
except:
pass
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.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(client=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"])
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
job_status = client.get_job(api_name, result_response_json["id"])
printer(json.dumps(job_status, indent=2))
td.Thread(
target=lambda: client.stream_job_logs(
api_name, result_response_json["id"]),
daemon=True,
).start()
time.sleep(5)
except:
pass
raise
finally:
delete_apis(client, [api_name])
def test_task_api(
printer: Callable,
client: cx.Client,
api: str,
deploy_timeout: int = None,
job_timeout: int = None,
retry_attempts: int = 0,
api_config_name: str = "cortex_cpu.yaml",
node_groups: List[str] = [],
local_operator: bool = False,
):
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
api_name = api_specs[0]["name"]
client.deploy(api_spec=api_specs[0])
try:
endpoint_override = f"http://localhost:8888/tasks/{api_name}" if local_operator else None
assert endpoint_ready(
client=client,
api_name=api_name,
timeout=deploy_timeout,
endpoint_override=endpoint_override,
), f"api {api_name} not ready"
response = None
for _ in range(retry_attempts + 1):
response = request_task(client,
api_name,
local_operator=local_operator)
if response.status_code == HTTPStatus.OK:
break
time.sleep(1)
job_spec = response.json()
job_id = job_spec["job_id"]
endpoint_override = (
f"http://localhost:8888/tasks/{api_name}?jobID={job_id}"
if local_operator else None)
assert job_done(
client=client,
api_name=api_name,
job_id=job_spec["job_id"],
timeout=job_timeout,
endpoint_override=endpoint_override,
), f"task job did not succeed (api_name: {api_name}, job_id: {job_spec['job_id']})"
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
job_status = client.get_job(api_name, job_spec["job_id"])
printer(json.dumps(job_status, indent=2))
td.Thread(target=lambda: stream_job_logs(client, api_name,
job_spec["job_id"]),
daemon=True).start()
time.sleep(5)
except:
pass
raise
finally:
delete_apis(client, [api_name])
def test_batch_api(
printer: Callable,
client: cx.Client,
api: str,
test_s3_path: str,
deploy_timeout: int = None,
job_timeout: int = None,
retry_attempts: int = 0,
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)
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 endpoint_ready(
client=client, api_name=api_name,
timeout=deploy_timeout), f"api {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
response = None
for _ in range(retry_attempts + 1):
response = request_batch_prediction(
client,
api_name,
item_list=payload,
batch_size=2,
config={"dest_s3_dir": test_s3_path},
)
if response.status_code == HTTPStatus.OK:
break
time.sleep(1)
assert (
response.status_code == HTTPStatus.OK
), f"status code: got {response.status_code}, expected {HTTPStatus.OK} ({response.text})"
job_spec = response.json()
# monitor job progress
assert job_done(
client=client,
api_name=job_spec["api_name"],
job_id=job_spec["job_id"],
timeout=job_timeout,
), f"job did not succeed (api_name: {api_name}, job_id: {job_spec['job_id']})"
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
job_status = client.get_job(api_name, job_spec["job_id"])
printer(json.dumps(job_status, indent=2))
td.Thread(target=lambda: client.stream_job_logs(
api_name, job_spec["job_id"]),
daemon=True).start()
time.sleep(5)
except:
pass
raise
finally:
delete_apis(client, [api_name])
def test_batch_api(
printer: Callable,
client: cx.Client,
api: str,
test_s3_path: str,
deploy_timeout: int = None,
job_timeout: int = None,
retry_attempts: int = 0,
api_config_name: str = "cortex_cpu.yaml",
node_groups: List[str] = [],
local_operator: bool = False,
):
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
api_name = api_specs[0]["name"]
client.deploy(api_spec=api_specs[0])
try:
endpoint_override = f"http://localhost:8888/batch/{api_name}" if local_operator else None
assert endpoint_ready(
client=client,
api_name=api_name,
timeout=deploy_timeout,
endpoint_override=endpoint_override,
), f"api {api_name} not ready"
with open(str(api_dir / "sample.json")) as f:
payload = json.load(f)
response = None
for _ in range(retry_attempts + 1):
response = request_batch_prediction(
client,
api_name,
item_list=payload,
batch_size=2,
config={"dest_s3_dir": test_s3_path},
local_operator=local_operator,
)
if response.status_code == HTTPStatus.OK:
break
time.sleep(1)
assert (
response.status_code == HTTPStatus.OK
), f"status code: got {response.status_code}, expected {HTTPStatus.OK} ({response.text})"
job_spec = response.json()
# monitor job progress
job_id = job_spec["job_id"]
endpoint_override = (
f"http://localhost:8888/batch/{api_name}?jobID={job_id}"
if local_operator else None)
assert job_done(
client=client,
api_name=api_name,
job_id=job_id,
timeout=job_timeout,
endpoint_override=endpoint_override,
), f"job did not succeed (api_name: {api_name}, job_id: {job_spec['job_id']})"
except:
# best effort
try:
api_info = client.get_api(api_name)
printer(json.dumps(api_info, indent=2))
job_status = client.get_job(api_name, job_spec["job_id"])
printer(json.dumps(job_status, indent=2))
td.Thread(
target=lambda: stream_job_logs(client, api_name, job_spec[
"job_id"]),
daemon=True,
).start()
time.sleep(5)
finally:
raise
finally:
delete_apis(client, [api_name])