def start(args): api = None try: ctx = Context(s3_path=args.context, cache_dir=args.cache_dir, workload_id=args.workload_id) api = ctx.apis_id_map[args.api] local_cache["api"] = api local_cache["ctx"] = ctx if api.get("onnx") is None: raise CortexException(api["name"], "onnx key not configured") _, prefix = ctx.storage.deconstruct_s3_path(api["onnx"]["model"]) model_path = os.path.join(args.model_dir, os.path.basename(prefix)) if api["onnx"].get("request_handler") is not None: local_cache["request_handler"] = ctx.get_request_handler_impl( api["name"], args.project_dir) request_handler = local_cache.get("request_handler") if request_handler is not None and util.has_function( request_handler, "pre_inference"): cx_logger().info( "using pre_inference request handler provided in {}".format( api["onnx"]["request_handler"])) else: cx_logger().info("pre_inference request handler not found") if request_handler is not None and util.has_function( request_handler, "post_inference"): cx_logger().info( "using post_inference request handler provided in {}".format( api["onnx"]["request_handler"])) else: cx_logger().info("post_inference request handler not found") sess = rt.InferenceSession(model_path) local_cache["sess"] = sess local_cache["input_metadata"] = sess.get_inputs() cx_logger().info("input_metadata: {}".format( truncate(extract_signature(local_cache["input_metadata"])))) local_cache["output_metadata"] = sess.get_outputs() cx_logger().info("output_metadata: {}".format( truncate(extract_signature(local_cache["output_metadata"])))) except Exception as e: cx_logger().exception("failed to start api") sys.exit(1) if api.get("tracker") is not None and api["tracker"].get( "model_type") == "classification": try: local_cache["class_set"] = api_utils.get_classes(ctx, api["name"]) except Exception as e: cx_logger().warn( "an error occurred while attempting to load classes", exc_info=True) cx_logger().info("API is ready") serve(app, listen="*:{}".format(args.port))
def start(args): api = None try: ctx = Context(s3_path=args.context, cache_dir=args.cache_dir, workload_id=args.workload_id) api = ctx.apis_id_map[args.api] local_cache["api"] = api local_cache["ctx"] = ctx if api.get("request_handler") is not None: local_cache["request_handler"] = ctx.get_request_handler_impl( api["name"], args.project_dir) except Exception as e: logger.exception("failed to start api") sys.exit(1) try: validate_model_dir(args.model_dir) except Exception as e: logger.exception("failed to validate model") sys.exit(1) if api.get("tracker") is not None and api["tracker"].get( "model_type") == "classification": 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) channel = grpc.insecure_channel("localhost:" + str(args.tf_serve_port)) local_cache["stub"] = prediction_service_pb2_grpc.PredictionServiceStub( channel) # wait a bit for tf serving to start before querying metadata limit = 60 for i in range(limit): try: local_cache["metadata"] = run_get_model_metadata() break except Exception as e: if i > 6: logger.warn( "unable to read model metadata - model is still loading. Retrying..." ) if i == limit - 1: logger.exception("retry limit exceeded") sys.exit(1) time.sleep(5) logger.info("model_signature: {}".format( extract_signature( local_cache["metadata"]["signatureDef"], local_cache["api"]["tf_serving"]["signature_key"], ))) serve(app, listen="*:{}".format(args.port))
def start(args): api = None try: ctx = Context(s3_path=args.context, cache_dir=args.cache_dir, workload_id=args.workload_id) api = ctx.apis_id_map[args.api] local_cache["api"] = api local_cache["ctx"] = ctx if api["predictor"]["type"] != "onnx": raise CortexException(api["name"], "predictor type is not onnx") cx_logger().info("loading the predictor from {}".format(api["predictor"]["path"])) _, prefix = ctx.storage.deconstruct_s3_path(api["predictor"]["model"]) model_path = os.path.join(args.model_dir, os.path.basename(prefix)) local_cache["client"] = ONNXClient(model_path) predictor_class = ctx.get_predictor_class(api["name"], args.project_dir) try: local_cache["predictor"] = predictor_class( local_cache["client"], api["predictor"]["config"] ) except Exception as e: raise UserRuntimeException(api["predictor"]["path"], "__init__", str(e)) from e finally: refresh_logger() except Exception as e: cx_logger().exception("failed to start api") sys.exit(1) if api.get("tracker") is not None and api["tracker"].get("model_type") == "classification": try: local_cache["class_set"] = api_utils.get_classes(ctx, api["name"]) except Exception as e: cx_logger().warn("an error occurred while attempting to load classes", exc_info=True) cx_logger().info("ONNX model signature: {}".format(local_cache["client"].input_signature)) waitress_kwargs = {} if api["predictor"].get("config") is not None: for key, value in api["predictor"]["config"].items(): if key.startswith("waitress_"): waitress_kwargs[key[len("waitress_") :]] = value if len(waitress_kwargs) > 0: cx_logger().info("waitress parameters: {}".format(waitress_kwargs)) waitress_kwargs["listen"] = "*:{}".format(args.port) cx_logger().info("{} api is live".format(api["name"])) open("/health_check.txt", "a").close() serve(app, **waitress_kwargs)
def start(args): api = None try: ctx = Context(s3_path=args.context, cache_dir=args.cache_dir, workload_id=args.workload_id) api = ctx.apis_id_map[args.api] local_cache["api"] = api local_cache["ctx"] = ctx if api.get("predictor") is None: raise CortexException(api["name"], "predictor key not configured") cx_logger().info("loading the predictor from {}".format( api["predictor"]["path"])) local_cache["predictor"] = ctx.get_predictor_impl( api["name"], args.project_dir) if util.has_function(local_cache["predictor"], "init"): try: model_path = None if api["predictor"].get("model") is not None: _, prefix = ctx.storage.deconstruct_s3_path( api["predictor"]["model"]) model_path = os.path.join( args.model_dir, os.path.basename(os.path.normpath(prefix))) cx_logger().info("calling the predictor's init() function") local_cache["predictor"].init(model_path, api["predictor"]["metadata"]) except Exception as e: raise UserRuntimeException(api["predictor"]["path"], "init", str(e)) from e finally: refresh_logger() except: cx_logger().exception("failed to start api") sys.exit(1) if api.get("tracker") is not None and api["tracker"].get( "model_type") == "classification": try: local_cache["class_set"] = api_utils.get_classes(ctx, api["name"]) except Exception as e: cx_logger().warn( "an error occurred while attempting to load classes", exc_info=True) cx_logger().info("{} api is live".format(api["name"])) serve(app, listen="*:{}".format(args.port))
def start(args): api = None try: ctx = Context(s3_path=args.context, cache_dir=args.cache_dir, workload_id=args.workload_id) api = ctx.apis_id_map[args.api] local_cache["api"] = api local_cache["ctx"] = ctx _, prefix = ctx.storage.deconstruct_s3_path(api["model"]) model_path = os.path.join(args.model_dir, os.path.basename(prefix)) if api.get("request_handler") is not None: local_cache["request_handler"] = ctx.get_request_handler_impl( api["name"], args.project_dir) sess = rt.InferenceSession(model_path) local_cache["sess"] = sess local_cache["input_metadata"] = sess.get_inputs() logger.info("input_metadata: {}".format( truncate(extract_signature(local_cache["input_metadata"])))) local_cache["output_metadata"] = sess.get_outputs() logger.info("output_metadata: {}".format( truncate(extract_signature(local_cache["output_metadata"])))) 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": 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) serve(app, listen="*:{}".format(args.port))
def start(args): api = None try: ctx = Context(s3_path=args.context, cache_dir=args.cache_dir, workload_id=args.workload_id) api = ctx.apis_id_map[args.api] local_cache["api"] = api local_cache["ctx"] = ctx if api.get("tensorflow") is None: raise CortexException(api["name"], "tensorflow key not configured") if api["tensorflow"].get("request_handler") is not None: cx_logger().info("loading the request handler from {}".format( api["tensorflow"]["request_handler"])) local_cache["request_handler"] = ctx.get_request_handler_impl( api["name"], args.project_dir) request_handler = local_cache.get("request_handler") if request_handler is not None and util.has_function( request_handler, "pre_inference"): cx_logger().info( "using pre_inference request handler defined in {}".format( api["tensorflow"]["request_handler"])) else: cx_logger().info("pre_inference request handler not defined") if request_handler is not None and util.has_function( request_handler, "post_inference"): cx_logger().info( "using post_inference request handler defined in {}".format( api["tensorflow"]["request_handler"])) else: cx_logger().info("post_inference request handler not defined") except Exception as e: cx_logger().exception("failed to start api") sys.exit(1) try: validate_model_dir(args.model_dir) except Exception as e: cx_logger().exception("failed to validate model") sys.exit(1) if api.get("tracker") is not None and api["tracker"].get( "model_type") == "classification": try: local_cache["class_set"] = api_utils.get_classes(ctx, api["name"]) except Exception as e: cx_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) # wait a bit for tf serving to start before querying metadata limit = 60 for i in range(limit): try: local_cache["model_metadata"] = run_get_model_metadata() break except Exception as e: if i > 6: cx_logger().warn( "unable to read model metadata - model is still loading. Retrying..." ) if i == limit - 1: cx_logger().exception("retry limit exceeded") sys.exit(1) time.sleep(5) signature_key, parsed_signature = extract_signature( local_cache["model_metadata"]["signatureDef"], api["tensorflow"]["signature_key"]) local_cache["signature_key"] = signature_key local_cache["parsed_signature"] = parsed_signature cx_logger().info("model_signature: {}".format( local_cache["parsed_signature"])) cx_logger().info("{} API is live".format(api["name"])) serve(app, listen="*:{}".format(args.port))