refactor: Removing Server subclass from tritonfrontend (#7683)

docs/customization_guide/tritonfrontend.md

@@ -59,11 +59,11 @@ Note: `model_path` may need to be edited depending on your setup.
 ```python
 from tritonfrontend import KServeHttp, KServeGrpc
 http_options = KServeHttp.Options(thread_count=5)
-http_service = KServeHttp.Server(server, http_options)
+http_service = KServeHttp(server, http_options)
 http_service.start()
 
 # Default options (if none provided)
-grpc_service = KServeGrpc.Server(server)
+grpc_service = KServeGrpc(server)
 grpc_service.start()
 ```
 
@@ -110,7 +110,7 @@ from tritonfrontend import KServeHttp
 import tritonclient.http as httpclient
 import numpy as np # Use version numpy < 2
 
-with KServeHttp.Server(server) as http_service:
+with KServeHttp(server) as http_service:
  # The identity model returns an exact duplicate of the input data as output
  model_name = "identity"
  url = "localhost:8000"

qa/L0_python_api/test_kserve.py

@@ -118,7 +118,7 @@ def test_invalid_options(self, frontend):
  tritonserver.InvalidArgumentError,
  match="Incorrect type for options. options argument must be of type",
  ):
- frontend.Server(server, {"port": 8001})
+ frontend(server, {"port": 8001})
 
  utils.teardown_server(server)
 

qa/L0_python_api/testing_utils.py

@@ -64,7 +64,7 @@ def setup_service(
  frontend: Union[KServeHttp, KServeGrpc],
  options=None,
 ) -> Union[KServeHttp, KServeGrpc]:
- service = frontend.Server(server=server, options=options)
+ service = frontend(server=server, options=options)
  service.start()
  return service
 

src/python/examples/example.py

@@ -51,7 +51,7 @@ def main():
  http_options = KServeHttp.Options(port=8005)
 
  # or http_service = KServeHttp.Server(server, http_options) & http_service.stop()
- with KServeHttp.Server(server, http_options) as http_service:
+ with KServeHttp(server, http_options) as http_service:
  # The identity model returns an exact duplicate of the input data as output
  model_name = "identity"
  url = "localhost:8005"
@@ -74,7 +74,6 @@ def main():
  output_data = results.as_numpy("OUTPUT0")
 
  print("--------------------- INFERENCE RESULTS ---------------------")
- print("Input data:", input_data)
  print("Output data:", output_data)
  print("-------------------------------------------------------------")
 

src/python/tritonfrontend/_api/_kservegrpc.py

@@ -90,48 +90,47 @@ def __post_init__(self):
  if isinstance(self.infer_compression_level, Grpc_compression_level):
  self.infer_compression_level = self.infer_compression_level.value
 
- class Server:
- def __init__(self, server: tritonserver, options: "KServeGrpc.Options" = None):
- try:
- server_ptr = server._ptr() # TRITONSERVER_Server pointer
-
- # If no options provided, default options are selected
- if options is None:
- options = KServeGrpc.Options()
-
- if not isinstance(options, KServeGrpc.Options):
- raise InvalidArgumentError(
- "Incorrect type for options. options argument must be of type KServeGrpc.Options"
- )
-
- # Converts dataclass instance -> python dictionary -> unordered_map<string, std::variant<...>>
- options_dict: dict[str, Union[int, bool, str]] = options.__dict__
-
- self.triton_frontend = TritonFrontendGrpc(server_ptr, options_dict)
- except TritonError:
- exc_type, exc_value, _ = sys.exc_info()
- # raise ... from None masks the tritonfrontend Error from being added in traceback
- raise ERROR_MAPPING[exc_type](exc_value) from None
-
- def __enter__(self):
+ def __init__(self, server: tritonserver, options: "KServeGrpc.Options" = None):
+ try:
+ server_ptr = server._ptr() # TRITONSERVER_Server pointer
+
+ # If no options provided, default options are selected
+ if options is None:
+ options = KServeGrpc.Options()
+
+ if not isinstance(options, KServeGrpc.Options):
+ raise InvalidArgumentError(
+ "Incorrect type for options. options argument must be of type KServeGrpc.Options"
+ )
+
+ # Converts dataclass instance -> python dictionary -> unordered_map<string, std::variant<...>>
+ options_dict: dict[str, Union[int, bool, str]] = options.__dict__
+
+ self.triton_frontend = TritonFrontendGrpc(server_ptr, options_dict)
+ except TritonError:
+ exc_type, exc_value, _ = sys.exc_info()
+ # raise ... from None masks the tritonfrontend Error from being added in traceback
+ raise ERROR_MAPPING[exc_type](exc_value) from None
+
+ def __enter__(self):
+ self.triton_frontend.start()
+ return self
+
+ def __exit__(self, exc_type, exc_value, traceback):
+ self.triton_frontend.stop()
+ if exc_type:
+ raise ERROR_MAPPING[exc_type](exc_value) from None
+
+ def start(self):
+ try:
  self.triton_frontend.start()
- return self
+ except TritonError:
+ exc_type, exc_value, _ = sys.exc_info()
+ raise ERROR_MAPPING[exc_type](exc_value) from None
 
- def __exit__(self, exc_type, exc_value, traceback):
+ def stop(self):
+ try:
  self.triton_frontend.stop()
- if exc_type:
- raise ERROR_MAPPING[exc_type](exc_value) from None
-
- def start(self):
- try:
- self.triton_frontend.start()
- except TritonError:
- exc_type, exc_value, _ = sys.exc_info()
- raise ERROR_MAPPING[exc_type](exc_value) from None
-
- def stop(self):
- try:
- self.triton_frontend.stop()
- except TritonError:
- exc_type, exc_value, _ = sys.exc_info()
- raise ERROR_MAPPING[exc_type](exc_value) from None
+ except TritonError:
+ exc_type, exc_value, _ = sys.exc_info()
+ raise ERROR_MAPPING[exc_type](exc_value) from None

src/python/tritonfrontend/_api/_kservehttp.py

@@ -50,48 +50,47 @@ class Options:
  # DLIS-7215: Add restricted protocol support
  # restricted_protocols: list
 
- class Server:
- def __init__(self, server: tritonserver, options: "KServeHttp.Options" = None):
- try:
- server_ptr = server._ptr() # TRITONSERVER_Server pointer
+ def __init__(self, server: tritonserver, options: "KServeHttp.Options" = None):
+ try:
+ server_ptr = server._ptr() # TRITONSERVER_Server pointer
 
-  # If no options provided, default options are selected
-  if options is None:
-  options = KServeHttp.Options()
+ # If no options provided, default options are selected
+ if options is None:
+ options = KServeHttp.Options()
 
-  if not isinstance(options, KServeHttp.Options):
-  raise InvalidArgumentError(
-  "Incorrect type for options. options argument must be of type KServeHttp.Options"
-  )
+ if not isinstance(options, KServeHttp.Options):
+ raise InvalidArgumentError(
+ "Incorrect type for options. options argument must be of type KServeHttp.Options"
+ )
 
-  options_dict: dict[str, Union[int, bool, str]] = options.__dict__
-  # Converts dataclass instance -> python dictionary -> unordered_map<string, std::variant<...>>
+ options_dict: dict[str, Union[int, bool, str]] = options.__dict__
+ # Converts dataclass instance -> python dictionary -> unordered_map<string, std::variant<...>>
 
-  self.triton_frontend = TritonFrontendHttp(server_ptr, options_dict)
-  except TritonError:
-  exc_type, exc_value, _ = sys.exc_info()
-  # raise ... from None masks the tritonfrontend Error from being added in traceback
-  raise ERROR_MAPPING[exc_type](exc_value) from None
+ self.triton_frontend = TritonFrontendHttp(server_ptr, options_dict)
+ except TritonError:
+ exc_type, exc_value, _ = sys.exc_info()
+ # raise ... from None masks the tritonfrontend Error from being added in traceback
+ raise ERROR_MAPPING[exc_type](exc_value) from None
 
-  def __enter__(self):
-  self.triton_frontend.start()
-  return self
+ def __enter__(self):
+ self.triton_frontend.start()
+ return self
 
-  def __exit__(self, exc_type, exc_value, traceback):
-  self.triton_frontend.stop()
-  if exc_type:
-  raise ERROR_MAPPING[exc_type](exc_value) from None
+ def __exit__(self, exc_type, exc_value, traceback):
+ self.triton_frontend.stop()
+ if exc_type:
+ raise ERROR_MAPPING[exc_type](exc_value) from None
 
-  def start(self):
-  try:
-  self.triton_frontend.start()
-  except TritonError:
-  exc_type, exc_value, _ = sys.exc_info()
-  raise ERROR_MAPPING[exc_type](exc_value) from None
+ def start(self):
+ try:
+ self.triton_frontend.start()
+ except TritonError:
+ exc_type, exc_value, _ = sys.exc_info()
+ raise ERROR_MAPPING[exc_type](exc_value) from None
 
-  def stop(self):
-  try:
-  self.triton_frontend.stop()
-  except TritonError:
-  exc_type, exc_value, _ = sys.exc_info()
-  raise ERROR_MAPPING[exc_type](exc_value) from None
+ def stop(self):
+ try:
+ self.triton_frontend.stop()
+ except TritonError:
+ exc_type, exc_value, _ = sys.exc_info()
+ raise ERROR_MAPPING[exc_type](exc_value) from None