ASGI refactoring attempt

docs/sanic/deploying.md

@@ -1,7 +1,12 @@
 # Deploying
 
-Deploying Sanic is made simple by the inbuilt webserver. After defining an
-instance of `sanic.Sanic`, we can call the `run` method with the following
+Deploying Sanic is very simple using one of three options: the inbuilt webserver,
+an [ASGI webserver](https://asgi.readthedocs.io/en/latest/implementations.html), or `gunicorn`.
+It is also very common to place Sanic behind a reverse proxy, like `nginx`. 
+
+## Running via Sanic webserver
+
+After defining an instance of `sanic.Sanic`, we can call the `run` method with the following
 keyword arguments:
 
 - `host` *(default `"127.0.0.1"`)*: Address to host the server on.
@@ -17,7 +22,13 @@ keyword arguments:
   [asyncio.protocol](https://docs.python.org/3/library/asyncio-protocol.html#protocol-classes).
 - `access_log` *(default `True`)*: Enables log on handling requests (significantly slows server).
 
-## Workers
+```python
+app.run(host='0.0.0.0', port=1337, access_log=False)
+```
+
+In the above example, we decided to turn off the access log in order to increase performance.
+
+### Workers
 
 By default, Sanic listens in the main process using only one CPU core. To crank
 up the juice, just specify the number of workers in the `run` arguments.
@@ -29,9 +40,9 @@ app.run(host='0.0.0.0', port=1337, workers=4)
 Sanic will automatically spin up multiple processes and route traffic between
 them. We recommend as many workers as you have available cores.
 
-## Running via command
+### Running via command
 
-If you like using command line arguments, you can launch a Sanic server by
+If you like using command line arguments, you can launch a Sanic webserver by
 executing the module. For example, if you initialized Sanic as `app` in a file
 named `server.py`, you could run the server like so:
 
@@ -46,6 +57,33 @@ if __name__ == '__main__':
     app.run(host='0.0.0.0', port=1337, workers=4)
 ```
 
+## Running via ASGI
+
+Sanic is also ASGI-compliant. This means you can use your preferred ASGI webserver
+to run Sanic. The three main implementations of ASGI are
+[Daphne](http://github.com/django/daphne), [Uvicorn](https://www.uvicorn.org/),
+and [Hypercorn](https://pgjones.gitlab.io/hypercorn/index.html).
+
+Follow their documentation for the proper way to run them, but it should look
+something like:
+
+```
+daphne myapp:app
+uvicorn myapp:app
+hypercorn myapp:app
+```
+
+A couple things to note when using ASGI:
+
+1. When using the Sanic webserver, websockets will run using the [`websockets`](https://websockets.readthedocs.io/) package. In ASGI mode, there is no need for this package since websockets are managed in the ASGI server.
+1. The ASGI [lifespan protocol](https://asgi.readthedocs.io/en/latest/specs/lifespan.html) supports
+only two server events: startup and shutdown. Sanic has four: before startup, after startup, 
+before shutdown, and after shutdown. Therefore, in ASGI mode, the startup and shutdown events will 
+run consecutively and not actually around the server process beginning and ending (since that 
+is now controlled by the ASGI server). Therefore, it is best to use `after_server_start` and 
+`before_server_stop`.
+1. ASGI mode is still in "beta" as of Sanic v19.6.
+
 ## Running via Gunicorn
 
 [Gunicorn](http://gunicorn.org/) ‘Green Unicorn’ is a WSGI HTTP Server for UNIX.
@@ -64,7 +102,9 @@ of the memory leak.
 
 See the [Gunicorn Docs](http://docs.gunicorn.org/en/latest/settings.html#max-requests) for more information.
 
-## Running behind a reverse proxy
+## Other deployment considerations
+
+### Running behind a reverse proxy
 
 Sanic can be used with a reverse proxy (e.g. nginx). There's a simple example of nginx configuration:
 
@@ -84,7 +124,7 @@ server {
 
 If you want to get real client ip, you should configure `X-Real-IP` and `X-Forwarded-For` HTTP headers and set `app.config.PROXIES_COUNT` to `1`; see the configuration page for more information.
 
-## Disable debug logging
+### Disable debug logging for performance
 
 To improve the performance add `debug=False` and `access_log=False` in the `run` arguments.
 
@@ -104,9 +144,10 @@ Or you can rewrite app config directly
 app.config.ACCESS_LOG = False
 ```
 
-## Asynchronous support
-This is suitable if you *need* to share the sanic process with other applications, in particular the `loop`.
-However be advised that this method does not support using multiple processes, and is not the preferred way
+### Asynchronous support and sharing the loop
+
+This is suitable if you *need* to share the Sanic process with other applications, in particular the `loop`.
+However, be advised that this method does not support using multiple processes, and is not the preferred way
 to run the app in general.
 
 Here is an incomplete example (please see `run_async.py` in examples for something more practical):
@@ -116,4 +157,4 @@ server = app.create_server(host="0.0.0.0", port=8000, return_asyncio_server=True
 loop = asyncio.get_event_loop()
 task = asyncio.ensure_future(server)
 loop.run_forever()
-```
+```

examples/run_asgi.py

@@ -0,0 +1,88 @@
+"""
+1. Create a simple Sanic app
+0. Run with an ASGI server:
+    $ uvicorn run_asgi:app
+    or
+    $ hypercorn run_asgi:app
+"""
+
+from pathlib import Path
+from sanic import Sanic, response
+
+
+app = Sanic(__name__)
+
+
+@app.route("/text")
+def handler_text(request):
+    return response.text("Hello")
+
+
+@app.route("/json")
+def handler_json(request):
+    return response.json({"foo": "bar"})
+
+
+@app.websocket("/ws")
+async def handler_ws(request, ws):
+    name = "<someone>"
+    while True:
+        data = f"Hello {name}"
+        await ws.send(data)
+        name = await ws.recv()
+
+        if not name:
+            break
+
+
+@app.route("/file")
+async def handler_file(request):
+    return await response.file(Path("../") / "setup.py")
+
+
+@app.route("/file_stream")
+async def handler_file_stream(request):
+    return await response.file_stream(
+        Path("../") / "setup.py", chunk_size=1024
+    )
+
+
+@app.route("/stream", stream=True)
+async def handler_stream(request):
+    while True:
+        body = await request.stream.read()
+        if body is None:
+            break
+        body = body.decode("utf-8").replace("1", "A")
+        # await response.write(body)
+    return response.stream(body)
+
+
+@app.listener("before_server_start")
+async def listener_before_server_start(*args, **kwargs):
+    print("before_server_start")
+
+
+@app.listener("after_server_start")
+async def listener_after_server_start(*args, **kwargs):
+    print("after_server_start")
+
+
+@app.listener("before_server_stop")
+async def listener_before_server_stop(*args, **kwargs):
+    print("before_server_stop")
+
+
+@app.listener("after_server_stop")
+async def listener_after_server_stop(*args, **kwargs):
+    print("after_server_stop")
+
+
+@app.middleware("request")
+async def print_on_request(request):
+    print("print_on_request")
+
+
+@app.middleware("response")
+async def print_on_response(request, response):
+    print("print_on_response")

sanic/app.py

@@ -15,6 +15,7 @@
 from urllib.parse import urlencode, urlunparse
 
 from sanic import reloader_helpers
+from sanic.asgi import ASGIApp
 from sanic.blueprint_group import BlueprintGroup
 from sanic.config import BASE_LOGO, Config
 from sanic.constants import HTTP_METHODS
@@ -25,7 +26,7 @@
 from sanic.router import Router
 from sanic.server import HttpProtocol, Signal, serve, serve_multiple
 from sanic.static import register as static_register
-from sanic.testing import SanicTestClient
+from sanic.testing import SanicASGITestClient, SanicTestClient
 from sanic.views import CompositionView
 from sanic.websocket import ConnectionClosed, WebSocketProtocol
 
@@ -53,6 +54,7 @@ def __init__(
             logging.config.dictConfig(log_config or LOGGING_CONFIG_DEFAULTS)
 
         self.name = name
+        self.asgi = False
         self.router = router or Router()
         self.request_class = request_class
         self.error_handler = error_handler or ErrorHandler()
@@ -80,7 +82,7 @@ def loop(self):
 
         Only supported when using the `app.run` method.
         """
-        if not self.is_running:
+        if not self.is_running and self.asgi is False:
             raise SanicException(
                 "Loop can only be retrieved after the app has started "
                 "running. Not supported with `create_server` function"
@@ -469,13 +471,23 @@ async def websocket_handler(request, *args, **kwargs):
                         getattr(handler, "__blueprintname__", "")
                         + handler.__name__
                     )
-                try:
-                    protocol = request.transport.get_protocol()
-                except AttributeError:
-                    # On Python3.5 the Transport classes in asyncio do not
-                    # have a get_protocol() method as in uvloop
-                    protocol = request.transport._protocol
-                ws = await protocol.websocket_handshake(request, subprotocols)
+
+                    pass
+
+                if self.asgi:
+                    ws = request.transport.get_websocket_connection()
+                else:
+                    try:
+                        protocol = request.transport.get_protocol()
+                    except AttributeError:
+                        # On Python3.5 the Transport classes in asyncio do not
+                        # have a get_protocol() method as in uvloop
+                        protocol = request.transport._protocol
+                    protocol.app = self
+
+                    ws = await protocol.websocket_handshake(
+                        request, subprotocols
+                    )
 
                 # schedule the application handler
                 # its future is kept in self.websocket_tasks in case it
@@ -983,8 +995,16 @@ async def handle_request(self, request, write_callback, stream_callback):
                 raise CancelledError()
 
         # pass the response to the correct callback
-        if isinstance(response, StreamingHTTPResponse):
-            await stream_callback(response)
+        if write_callback is None or isinstance(
+            response, StreamingHTTPResponse
+        ):
+            if stream_callback:
+                await stream_callback(response)
+            else:
+                # Should only end here IF it is an ASGI websocket.
+                # TODO:
+                # - Add exception handling
+                pass
         else:
             write_callback(response)
 
@@ -996,6 +1016,10 @@ async def handle_request(self, request, write_callback, stream_callback):
     def test_client(self):
         return SanicTestClient(self)
 
+    @property
+    def asgi_client(self):
+        return SanicASGITestClient(self)
+
     # -------------------------------------------------------------------- #
     # Execution
     # -------------------------------------------------------------------- #
@@ -1122,10 +1146,6 @@ def stop(self):
         """This kills the Sanic"""
         get_event_loop().stop()
 
-    def __call__(self):
-        """gunicorn compatibility"""
-        return self
-
     async def create_server(
         self,
         host: Optional[str] = None,
@@ -1367,3 +1387,15 @@ def _helper(
     def _build_endpoint_name(self, *parts):
         parts = [self.name, *parts]
         return ".".join(parts)
+
+    # -------------------------------------------------------------------- #
+    # ASGI
+    # -------------------------------------------------------------------- #
+
+    async def __call__(self, scope, receive, send):
+        """To be ASGI compliant, our instance must be a callable that accepts
+        three arguments: scope, receive, send. See the ASGI reference for more
+        details: https://asgi.readthedocs.io/en/latest/"""
+        self.asgi = True
+        asgi_app = await ASGIApp.create(self, scope, receive, send)
+        await asgi_app()