fix: make gateway load balancer should stream results (#6024)

Co-authored-by: Jina Dev Bot <[email protected]>
Co-authored-by: Joan Fontanals <[email protected]>

Author: 3 people

jina/serve/runtimes/gateway/request_handling.py

@@ -11,6 +11,7 @@
 
 if TYPE_CHECKING:  # pragma: no cover
     from types import SimpleNamespace
+
     import grpc
 
     from jina.logging.logger import JinaLogger
@@ -185,12 +186,23 @@ async def _load_balance(self, request):
             async with aiohttp.ClientSession() as session:
                 if request.method == 'GET':
                     async with session.get(target_url) as response:
-                        content = await response.read()
-                        return web.Response(
-                            body=content,
+                        # Create a StreamResponse with the same headers and status as the target response
+                        stream_response = web.StreamResponse(
                             status=response.status,
-                            content_type=response.content_type,
+                            headers=response.headers,
                         )
+
+                        # Prepare the response to send headers
+                        await stream_response.prepare(request)
+
+                        # Stream the response from the target server to the client
+                        async for chunk in response.content.iter_any():
+                            await stream_response.write(chunk)
+
+                        # Close the stream response once all chunks are sent
+                        await stream_response.write_eof()
+                        return stream_response
+
                 elif request.method == 'POST':
                     d = await request.read()
                     import json
@@ -282,7 +294,7 @@ async def stream(
             yield resp
 
     async def stream_doc(
-            self, request: SingleDocumentRequest, context: 'grpc.aio.ServicerContext'
+        self, request: SingleDocumentRequest, context: 'grpc.aio.ServicerContext'
     ) -> SingleDocumentRequest:
         """
         Process the received requests and return the result as a new request
@@ -293,7 +305,7 @@ async def stream_doc(
         """
         self.logger.debug('recv a stream_doc request')
         async for result in self.streamer.rpc_stream_doc(
-                request=request,
+            request=request,
         ):
             yield result
 
@@ -317,6 +329,7 @@ async def endpoint_discovery(self, empty, context) -> jina_pb2.EndpointsProto:
         :returns: the response request
         """
         from google.protobuf import json_format
+
         self.logger.debug('got an endpoint discovery request')
         response = jina_pb2.EndpointsProto()
         await self.streamer._get_endpoints_input_output_models(is_cancel=None)
@@ -332,7 +345,9 @@ async def endpoint_discovery(self, empty, context) -> jina_pb2.EndpointsProto:
             response.endpoints.extend(schema_maps.keys())
             json_format.ParseDict(schema_maps, response.schemas)
         else:
-            endpoints = await self.streamer.topology_graph._get_all_endpoints(self.streamer._connection_pool,  retry_forever=True, is_cancel=None)
+            endpoints = await self.streamer.topology_graph._get_all_endpoints(
+                self.streamer._connection_pool, retry_forever=True, is_cancel=None
+            )
             response.endpoints.extend(list(endpoints))
         return response
 

tests/integration/docarray_v2/test_streaming.py

@@ -1,9 +1,11 @@
+import asyncio
+import time
 from typing import AsyncGenerator, Generator, Optional
 
 import pytest
+from docarray import BaseDoc, DocList
 
-from jina import Client, Executor, requests, Flow, Deployment
-from docarray import DocList, BaseDoc
+from jina import Client, Deployment, Executor, Flow, requests
 from jina.helper import random_port
 
 
@@ -67,20 +69,21 @@ async def test_streaming_deployment(protocol, include_gateway):
             assert doc.text == f'hello world {i}'
             i += 1
 
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize('protocol', ['http', 'grpc'])
 async def test_streaming_flow(protocol):
     port = random_port()
 
     with Flow(protocol=protocol, port=port, cors=True).add(
-            uses=MyExecutor,
+        uses=MyExecutor,
     ):
         client = Client(port=port, protocol=protocol, asyncio=True)
         i = 10
         async for doc in client.stream_doc(
-                on='/hello',
-                inputs=MyDocument(text='hello world', number=i),
-                return_type=MyDocument,
+            on='/hello',
+            inputs=MyDocument(text='hello world', number=i),
+            return_type=MyDocument,
         ):
             assert doc.text == f'hello world {i}'
             i += 1
@@ -111,27 +114,66 @@ async def test_streaming_custom_response(protocol, endpoint, include_gateway):
             i += 1
 
 
+class WaitStreamExecutor(Executor):
+    @requests(on='/hello')
+    async def task(self, doc: MyDocument, **kwargs) -> MyDocument:
+        for i in range(5):
+            yield MyDocument(text=f'{doc.text} {doc.number + i}')
+            await asyncio.sleep(0.5)
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize('protocol', ['http', 'grpc'])
+@pytest.mark.parametrize('include_gateway', [False, True])
+async def test_streaming_delay(protocol, include_gateway):
+    from jina import Deployment
+
+    port = random_port()
+
+    with Deployment(
+        uses=WaitStreamExecutor,
+        timeout_ready=-1,
+        protocol=protocol,
+        port=port,
+        include_gateway=include_gateway,
+    ):
+        client = Client(port=port, protocol=protocol, asyncio=True)
+        i = 0
+        start_time = time.time()
+        async for doc in client.stream_doc(
+            on='/hello',
+            inputs=MyDocument(text='hello world', number=i),
+            return_type=MyDocument,
+        ):
+            assert doc.text == f'hello world {i}'
+            i += 1
+
+            # 0.5 seconds between each request + 0.5 seconds tolerance interval
+            assert time.time() - start_time < (0.5 * i) + 0.5
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize('protocol', ['http', 'grpc'])
 @pytest.mark.parametrize('endpoint', ['task1', 'task2', 'task3'])
 async def test_streaming_custom_response_flow_one_executor(protocol, endpoint):
     port = random_port()
 
     with Flow(
-            protocol=protocol,
-            cors=True,
-            port=port,
+        protocol=protocol,
+        cors=True,
+        port=port,
     ).add(uses=CustomResponseExecutor):
         client = Client(port=port, protocol=protocol, cors=True, asyncio=True)
         i = 0
         async for doc in client.stream_doc(
-                on=f'/{endpoint}',
-                inputs=MyDocument(text='hello world', number=5),
-                return_type=OutputDocument,
+            on=f'/{endpoint}',
+            inputs=MyDocument(text='hello world', number=5),
+            return_type=OutputDocument,
         ):
             assert doc.text == f'hello world 5-{i}-{endpoint}'
             i += 1
 
+
 class Executor1(Executor):
     @requests
     def generator(self, doc: MyDocument, **kwargs) -> MyDocument:

tests/integration/streaming/test_streaming.py

@@ -1,3 +1,6 @@
+import asyncio
+import time
+
 import pytest
 
 from jina import Client, Deployment, Executor, requests
@@ -29,11 +32,10 @@ async def test_streaming_deployment(protocol, include_gateway):
         uses=MyExecutor,
         timeout_ready=-1,
         protocol=protocol,
-        cors=True,
         port=port,
         include_gateway=include_gateway,
     ):
-        client = Client(port=port, protocol=protocol, cors=True, asyncio=True)
+        client = Client(port=port, protocol=protocol, asyncio=True)
         i = 0
         async for doc in client.stream_doc(
             on='/hello', inputs=Document(text='hello world')
@@ -42,6 +44,42 @@ async def test_streaming_deployment(protocol, include_gateway):
             i += 1
 
 
+class WaitStreamExecutor(Executor):
+    @requests(on='/hello')
+    async def task(self, doc: Document, **kwargs):
+        for i in range(5):
+            yield Document(text=f'{doc.text} {i}')
+            await asyncio.sleep(0.5)
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize('protocol', ['http', 'grpc'])
+@pytest.mark.parametrize('include_gateway', [False, True])
+async def test_streaming_delay(protocol, include_gateway):
+    from jina import Deployment
+
+    port = random_port()
+
+    with Deployment(
+        uses=WaitStreamExecutor,
+        timeout_ready=-1,
+        protocol=protocol,
+        port=port,
+        include_gateway=include_gateway,
+    ):
+        client = Client(port=port, protocol=protocol, asyncio=True)
+        i = 0
+        start_time = time.time()
+        async for doc in client.stream_doc(
+            on='/hello', inputs=Document(text='hello world')
+        ):
+            assert doc.text == f'hello world {i}'
+            i += 1
+
+            # 0.5 seconds between each request + 0.5 seconds tolerance interval
+            assert time.time() - start_time < (0.5 * i) + 0.5
+
+
 @pytest.mark.asyncio
 @pytest.mark.parametrize('protocol', ['grpc'])
 async def test_streaming_client_non_gen_endpoint(protocol):
@@ -53,11 +91,10 @@ async def test_streaming_client_non_gen_endpoint(protocol):
         uses=MyExecutor,
         timeout_ready=-1,
         protocol=protocol,
-        cors=True,
         port=port,
         include_gateway=False,
     ):
-        client = Client(port=port, protocol=protocol, cors=True, asyncio=True)
+        client = Client(port=port, protocol=protocol, asyncio=True)
         i = 0
         with pytest.raises(BadServer):
             async for _ in client.stream_doc(
@@ -67,7 +104,6 @@ async def test_streaming_client_non_gen_endpoint(protocol):
 
 
 def test_invalid_executor():
-
     with pytest.raises(RuntimeError) as exc_info:
 
         class InvalidExecutor3(Executor):