ci: Merge change for release 1.7.0

Dockerfile

@@ -1,7 +1,9 @@
 FROM python:3.10-slim AS wheel-builder
 SHELL ["/bin/bash", "-l", "-c"]
 
-ARG POETRY_VERSION="1.8.3"
+# TODO: Upgrade poetry version to 2.x
+# Currently it fails in `poetry export --with all-runtimes --without-hashes --format constraints.txt -o /opt/mlserver/dist/constraints.txt`
+ARG POETRY_VERSION="1.8.5"
 
 COPY ./hack/build-wheels.sh ./hack/build-wheels.sh
 COPY ./mlserver ./mlserver

Makefile

@@ -18,11 +18,11 @@ install-dev:
 
 lock:
 	echo "Locking mlserver deps..."
-	poetry lock --no-update
+	poetry lock
 	for _runtime in ./runtimes/*; \
 	do \
 		echo "Locking $$_runtime deps..."; \
-		poetry lock --no-update -C $$_runtime; \
+		poetry lock -C $$_runtime; \
 	done
 
 _generate_model_repository: # "private" target to call `fmt` after `generate`

mlserver/batching/adaptive.py

@@ -136,7 +136,7 @@ async def _batch_requests(self) -> AsyncIterator[BatchedRequests]:
 
                     # Update remaining timeout
                     current = time.time()
-                    timeout = timeout - (current - start)
+                    timeout = self._max_batch_time - (current - start)
             except asyncio.TimeoutError:
                 # NOTE: Hit timeout, continue
                 pass

mlserver/codecs/json.py

@@ -1,14 +1,17 @@
 # seperate file to side step circular dependency on the decode_str function
 
-from typing import Any, Union
 import json
+import numpy as np
+from typing import Any, List, Union
 
 try:
     import orjson
 except ImportError:
     orjson = None  # type: ignore
 
 from .string import decode_str
+from .lists import as_list
+from .utils import InputOrOutput
 
 
 # originally taken from: mlserver/rest/responses.py
@@ -57,5 +60,36 @@ def encode_to_json_bytes(v: Any) -> bytes:
 def decode_from_bytelike_json_to_dict(v: Union[bytes, str]) -> dict:
     if orjson is None:
         return json.loads(v)
-
     return orjson.loads(v)
+
+
+class JSONEncoderWithArray(json.JSONEncoder):
+    def default(self, obj):
+        if isinstance(obj, np.ndarray):
+            return obj.tolist()
+        elif isinstance(obj, np.floating):
+            return float(obj)
+        elif isinstance(obj, np.integer):
+            return int(obj)
+        else:
+            return json.JSONEncoder.default(self, obj)
+
+
+def encode_to_json(v: Any, use_bytes: bool = True) -> Union[str, bytes]:
+    enc_v = json.dumps(
+        v,
+        ensure_ascii=False,
+        allow_nan=False,
+        indent=None,
+        separators=(",", ":"),
+        cls=JSONEncoderWithArray,
+    )
+    if use_bytes:
+        enc_v = enc_v.encode("utf-8")  # type: ignore[assignment]
+    return enc_v
+
+
+def decode_json_input_or_output(input_or_output: InputOrOutput) -> List[Any]:
+    packed = input_or_output.data.root
+    unpacked = map(json.loads, as_list(packed))
+    return list(unpacked)

mlserver/codecs/pandas.py

@@ -4,7 +4,8 @@
 from typing import Optional, Any, List, Tuple
 
 from .base import RequestCodec, register_request_codec
-from .numpy import to_datatype, to_dtype, convert_nan
+from .numpy import to_dtype, convert_nan, to_datatype
+from .json import decode_json_input_or_output, encode_to_json
 from .string import encode_str, StringCodec
 from .utils import get_decoded_or_raw, InputOrOutput, inject_batch_dimension
 from .lists import ListElement
@@ -19,8 +20,12 @@
 
 
 def _to_series(input_or_output: InputOrOutput) -> pd.Series:
-    payload = get_decoded_or_raw(input_or_output)
+    parameters = input_or_output.parameters
+
+    if parameters and parameters.content_type == PandasCodec.JsonContentType:
+        return pd.Series(decode_json_input_or_output(input_or_output))
 
+    payload = get_decoded_or_raw(input_or_output)
     if Datatype(input_or_output.datatype) == Datatype.BYTES:
         # Don't convert the dtype of BYTES
         return pd.Series(payload)
@@ -43,7 +48,13 @@ def _to_response_output(series: pd.Series, use_bytes: bool = True) -> ResponseOu
 
     content_type = None
     if datatype == Datatype.BYTES:
-        data, content_type = _process_bytes(data, use_bytes)
+        processed_data, content_type = _process_bytes(data, use_bytes)
+
+        if content_type is None:
+            data = [encode_to_json(elem, use_bytes) for elem in data]
+            content_type = PandasCodec.JsonContentType
+        else:
+            data = processed_data
 
     shape = inject_batch_dimension(list(series.shape))
     parameters = None
@@ -90,6 +101,7 @@ class PandasCodec(RequestCodec):
     """
 
     ContentType = "pd"
+    JsonContentType = "pd_json"
     TypeHint = pd.DataFrame
 
     @classmethod

mlserver/grpc/server.py

@@ -70,7 +70,7 @@ def _create_server(self):
 
         self._server = aio.server(
             ThreadPoolExecutor(max_workers=DefaultGrpcWorkers),
-            interceptors=tuple(self._interceptors),
+            interceptors=list(self._interceptors),
             options=self._get_options(),
         )