Add rest request gzip decompress

.dockerignore

@@ -26,6 +26,7 @@ eggs/
 .tox
 build/
 dist/
+venv/
 
 # Notebook Checkpoints
 .ipynb_checkpoints

.gitignore

@@ -29,6 +29,7 @@ eggs/
 .tox
 build/
 dist/
+venv/
 
 # Notebook Checkpoints
 .ipynb_checkpoints

mlserver/rest/app.py

@@ -117,7 +117,7 @@ def create_app(
         default_response_class=Response,
         exception_handlers=_EXCEPTION_HANDLERS,  # type: ignore
     )
-
+    app.router.route_class = APIRoute
     app.add_middleware(GZipMiddleware)
     if settings.cors_settings is not None:
         app.add_middleware(

mlserver/rest/requests.py

@@ -1,5 +1,5 @@
 from typing import Any
-
+from gzip import decompress as gzip_decompress
 from fastapi import Request as _Request
 
 try:
@@ -13,6 +13,13 @@ class Request(_Request):
     Custom request class which uses `orjson` if present.
     Otherwise, it falls back to the standard FastAPI request.
     """
+    async def body(self) -> bytes:
+        if not hasattr(self, "_body"):
+            body = await super().body()
+            if "gzip" in self.headers.getlist("Content-Encoding"):
+                body = gzip_decompress(body)
+            self._body = body
+        return self._body
 
     async def json(self) -> Any:
         if orjson is None:

tests/rest/test_custom.py

@@ -1,5 +1,6 @@
 from mlserver.types import InferenceRequest
-
+import gzip
+import json
 
 async def test_custom_handler(
     rest_client,
@@ -11,6 +12,18 @@ async def test_custom_handler(
     assert response.json() == 10
 
 
+async def test_custom_handler_gzip(rest_client, inference_request: InferenceRequest):
+    response = await rest_client.post(
+        "/my-custom-endpoint",
+        content = gzip.compress(json.dumps([1, 2, 3, 4]).encode('utf-8')),
+        headers = {
+            "Content-Encoding": "gzip",
+        }
+    )
+
+    assert response.status_code == 200
+    assert response.json() == 10
+
 async def test_gzip_compression(
     rest_client,
     inference_request,