Tracebacks from endpoint failures are propagated to client

launch/cli/endpoints.py

@@ -37,11 +37,15 @@ def list_endpoints(ctx: click.Context):
  servable_endpoint.model_endpoint.bundle_name,
  servable_endpoint.model_endpoint.status,
  servable_endpoint.model_endpoint.endpoint_type,
- (servable_endpoint.model_endpoint.worker_settings or {}).get(
- "min_workers"
+ str(
+ (servable_endpoint.model_endpoint.worker_settings or {}).get(
+ "min_workers"
+ )
  ),
- (servable_endpoint.model_endpoint.worker_settings or {}).get(
- "max_workers"
+ str(
+ (servable_endpoint.model_endpoint.worker_settings or {}).get(
+ "max_workers"
+ )
  ),
  servable_endpoint.model_endpoint.metadata or "{}",
  )

launch/connection.py

@@ -26,21 +26,38 @@ def __eq__(self, other):
  )
 
  def delete(self, route: str):
- return self.make_request({}, route, requests_command=requests.delete)
+ return self.make_request(
+ {},
+ route,
+ requests_command=requests.delete,
+ )
 
  def get(self, route: str):
- return self.make_request({}, route, requests_command=requests.get)
+ return self.make_request(
+ {},
+ route,
+ requests_command=requests.get,
+ )
 
  def post(self, payload: dict, route: str):
  return self.make_request(
- payload, route, requests_command=requests.post
+ payload,
+ route,
+ requests_command=requests.post,
  )
 
  def put(self, payload: dict, route: str):
- return self.make_request(payload, route, requests_command=requests.put)
+ return self.make_request(
+ payload,
+ route,
+ requests_command=requests.put,
+ )
 
  def make_request(
- self, payload: dict, route: str, requests_command=requests.post
+ self,
+ payload: dict,
+ route: str,
+ requests_command=requests.post,
  ) -> dict:
  """
  Makes a request to Launch endpoint and logs a warning if not

launch/model_endpoint.py

@@ -122,6 +122,7 @@ def __init__(
  status: str,
  result_url: Optional[str] = None,
  result: Optional[str] = None,
+ traceback: Optional[str] = None,
  ):
  """
  Parameters:
@@ -140,14 +141,17 @@ def __init__(
  Exactly one of ``result_url`` or ``result`` will be populated,
  depending on the value of ``return_pickled`` in the request.
 
+ traceback: The stack trace if the inference endpoint raised an error. Can be used for debugging
+
  """
  self.client = client
  self.status = status
  self.result_url = result_url
  self.result = result
+ self.traceback = traceback
 
  def __str__(self) -> str:
- return f"status: {self.status}, result: {self.result}, result_url: {self.result_url}"
+ return f"status: {self.status}, result: {self.result}, result_url: {self.result_url}, traceback: {self.traceback}"
 
 
 class EndpointResponseFuture:
@@ -188,13 +192,15 @@ def get(self) -> EndpointResponse:
  status=async_response["state"],
  result_url=async_response.get("result_url", None),
  result=async_response.get("result", None),
+ traceback=None,
  )
  elif async_response["state"] == "FAILURE":
  return EndpointResponse(
  client=self.client,
  status=async_response["state"],
  result_url=None,
  result=None,
+ traceback=async_response.get("traceback", None),
  )
  else:
  raise ValueError(
@@ -236,9 +242,10 @@ def predict(self, request: EndpointRequest) -> EndpointResponse:
  )
  return EndpointResponse(
  client=self.client,
- status=TASK_SUCCESS_STATE,
+ status=raw_response.get("state"),
  result_url=raw_response.get("result_url", None),
  result=raw_response.get("result", None),
+ traceback=raw_response.get("traceback", None),
  )
 
  def status(self):
@@ -429,6 +436,7 @@ def single_request(inner_url, inner_task_id):
  status=raw_response["state"],
  result_url=raw_response.get("result_url", None),
  result=raw_response.get("result", None),
+ traceback=raw_response.get("traceback", None),
  )
  self.responses[url] = response_object
 

pyproject.toml

@@ -26,7 +26,7 @@ exclude = [
 
 [tool.poetry]
 name = "scale-launch"
-version = "0.2.0"
+version = "0.2.1"
 description = "The official Python client library for Launch, the Data Platform for AI"
 authors = ["Your Name <you@example.com>"]
 readme = "README.md"