Tweak python exception handling (#405)

Author: vishalbollu

pkg/workloads/cortex/lib/api_utils.py

@@ -138,9 +138,9 @@ def post_request_metrics(ctx, api, response, prediction_payload, class_set):
 
                 metrics_list += prediction_metrics(api_dimensions, api, prediction)
             except Exception as e:
-                logger.warn(str(e), exc_info=True)
+                logger.warn("unable to record prediction metric", exc_info=True)
 
     try:
         ctx.publish_metrics(metrics_list)
     except Exception as e:
-        logger.warn(str(e), exc_info=True)
+        logger.warn("failure encountered while publishing metrics", exc_info=True)

pkg/workloads/cortex/lib/context.py

@@ -114,7 +114,7 @@ def load_module(self, module_prefix, module_name, impl_key):
         try:
             impl = imp.load_source(full_module_name, impl_path)
         except Exception as e:
-            raise UserException("unable to load python file") from e
+            raise UserException("unable to load python file", str(e)) from e
 
         return impl, impl_path
 
@@ -200,7 +200,7 @@ def publish_metrics(self, metrics):
 
         if int(response["ResponseMetadata"]["HTTPStatusCode"] / 100) != 2:
             logger.warn(response)
-            raise Exception("failed to publish metrics")
+            raise Exception("cloudwatch returned a non-200 status")
 
 
 REQUEST_HANDLER_IMPL_VALIDATION = {

pkg/workloads/cortex/lib/exceptions.py

@@ -17,19 +17,20 @@
 
 class CortexException(Exception):
     def __init__(self, *messages):
-        super().__init__(self.stringify(messages))
-
+        super().__init__(": ".join(messages))
         self.errors = deque(messages)
 
     def wrap(self, *messages):
         self.errors.extendleft(reversed(messages))
 
     def __str__(self):
-        return self.stringify(self.errors)
+        return self.stringify()
+
+    def __repr__(self):
+        return self.stringify()
 
-    @staticmethod
-    def stringify(str_list):
-        return ": ".join(str_list)
+    def stringify(self):
+        return "error: " + ": ".join(self.errors)
 
 
 class UserException(CortexException):

pkg/workloads/cortex/onnx_serve/api.py

@@ -124,7 +124,7 @@ def transform_to_numpy(input_pyobj, input_metadata):
         np_arr = np_arr.reshape(target_shape)
         return np_arr
     except Exception as e:
-        raise UserException(str(e)) from e
+        raise UserException("failed to convert to numpy array", str(e)) from e
 
 
 def convert_to_onnx_input(sample, input_metadata_list):
@@ -170,7 +170,7 @@ def predict(app_name, api_name):
     try:
         sample = request.get_json()
     except Exception as e:
-        return "Malformed JSON", status.HTTP_400_BAD_REQUEST
+        return "malformed json", status.HTTP_400_BAD_REQUEST
 
     sess = local_cache["sess"]
     api = local_cache["api"]
@@ -189,7 +189,7 @@ def predict(app_name, api_name):
                 debug_obj("pre_inference", prepared_sample, debug)
             except Exception as e:
                 raise UserRuntimeException(
-                    api["request_handler"], "pre_inference request handler"
+                    api["request_handler"], "pre_inference request handler", str(e)
                 ) from e
 
         inference_input = convert_to_onnx_input(prepared_sample, input_metadata)
@@ -208,16 +208,12 @@ def predict(app_name, api_name):
                 result = request_handler.post_inference(result, output_metadata)
             except Exception as e:
                 raise UserRuntimeException(
-                    api["request_handler"], "post_inference request handler"
+                    api["request_handler"], "post_inference request handler", str(e)
                 ) from e
 
             debug_obj("post_inference", result, debug)
-    except CortexException as e:
-        e.wrap("error")
-        logger.exception(str(e))
-        return prediction_failed(str(e))
     except Exception as e:
-        logger.exception(str(e))
+        logger.exception("prediction failed")
         return prediction_failed(str(e))
 
     g.prediction = result
@@ -270,9 +266,9 @@ def start(args):
                 truncate(extract_signature(local_cache["output_metadata"]))
             )
         )
-    except CortexException as e:
-        e.wrap("error")
-        logger.error(str(e))
+
+    except Exception as e:
+        logger.exception("failed to start api")
         sys.exit(1)
 
     if api.get("tracker") is not None and api["tracker"].get("model_type") == "classification":

pkg/workloads/cortex/tf_api/api.py

@@ -31,7 +31,7 @@
 from cortex.lib import util, package, Context, api_utils
 from cortex.lib.storage import S3
 from cortex.lib.log import get_logger, debug_obj
-from cortex.lib.exceptions import CortexException, UserRuntimeException, UserException
+from cortex.lib.exceptions import UserRuntimeException, UserException
 from cortex.lib.stringify import truncate
 
 
@@ -134,7 +134,7 @@ def create_prediction_request(sample):
             prediction_request.inputs[column_name].CopyFrom(tensor_proto)
         except Exception as e:
             raise UserException(
-                'key "{}"'.format(column_name), "expected shape {}".format(shape)
+                'key "{}"'.format(column_name), "expected shape {}".format(shape), str(e)
             ) from e
 
     return prediction_request
@@ -185,7 +185,7 @@ def run_predict(sample, debug=False):
             debug_obj("pre_inference", prepared_sample, debug)
         except Exception as e:
             raise UserRuntimeException(
-                api["request_handler"], "pre_inference request handler"
+                api["request_handler"], "pre_inference request handler", str(e)
             ) from e
 
     validate_sample(prepared_sample)
@@ -201,7 +201,7 @@ def run_predict(sample, debug=False):
             debug_obj("post_inference", result, debug)
         except Exception as e:
             raise UserRuntimeException(
-                api["request_handler"], "post_inference request handler"
+                api["request_handler"], "post_inference request handler", str(e)
             ) from e
 
     return result
@@ -243,12 +243,8 @@ def predict(deployment_name, api_name):
 
     try:
         result = run_predict(sample, debug)
-    except CortexException as e:
-        e.wrap("error")
-        logger.exception(str(e))
-        return prediction_failed(str(e))
     except Exception as e:
-        logger.exception(str(e))
+        logger.exception("prediction failed")
         return prediction_failed(str(e))
 
     g.prediction = result
@@ -284,12 +280,9 @@ def get_signature(app_name, api_name):
             local_cache["metadata"]["signatureDef"],
             local_cache["api"]["tf_serving"]["signature_key"],
         )
-    except CortexException as e:
-        logger.exception(str(e))
-        return str(e), HTTP_404_NOT_FOUND
     except Exception as e:
-        logger.exception(str(e))
-        return str(e), HTTP_404_NOT_FOUND
+        logger.exception("failed to get signature")
+        return jsonify(error=str(e)), 404
 
     response = {"signature": metadata}
     return jsonify(response)
@@ -336,12 +329,8 @@ def start(args):
         if api.get("request_handler") is not None:
             package.install_packages(ctx.python_packages, ctx.storage)
             local_cache["request_handler"] = ctx.get_request_handler_impl(api["name"])
-    except CortexException as e:
-        e.wrap("error")
-        logger.exception(str(e))
-        sys.exit(1)
     except Exception as e:
-        logger.exception(str(e))
+        logger.exception("failed to start api")
         sys.exit(1)
 
     try:
@@ -354,7 +343,7 @@ def start(args):
         try:
             local_cache["class_set"] = api_utils.get_classes(ctx, api["name"])
         except Exception as e:
-            logger.warn("An error occurred while attempting to load classes", exc_info=True)
+            logger.warn("an error occurred while attempting to load classes", exc_info=True)
 
     channel = grpc.insecure_channel("localhost:" + str(args.tf_serve_port))
     local_cache["stub"] = prediction_service_pb2_grpc.PredictionServiceStub(channel)