chore: fixing bad merge

Signed-off-by: Aaron Pham <29749331+aarnphm@users.noreply.github.com>

MANIFEST.in

@@ -1,11 +1,14 @@
 # All files tracked by Git are included in PyPI distribution
+include bentoml/grpc/**/*_pb2*.py
+include bentoml/grpc/**/*.pyi
 
 # Files to exclude in PyPI distribution
 exclude CONTRIBUTING.md GOVERNANCE.md CODE_OF_CONDUCT.md DEVELOPMENT.md SECURITY.md
 exclude Makefile MANIFEST.in
 exclude *.yml
 exclude .git*
 exclude bentoml/grpc/buf.yaml
+exclude bentoml/_internal/frameworks/FRAMEWORK_TEMPLATE_PY
 
 # Directories to exclude in PyPI package
 prune requirements
@@ -19,6 +22,7 @@ prune .git*
 prune */__pycache__
 prune */.DS_Store
 prune */.ipynb_checkpoints
+prune */README*
 
 # Patterns to exclude from any directory
 global-exclude *.py[cod]

bentoml/_internal/bento/build_config.py

@@ -484,7 +484,7 @@ def write_to_bento(self, bento_fs: FS, build_ctx: str) -> None:
             install_script_content = (
                 """\
 #!/usr/bin/env bash
-set -ex
+set -exuo pipefail
 
 # Parent directory https://stackoverflow.com/a/246128/8643197
 BASEDIR="$( cd -- "$( dirname -- "${BASH_SOURCE[0]:-$0}"; )" &> /dev/null && pwd 2> /dev/null; )"

bentoml/_internal/bento/build_dev_bentoml_whl.py

@@ -4,13 +4,15 @@
 import logging
 
 from bentoml.exceptions import BentoMLException
+from bentoml.exceptions import MissingDependencyException
 
 from ..utils.pkg import source_locations
 from ..configuration import is_pypi_installed_bentoml
 
 logger = logging.getLogger(__name__)
 
 BENTOML_DEV_BUILD = "BENTOML_BUNDLE_LOCAL_BUILD"
+_exc_message = f"'{BENTOML_DEV_BUILD}=True', which requires the 'pypa/build' package. Install development dependencies with 'pip install -r requirements/dev-requirements.txt' and try again."
 
 
 def build_bentoml_editable_wheel(target_path: str) -> None:
@@ -26,6 +28,12 @@ def build_bentoml_editable_wheel(target_path: str) -> None:
         # skip this entirely if BentoML is installed from PyPI
         return
 
+    try:
+        from build import ProjectBuilder
+        from build.env import IsolatedEnvBuilder
+    except ModuleNotFoundError:
+        raise MissingDependencyException(_exc_message)
+
     # Find bentoml module path
     module_location = source_locations("bentoml")
     if not module_location:
@@ -41,15 +49,6 @@ def build_bentoml_editable_wheel(target_path: str) -> None:
             "BentoML is installed in `editable` mode; building BentoML distribution with the local BentoML code base. The built wheel file will be included in the target bento."
         )
 
-        try:
-            from build.env import IsolatedEnvBuilder
-
-            from build import ProjectBuilder
-        except ModuleNotFoundError:
-            raise BentoMLException(
-                f"'{BENTOML_DEV_BUILD}=True', which requires the 'pypa/build' package. Install development dependencies with 'pip install -r requirements/dev-requirements.txt' and try again."
-            )
-
         with IsolatedEnvBuilder() as env:
             builder = ProjectBuilder(os.path.dirname(pyproject))
             builder.python_executable = env.executable

bentoml/_internal/bento/docker/templates/base.j2

@@ -75,6 +75,10 @@ RUN {{ __setup_script__ }}
 {% block SETUP_BENTO_ENTRYPOINT %}
 # Default port for BentoServer
 EXPOSE 3000
+{% if __options__grpc is true %}
+# expose Prometheus port
+EXPOSE 9090
+{% endif %}
 
 RUN chmod +x {{ bento__entrypoint }}
 

bentoml/_internal/configuration/containers.py

@@ -395,30 +395,23 @@ def tracer_provider(
 
         if tracer_type == "zipkin" and zipkin_server_url is not None:
             # pylint: disable=no-name-in-module # https://github.com/open-telemetry/opentelemetry-python-contrib/issues/290
-            from opentelemetry.exporter.zipkin.json import (
-                ZipkinExporter,  # type: ignore (no opentelemetry types)
-            )
+            from opentelemetry.exporter.zipkin.json import ZipkinExporter
 
-            exporter = ZipkinExporter(  # type: ignore (no opentelemetry types)
-                endpoint=zipkin_server_url,
-            )
-            provider.add_span_processor(BatchSpanProcessor(exporter))  # type: ignore (no opentelemetry types)
+            exporter = ZipkinExporter(endpoint=zipkin_server_url)
+            provider.add_span_processor(BatchSpanProcessor(exporter))
             return provider
         elif (
             tracer_type == "jaeger"
             and jaeger_server_address is not None
             and jaeger_server_port is not None
         ):
             # pylint: disable=no-name-in-module # https://github.com/open-telemetry/opentelemetry-python-contrib/issues/290
-            from opentelemetry.exporter.jaeger.thrift import (
-                JaegerExporter,  # type: ignore (no opentelemetry types)
-            )
+            from opentelemetry.exporter.jaeger.thrift import JaegerExporter
 
-            exporter = JaegerExporter(  # type: ignore (no opentelemetry types)
-                agent_host_name=jaeger_server_address,
-                agent_port=jaeger_server_port,
+            exporter = JaegerExporter(
+                agent_host_name=jaeger_server_address, agent_port=jaeger_server_port
             )
-            provider.add_span_processor(BatchSpanProcessor(exporter))  # type: ignore (no opentelemetry types)
+            provider.add_span_processor(BatchSpanProcessor(exporter))
             return provider
         else:
             return provider
@@ -436,7 +429,7 @@ def tracing_excluded_urls(
         if isinstance(excluded_urls, list):
             return ExcludeList(excluded_urls)
         else:
-            return parse_excluded_urls(excluded_urls)
+            return parse_excluded_urls(excluded_urls)  # type: ignore (already handle None)
 
     # Mapping from runner name to RunnerApp file descriptor
     remote_runner_mapping = providers.Static[t.Dict[str, str]]({})

bentoml/_internal/configuration/default_configuration.yaml

@@ -26,7 +26,7 @@ api_server:
   grpc:
     # https://github.com/grpc/grpc/blob/ebfb028a29ef5f0f4be6f33c131c740ae0a32c84/src/core/ext/transport/chttp2/transport/http2_settings.cc#L51
     max_concurrent_streams: ~
-    max_message_length: ~
+    max_message_length: -1
     maximum_concurrent_rpcs: ~
     metrics_port: 9090
     metrics_host: 0.0.0.0

bentoml/_internal/external_typing/__init__.py

@@ -4,6 +4,8 @@
 if TYPE_CHECKING:
     from typing import Literal
 
+    F = t.Callable[..., t.Any]
+
     from pandas import Series as _PdSeries
     from pandas import DataFrame as PdDataFrame
     from pyarrow.plasma import ObjectID
@@ -30,6 +32,8 @@
     from .starlette import ASGIReceive
     from .starlette import AsgiMiddleware
 
+    WSGIApp = t.Callable[[F, t.Mapping[str, t.Any]], t.Iterable[bytes]]
+
     __all__ = [
         "PdSeries",
         "PdDataFrame",
@@ -51,4 +55,6 @@
         "ASGISend",
         "ASGIReceive",
         "ASGIMessage",
+        # misc
+        "WSGIApp",
     ]

bentoml/_internal/io_descriptors/base.py

@@ -24,6 +24,13 @@
     from ..service.openapi.specification import Reference
     from ..service.openapi.specification import RequestBody
 
+    InputType = (
+        UnionType
+        | t.Type[t.Any]
+        | LazyType[t.Any]
+        | dict[str, t.Type[t.Any] | UnionType | LazyType[t.Any]]
+    )
+
 
 IOType = t.TypeVar("IOType")
 
@@ -78,14 +85,7 @@ def accepted_proto_fields(self) -> list[str]:
         return self._proto_fields
 
     @abstractmethod
-    def input_type(
-        self,
-    ) -> t.Union[
-        "UnionType",
-        t.Type[t.Any],
-        "LazyType[t.Any]",
-        t.Dict[str, t.Union[t.Type[t.Any], "UnionType", "LazyType[t.Any]"]],
-    ]:
+    def input_type(self) -> InputType:
         ...
 
     @abstractmethod

bentoml/_internal/io_descriptors/numpy.py

@@ -188,13 +188,17 @@ def __init__(
             try:
                 dtype = np.dtype(dtype)
             except TypeError as e:
-                raise UnprocessableEntity(f'NumpyNdarray: Invalid dtype "{dtype}": {e}')
+                raise UnprocessableEntity(
+                    f'NumpyNdarray: Invalid dtype "{dtype}": {e}'
+                ) from e
 
-        self._dtype: ext.NpDTypeLike | None = dtype
+        self._dtype = dtype
         self._shape = shape
         self._enforce_dtype = enforce_dtype
         self._enforce_shape = enforce_shape
 
+        self._sample_input = None
+
         # whether to use packed representation of numpy while sending protobuf
         # this means users should be using raw_value instead of array_value or multi_dimensional_array_value
         self._packed = packed
@@ -207,8 +211,6 @@ def __init__(
             # https://numpy.org/doc/stable/user/basics.byteswapping.html#introduction-to-byte-ordering-and-ndarrays
         self._bytesorder: t.Literal["C", "F", "A"] = bytesorder
 
-        self._sample_input = None
-
     def _openapi_types(self) -> str:
         # convert numpy dtypes to openapi compatible types.
         var_type = "integer"
@@ -270,11 +272,11 @@ def openapi_responses(self) -> OpenAPIResponse:
 
     def _verify_ndarray(
         self,
-        obj: "ext.NpNDArray",
-        dtype: np.dtype[t.Any] | None,
+        obj: ext.NpNDArray,
+        dtype: ext.NpDTypeLike | None,
         shape: tuple[int, ...] | None,
         exception_cls: t.Type[Exception] = BadInput,
-    ) -> "ext.NpNDArray":
+    ) -> ext.NpNDArray:
         if dtype is not None and dtype != obj.dtype:
             # ‘same_kind’ means only safe casts or casts within a kind, like float64
             # to float32, are allowed.
@@ -408,6 +410,7 @@ async def from_grpc_request(
         else:
             # {'float_contents': [1.0, 2.0, 3.0]}
             array = serialized
+            shape = self._shape
 
         dtype_string, content = get_array_proto(array)
         dtype = np.dtype(_VALUES_TO_NP_DTYPE_MAP[dtype_string])

bentoml/_internal/server/grpc/interceptors/__init__.py

@@ -1,25 +1,17 @@
 from __future__ import annotations
 
 import typing as t
-import logging
 import functools
-from timeit import default_timer
 from typing import TYPE_CHECKING
 
-import grpc
 from grpc import aio
 
 from ....utils import LazyLoader
 from ....utils.grpc import ProtoCodec
-from ....utils.grpc import to_http_status
 from ....utils.grpc import wrap_rpc_handler
 from ....utils.grpc import get_grpc_content_type
-from ....utils.grpc.codec import GRPC_CONTENT_TYPE
 
 if TYPE_CHECKING:
-    from grpc.aio._typing import MetadataType
-
-    from bentoml.grpc.v1 import service_pb2
 
     from ..types import Request
     from ..types import Response
@@ -31,13 +23,11 @@
 else:
     service_pb2 = LazyLoader("service_pb2", globals(), "bentoml.grpc.v1.service_pb2")
 
-logger = logging.getLogger(__name__)
-
 
 class GenericHeadersServerInterceptor(aio.ServerInterceptor):
     """
     A light header interceptor that provides some initial metadata to the client.
-    TODO: https://chromium.googlesource.com/external/github.com/grpc/grpc/+/HEAD/doc/PROTOCOL-HTTP2.md
+    Refers to https://chromium.googlesource.com/external/github.com/grpc/grpc/+/HEAD/doc/PROTOCOL-HTTP2.md
     """
 
     def __init__(self, *, codec: Codec | None = None):
@@ -82,68 +72,3 @@ async def new_behaviour(
             return new_behaviour
 
         return wrap_rpc_handler(wrapper, handler)
-
-
-class AccessLogServerInterceptor(aio.ServerInterceptor):
-    """
-    An asyncio interceptor for access log.
-    """
-
-    async def intercept_service(
-        self,
-        continuation: t.Callable[[HandlerCallDetails], t.Awaitable[RpcMethodHandler]],
-        handler_call_details: HandlerCallDetails,
-    ) -> RpcMethodHandler:
-        logger = logging.getLogger("bentoml.access")
-        handler = await continuation(handler_call_details)
-        method_name = handler_call_details.method
-
-        if handler and (handler.response_streaming or handler.request_streaming):
-            return handler
-
-        def wrapper(behaviour: AsyncHandlerMethod[Response]):
-            @functools.wraps(behaviour)
-            async def new_behaviour(
-                request: Request, context: BentoServicerContext
-            ) -> Response | t.Awaitable[Response]:
-
-                content_type = GRPC_CONTENT_TYPE
-
-                trailing_metadata: MetadataType | None = context.trailing_metadata()
-                if trailing_metadata:
-                    trailing = dict(trailing_metadata)
-                    content_type = trailing.get("content-type", GRPC_CONTENT_TYPE)
-
-                start = default_timer()
-                try:
-                    response = behaviour(request, context)
-                    if not hasattr(response, "__aiter__"):
-                        response = await response
-                except Exception as e:
-                    context.set_code(grpc.StatusCode.INTERNAL)
-                    context.set_details(str(e))
-                    response = service_pb2.Response()
-                finally:
-                    latency = max(default_timer() - start, 0)
-                    req = [
-                        "scheme=http",  # TODO: support https when ssl is added
-                        f"path={method_name}",
-                        f"type={content_type}",
-                        f"size={request.ByteSize()}",
-                    ]
-                    resp = [
-                        f"http_status={to_http_status(context.code())}",
-                        f"grpc_status={context.code().value[0]}",
-                        f"type={content_type}",
-                        f"size={response.ByteSize()}",
-                    ]
-
-                    # TODO: fix ports
-                    logger.info(
-                        f"{context.peer()} ({','.join(req)}) ({','.join(resp)}) {latency:.3f}ms"
-                    )
-                return response
-
-            return new_behaviour
-
-        return wrap_rpc_handler(wrapper, handler)