feat: gzip compression to web request and response

Author: limcheekin

.github/workflows/dev.yml

@@ -43,4 +43,6 @@ jobs:
       - name: Echo Endpoint URL
         run: echo $(cat endpoint)
       - name: Test Lambda functions
-        run: "curl -X POST -H 'Content-Type: application/json' -d @text.json $(cat endpoint)v1/embeddings"
+        run: |
+          "curl -X POST -H 'Content-Type: application/json' -d @text.json $(cat endpoint)v1/embeddings"
+          "./test.sh $(cat endpoint)v1/embeddings"

open/text/embeddings/server/app.py

@@ -10,6 +10,8 @@
 import os
 import torch
 
+from open.text.embeddings.server.gzip import GZipRequestMiddleware
+from fastapi.middleware.gzip import GZipMiddleware
 router = APIRouter()
 
 DEFAULT_MODEL_NAME = "intfloat/e5-large-v2"
@@ -23,7 +25,7 @@ def create_app():
     initialize_embeddings()
     app = FastAPI(
         title="Open Text Embeddings API",
-        version="1.0.0",
+        version="1.0.2",
     )
     app.add_middleware(
         CORSMiddleware,
@@ -32,6 +34,12 @@ def create_app():
         allow_methods=["*"],
         allow_headers=["*"],
     )
+
+    app.add_middleware(GZipRequestMiddleware)
+
+    # handling gzip response only
+    app.add_middleware(GZipMiddleware, minimum_size=1000)
+
     app.include_router(router)
 
     return app

open/text/embeddings/server/gzip.py

@@ -0,0 +1,56 @@
+# Created by ChatGPT (GPT-3.5)
+# REF: https://chat.openai.com/share/4ab741e7-8059-4be1-b3da-46b8e78d98cc
+# REF: https://gealber.com/gzip-middleware-fastapi
+import gzip
+import json
+from starlette.types import Message
+from starlette.requests import Request
+from starlette.responses import JSONResponse
+from starlette.middleware.base import BaseHTTPMiddleware
+
+
+class GZipRequestMiddleware(BaseHTTPMiddleware):
+    async def set_body(self, request: Request):
+        receive_ = await request._receive()
+        content_encoding = request.headers.get('Content-Encoding', '').lower()
+        print("content_encoding", content_encoding)
+        if 'gzip' in content_encoding:
+            print("receive_", receive_)
+
+            try:
+                content_length = int(
+                    request.headers.get('Content-Length', '0'))
+                body = receive_.get('body')
+                if len(body) != content_length:
+                    return JSONResponse(
+                        content={"error": "Invalid Content-Length header"},
+                        status_code=400,
+                    )
+                json_byte_string = gzip.decompress(body)
+                # json_string = json_byte_string.decode('utf-8')
+                # json_object = json.dumps(json_string)
+                receive_['body'] = json_byte_string
+                print("content_length", content_length)
+                print("body", body)
+                print("receive_['body']", receive_['body'])
+            except ValueError:
+                return JSONResponse(
+                    content={"error": "Invalid Content-Length header"},
+                    status_code=400,
+                )
+            except Exception as e:
+                print(e)
+                return JSONResponse(
+                    content={"error": "Failed to decompress gzip content"},
+                    status_code=400,
+                )
+
+        async def receive() -> Message:
+            return receive_
+
+        request._receive = receive
+
+    async def dispatch(self, request, call_next):
+        await self.set_body(request)
+        response = await call_next(request)
+        return response

server-requirements.txt

@@ -1,5 +1,4 @@
 fastapi
-sse-starlette
 mangum
 sentence_transformers
 langchain

setup.py

@@ -11,7 +11,7 @@
     description="Open Source Text Embedding Models with OpenAI API-Compatible Endpoint",
     long_description=long_description,
     long_description_content_type="text/markdown",
-    version="1.0.1",
+    version="1.0.2",
     author="Lim Chee Kin",
     author_email="limcheekin@vobject.com",
     license="MIT",
@@ -23,7 +23,6 @@
         "server": ["uvicorn>=0.22.0",
                    "fastapi>=0.100.0",
                    "pydantic-settings>=2.0.1",
-                   "sse-starlette>=1.6.1",
                    "sentence_transformers>=2.2.2",
                    ],
     },

test.sh

@@ -0,0 +1 @@
+gzip -c text.json | curl -H "Content-Encoding: gzip" -H "Content-Type: application/json" -X POST --data-binary @- $1