Add gzip support

inference/core/interfaces/http/http_api.py

@@ -7,7 +7,7 @@
 
 import asgi_correlation_id
 import uvicorn
-from fastapi import BackgroundTasks, FastAPI, Path, Query, Request
+from fastapi import BackgroundTasks, Depends, FastAPI, Path, Query, Request
 from fastapi.middleware.cors import CORSMiddleware
 from fastapi.responses import JSONResponse, RedirectResponse, Response
 from fastapi.staticfiles import StaticFiles
@@ -158,6 +158,7 @@
     handle_describe_workflows_blocks_request,
     handle_describe_workflows_interface,
 )
+from inference.core.interfaces.http.middlewares.gzip import gzip_response_if_requested
 from inference.core.interfaces.http.orjson_utils import (
     orjson_response,
     serialise_workflow_result,
@@ -1243,8 +1244,11 @@ async def get_execution_engine_versions() -> ExecutionEngineVersions:
                 deprecated=True,
             )
             @with_route_exceptions
-            async def describe_workflows_blocks() -> WorkflowsBlocksDescription:
-                return handle_describe_workflows_blocks_request()
+            async def describe_workflows_blocks(
+                request: Request,
+            ) -> Union[WorkflowsBlocksDescription, Response]:
+                result = handle_describe_workflows_blocks_request()
+                return gzip_response_if_requested(request=request, response=result)
 
             @app.post(
                 "/workflows/blocks/describe",
@@ -1258,20 +1262,24 @@ async def describe_workflows_blocks() -> WorkflowsBlocksDescription:
             )
             @with_route_exceptions
             async def describe_workflows_blocks(
-                request: Optional[DescribeBlocksRequest] = None,
-            ) -> WorkflowsBlocksDescription:
+                request: Request,
+                request_payload: Optional[DescribeBlocksRequest] = None,
+            ) -> Union[WorkflowsBlocksDescription, Response]:
                 # TODO: get rid of async: https://github.com/roboflow/inference/issues/569
                 dynamic_blocks_definitions = None
                 requested_execution_engine_version = None
-                if request is not None:
-                    dynamic_blocks_definitions = request.dynamic_blocks_definitions
+                if request_payload is not None:
+                    dynamic_blocks_definitions = (
+                        request_payload.dynamic_blocks_definitions
+                    )
                     requested_execution_engine_version = (
-                        request.execution_engine_version
+                        request_payload.execution_engine_version
                     )
-                return handle_describe_workflows_blocks_request(
+                result = handle_describe_workflows_blocks_request(
                     dynamic_blocks_definitions=dynamic_blocks_definitions,
                     requested_execution_engine_version=requested_execution_engine_version,
                 )
+                return gzip_response_if_requested(request=request, response=result)
 
             @app.get(
                 "/workflows/definition/schema",

inference/core/interfaces/http/middlewares/gzip.py

@@ -0,0 +1,22 @@
+import gzip
+from typing import TypeVar, Union
+
+from fastapi import Request, Response
+from pydantic import BaseModel
+
+T = TypeVar("T", bound=BaseModel)
+
+
+def gzip_response_if_requested(
+    request: Request,
+    response: T,
+) -> Union[Response, T]:
+    if "gzip" not in request.headers.get("Accept-Encoding", ""):
+        return response
+    response = Response(
+        content=response.json(),
+    )
+    response.body = gzip.compress(response.body)
+    response.headers["Content-Encoding"] = "gzip"
+    response.headers["Content-Length"] = str(len(response.body))
+    return response