feat: gRPC client implementation

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

grpc-client/python/client.py

@@ -1,32 +1,43 @@
+from __future__ import annotations
+
 import asyncio
 
-import grpc
+import numpy as np
+
+from bentoml.client import Client
+
 
-from bentoml.grpc.utils import import_generated_stubs
+async def arun(client: Client):
 
-pb, services = import_generated_stubs()
+    res = await client.async_classify(np.array([[5.9, 3, 5.1, 1.8]]))
+    print("Result from 'client.async_classify':\n", res)
+    res = await client.async_call("classify", np.array([[5.9, 3, 5.1, 1.8]]))
+    print("Result from 'client.async_call':\n", res)
 
 
-async def run():
-    async with grpc.aio.insecure_channel("localhost:3000") as channel:
-        stub = services.BentoServiceStub(channel)
-        req = await stub.Call(
-            request=pb.Request(
-                api_name="classify",
-                ndarray=pb.NDArray(
-                    dtype=pb.NDArray.DTYPE_FLOAT,
-                    shape=(1, 4),
-                    float_values=[5.9, 3, 5.1, 1.8],
-                ),
-            )
-        )
-    print(req)
+def run(client: Client):
+    res = client.classify(np.array([[5.9, 3, 5.1, 1.8]]))
+    print("Result from 'client.classify':\n", res)
+    res = client.call("classify", np.array([[5.9, 3, 5.1, 1.8]]))
+    print("Result from 'client.call(bentoml_api_name='classify')':\n", res)
 
 
 if __name__ == "__main__":
-    loop = asyncio.new_event_loop()
-    try:
-        loop.run_until_complete(run())
-    finally:
-        loop.close()
-        assert loop.is_closed()
+    import argparse
+
+    parser = argparse.ArgumentParser()
+    parser.add_argument("-rwa", "--run-with-async", action="store_true", default=False)
+    parser.add_argument("--grpc", action="store_true", default=False)
+    args = parser.parse_args()
+
+    c = Client.from_url("localhost:3000", grpc=args.grpc)
+
+    if args.run_with_async:
+        loop = asyncio.new_event_loop()
+        try:
+            loop.run_until_complete(arun(c))
+        finally:
+            loop.close()
+            assert loop.is_closed()
+    else:
+        run(c)

pyproject.toml

@@ -136,17 +136,15 @@ grpc = [
   # Restrict maximum version due to breaking protobuf 4.21.0 changes
   # (see https://github.com/protocolbuffers/protobuf/issues/10051)
   # 3.19.5 is currently breaking on a lot of system.
-  "protobuf>=3.5.0, <3.20, !=3.19.5",
+  "protobuf>=3.5.0,<4.0dev,!=3.19.5",
   # Lowest version that support 3.10. We need to set an upper bound
   # We can't use 1.48.2 since it depends on 3.19.5
-  "grpcio>=1.41.0,!=1.48.2",
-  # grpcio>=1.48.0 provides a pre-built M1 wheel.
-  "grpcio>=1.48.0,!=1.48.2;platform_machine=='arm64' and platform_system=='Darwin'",
-  "grpcio-health-checking>=1.41.0,!=1.48.2",
+  "grpcio>=1.41.0",
+  "grpcio-health-checking>=1.41.0",
   "opentelemetry-instrumentation-grpc==0.35b0",
 ]
-grpc-reflection = ["bentoml[grpc]", "grpcio-reflection>=1.41.0,!=1.48.2"]
-grpc-channelz = ["bentoml[grpc]", "grpcio-channelz>=1.41.0,!=1.48.2"]
+grpc-reflection = ["bentoml[grpc]", "grpcio-reflection>=1.41.0"]
+grpc-channelz = ["bentoml[grpc]", "grpcio-channelz>=1.41.0"]
 # We kept for compatibility with previous
 # versions of BentoML. It is discouraged to use this, instead use any
 # of the above tracing.* extras.
@@ -316,7 +314,7 @@ skip_glob = [
 ]
 
 [tool.pyright]
-pythonVersion = "3.10"
+pythonVersion = "3.11"
 include = ["src/", "examples/", "tests/"]
 exclude = [
   'src/bentoml/_version.py',

requirements/tests-requirements.txt

@@ -17,6 +17,4 @@ imageio==2.22.4
 pyarrow==10.0.1
 build[virtualenv]==0.9.0
 protobuf==3.19.6
-grpcio>=1.41.0, <1.49, !=1.48.2
-grpcio-health-checking>=1.41.0, <1.49, !=1.48.2
 opentelemetry-instrumentation-grpc==0.35b0

scripts/generate_grpc_stubs.sh

@@ -1,15 +1,18 @@
 #!/usr/bin/env bash
 
+set -e
+
 GIT_ROOT=$(git rev-parse --show-toplevel)
 STUBS_GENERATOR="bentoml/stubs-generator"
+BASEDIR="$(cd -- "$(dirname -- "${BASH_SOURCE[0]:-$0}")" &>/dev/null && pwd 2>/dev/null)"
 
 cd "$GIT_ROOT/src" || exit 1
 
 main() {
 	local VERSION="${1:-v1}"
 	# Use inline heredoc for even faster build
 	# Keeping image as cache should be fine since we only want to generate the stubs.
-	if [[ $(docker images --filter=reference="$STUBS_GENERATOR" -q) == "" ]] || test "$(git diff --name-only --diff-filter=d -- "$0")"; then
+	if test "$(git diff --name-only --diff-filter=d -- "$BASEDIR/$(basename "$0")")" || [[ $(docker images --filter=reference="$STUBS_GENERATOR" -q) == "" ]]; then
 		docker buildx build --platform=linux/amd64 -t "$STUBS_GENERATOR" --load -f- . <<EOF
 # syntax=docker/dockerfile:1.4-labs
 

src/bentoml/_internal/io_descriptors/json.py

@@ -240,11 +240,11 @@ def from_spec(cls, spec: SpecDict) -> Self:
         if "args" not in spec:
             raise InvalidArgument(f"Missing args key in JSON spec: {spec}")
         if "has_pydantic_model" in spec["args"] and spec["args"]["has_pydantic_model"]:
-            logger.warning(
+            logger.debug(
                 "BentoML does not support loading pydantic models from URLs; output will be a normal dictionary."
             )
         if "has_json_encoder" in spec["args"] and spec["args"]["has_json_encoder"]:
-            logger.warning(
+            logger.debug(
                 "BentoML does not support loading JSON encoders from URLs; output will be a normal dictionary."
             )
 

src/bentoml/_internal/server/grpc/server.py

@@ -48,7 +48,7 @@
     )
 
 
-def _load_from_file(p: str) -> bytes:
+def load_from_file(p: str) -> bytes:
     rp = resolve_user_filepath(p, ctx=None)
     with open(rp, "rb") as f:
         return f.read()
@@ -164,12 +164,12 @@ def configure_port(self, addr: str):
             ), "'ssl_keyfile' is required when 'ssl_certfile' is provided."
             if self.ssl_ca_certs is not None:
                 client_auth = True
-                ca_cert = _load_from_file(self.ssl_ca_certs)
+                ca_cert = load_from_file(self.ssl_ca_certs)
             server_credentials = grpc.ssl_server_credentials(
                 (
                     (
-                        _load_from_file(self.ssl_keyfile),
-                        _load_from_file(self.ssl_certfile),
+                        load_from_file(self.ssl_keyfile),
+                        load_from_file(self.ssl_certfile),
                     ),
                 ),
                 root_certificates=ca_cert,

src/bentoml/_internal/server/grpc/servicer.py

@@ -169,6 +169,7 @@ async def Call(
                 else:
                     output = await api.func(input_data)
             else:
+                assert api.func is not None
                 if api.multi_input:
                     output = await anyio.to_thread.run_sync(api.func, **input_data)
                 else:
@@ -212,10 +213,12 @@ async def ServiceMetadata(
                         name=api.name,
                         docs=api.doc,
                         input=make_descriptor_spec(
-                            api.input.to_spec(), ServiceMetadataResponse
+                            t.cast("SpecDict", api.input.to_spec()),
+                            ServiceMetadataResponse,
                         ),
                         output=make_descriptor_spec(
-                            api.output.to_spec(), ServiceMetadataResponse
+                            t.cast("SpecDict", api.output.to_spec()),
+                            ServiceMetadataResponse,
                         ),
                     )
                     for api in service.apis.values()

src/bentoml/_internal/service/inference_api.py

@@ -26,7 +26,7 @@
 class InferenceAPI:
     def __init__(
         self,
-        user_defined_callback: t.Callable[..., t.Any],
+        user_defined_callback: t.Callable[..., t.Any] | None,
         input_descriptor: IODescriptor[t.Any],
         output_descriptor: IODescriptor[t.Any],
         name: Optional[str],