Server Usage

Run a Simple Web Server

In order to implement a web server, first create a request handler.

A request handler is a coroutine or regular function that accepts a Request instance as its only parameter and returns a Response instance:

from aiohttp import web

async def hello(request):
    return web.Response(text="Hello, world")

Next, create an Application instance and register the request handler with the application’s router on a particular HTTP method and path:

app = web.Application()
app.router.add_get('/', hello)

After that, run the application by run_app() call:


That’s it. Now, head over to http://localhost:8080/ to see the results.

See also

Graceful shutdown section explains what run_app() does and how to implement complex server initialization/finalization from scratch.

Command Line Interface (CLI)

aiohttp.web implements a basic CLI for quickly serving an Application in development over TCP/IP:

$ python -m aiohttp.web -H localhost -P 8080 package.module:init_func

package.module:init_func should be an importable callable that accepts a list of any non-parsed command-line arguments and returns an Application instance after setting it up:

def init_func(argv):
    app = web.Application()
    app.router.add_get("/", index_handler)
    return app


A request handler can be any callable that accepts a Request instance as its only argument and returns a StreamResponse derived (e.g. Response) instance:

def handler(request):
    return web.Response()

A handler may also be a coroutine, in which case aiohttp.web will await the handler:

async def handler(request):
    return web.Response()

Handlers are setup to handle requests by registering them with the Application.router on a particular route (HTTP method and path pair) using methods like UrlDispatcher.add_get and UrlDispatcher.add_post:

app.router.add_get('/', handler)
app.router.add_post('/post', post_handler)
app.router.add_put('/put', put_handler)

add_route() also supports the wildcard HTTP method, allowing a handler to serve incoming requests on a path having any HTTP method:

app.router.add_route('*', '/path', all_handler)

The HTTP method can be queried later in the request handler using the Request.method property.

By default endpoints added with add_get() will accept HEAD requests and return the same response headers as they would for a GET request. You can also deny HEAD requests on a route:

app.router.add_get('/', handler, allow_head=False)

Here handler won’t be called and the server will response with 405.


This is a change as of aiohttp v2.0 to act in accordance with RFC 7231.

Previous version always returned 405 for HEAD requests to routes added with add_get().

If you have handlers which perform lots of processing to write the response body you may wish to improve performance by skipping that processing in the case of HEAD requests while still taking care to respond with the same headers as with GET requests.

Resources and Routes

Internally router is a list of resources.

Resource is an entry in route table which corresponds to requested URL.

Resource in turn has at least one route.

Route corresponds to handling HTTP method by calling web handler.

UrlDispatcher.add_get() / UrlDispatcher.add_post() and family are plain shortcuts for UrlDispatcher.add_route().

UrlDispatcher.add_route() in turn is just a shortcut for pair of UrlDispatcher.add_resource() and Resource.add_route():

resource = app.router.add_resource(path, name=name)
route = resource.add_route(method, handler)
return route

See also

Router refactoring in 0.21 for more details

New in version 0.21.0: Introduce resources.

Variable Resources

Resource may have variable path also. For instance, a resource with the path '/a/{name}/c' would match all incoming requests with paths such as '/a/b/c', '/a/1/c', and '/a/etc/c'.

A variable part is specified in the form {identifier}, where the identifier can be used later in a request handler to access the matched value for that part. This is done by looking up the identifier in the Request.match_info mapping:

async def variable_handler(request):
    return web.Response(
        text="Hello, {}".format(request.match_info['name']))

resource = app.router.add_resource('/{name}')
resource.add_route('GET', variable_handler)

By default, each part matches the regular expression [^{}/]+.

You can also specify a custom regex in the form {identifier:regex}:

resource = app.router.add_resource(r'/{name:\d+}')


Regex should match against percent encoded URL (request.raw_path). E.g. space character is encoded as %20.

According to RFC 3986 allowed in path symbols are:

allowed       = unreserved / pct-encoded / sub-delims
              / ":" / "@" / "/"

pct-encoded   = "%" HEXDIG HEXDIG

unreserved    = ALPHA / DIGIT / "-" / "." / "_" / "~"

sub-delims    = "!" / "$" / "&" / "'" / "(" / ")"
              / "*" / "+" / "," / ";" / "="

Reverse URL Constructing using Named Resources

Routes can also be given a name:

resource = app.router.add_resource('/root', name='root')

Which can then be used to access and build a URL for that resource later (e.g. in a request handler):

>>>['root'].url_for().with_query({"a": "b", "c": "d"})

A more interesting example is building URLs for variable resources:

app.router.add_resource(r'/{user}/info', name='user-info')

In this case you can also pass in the parts of the route:

...                                         .with_query("a=b")

Organizing Handlers in Classes

As discussed above, handlers can be first-class functions or coroutines:

async def hello(request):
    return web.Response(text="Hello, world")

app.router.add_get('/', hello)

But sometimes it’s convenient to group logically similar handlers into a Python class.

Since aiohttp.web does not dictate any implementation details, application developers can organize handlers in classes if they so wish:

class Handler:

    def __init__(self):

    def handle_intro(self, request):
        return web.Response(text="Hello, world")

    async def handle_greeting(self, request):
        name = request.match_info.get('name', "Anonymous")
        txt = "Hello, {}".format(name)
        return web.Response(text=txt)

handler = Handler()
app.router.add_get('/intro', handler.handle_intro)
app.router.add_get('/greet/{name}', handler.handle_greeting)

Class Based Views

aiohttp.web has support for django-style class based views.

You can derive from View and define methods for handling http requests:

class MyView(web.View):
    async def get(self):
        return await get_resp(self.request)

    async def post(self):
        return await post_resp(self.request)

Handlers should be coroutines accepting self only and returning response object as regular web-handler. Request object can be retrieved by View.request property.

After implementing the view (MyView from example above) should be registered in application’s router:

app.router.add_route('*', '/path/to', MyView)

Example will process GET and POST requests for /path/to but raise 405 Method not allowed exception for unimplemented HTTP methods.

Resource Views

All registered resources in a router can be viewed using the UrlDispatcher.resources() method:

for resource in app.router.resources():

Similarly, a subset of the resources that were registered with a name can be viewed using the UrlDispatcher.named_resources() method:

for name, resource in app.router.named_resources().items():
    print(name, resource)

New in version 0.18: UrlDispatcher.routes()

New in version 0.19: UrlDispatcher.named_routes()

Alternative ways for registering routes

Code examples shown above use imperative style for adding new routes: they call app.router.add_get(...) etc.

There are two alternatives: route tables and route decorators.

Route tables look like Django way:

async def handle_get(request):

async def handle_post(request):

app.router.add_routes([web.get('/get', handle_get),
             '/post', handle_post),

The snippet calls add_routes() to register a list of route definitions (aiohttp.web.RouteDef instances) created by aiohttp.web.get() or functions.

See also

RouteDef reference.

Route decorators are closer to Flask approach:

routes = web.RouteTableDef()

async def handle_get(request):
async def handle_post(request):


The example creates a aiohttp.web.RouteTableDef container first.

The container is a list-like object with additional decorators aiohttp.web.RouteTableDef.get(), etc. for registering new routes.

After filling the container add_routes() is used for adding registered route definitions into application’s router.

See also

RouteTableDef reference.

All tree ways (imperative calls, route tables and decorators) are equivalent, you could use what do you prefer or even mix them on your own.

New in version 2.3.

Web Handler Cancellation


web-handler execution could be canceled on every await if client drops connection without reading entire response’s BODY.

The behavior is very different from classic WSGI frameworks like Flask and Django.

Sometimes it is a desirable behavior: on processing GET request the code might fetch data from database or other web resource, the fetching is potentially slow.

Canceling this fetch is very good: the peer dropped connection already, there is no reason to waste time and resources (memory etc) by getting data from DB without any chance to send it back to peer.

But sometimes the cancellation is bad: on POST request very often is needed to save data to DB regardless to peer closing.

Cancellation prevention could be implemented in several ways: * Applying asyncio.shield() to coroutine that saves data into DB. * Spawning a new task for DB saving * Using aiojobs or other third party library.

asyncio.shield() works pretty good. The only disadvantage is you need to split web handler into exactly two async functions: one for handler itself and other for protected code.

For example the following snippet is not safe:

async def handler(request):
    await asyncio.shield(write_to_redis(request))
    await asyncio.shield(write_to_postgres(request))
    return web.Response('OK')

Cancellation might be occurred just after saving data in REDIS, write_to_postgres will be not called.

Spawning a new task is much worse: there is no place to await spawned tasks:

async def handler(request):
    return web.Response('OK')

In this case errors from write_to_redis are not awaited, it leads to many asyncio log messages Future exception was never retrieved and Task was destroyed but it is pending!.

Moreover on Graceful shutdown phase aiohttp don’t wait for these tasks, you have a great chance to loose very important data.

On other hand aiojobs provides an API for spawning new jobs and awaiting their results etc. It stores all scheduled activity in internal data structures and could terminate them gracefully:

from aiojobs.aiohttp import setup, spawn

async def coro(timeout):
    await asyncio.sleep(timeout)  # do something in background

async def handler(request):
    await spawn(request, coro())
    return web.Response()

app = web.Application()
app.router.add_get('/', handler)

All not finished jobs will be terminated on aiohttp.web.Application.on_cleanup signal.

To prevent cancellation of the whole web-handler use @atomic decorator:

from aiojobs.aiohttp import atomic

async def handler(request):
    await write_to_db()
    return web.Response()

app = web.Application()
app.router.add_post('/', handler)

It prevents all handler async function from cancellation, write_to_db will be never interrupted.

Custom Routing Criteria

Sometimes you need to register handlers on more complex criteria than simply a HTTP method and path pair.

Although UrlDispatcher does not support any extra criteria, routing based on custom conditions can be accomplished by implementing a second layer of routing in your application.

The following example shows custom routing based on the HTTP Accept header:

class AcceptChooser:

    def __init__(self):
        self._accepts = {}

    async def do_route(self, request):
        for accept in request.headers.getall('ACCEPT', []):
            acceptor = self._accepts.get(accept)
            if acceptor is not None:
                return (await acceptor(request))
        raise HTTPNotAcceptable()

    def reg_acceptor(self, accept, handler):
        self._accepts[accept] = handler

async def handle_json(request):
    # do json handling

async def handle_xml(request):
    # do xml handling

chooser = AcceptChooser()
app.router.add_get('/', chooser.do_route)

chooser.reg_acceptor('application/json', handle_json)
chooser.reg_acceptor('application/xml', handle_xml)

Static file handling

The best way to handle static files (images, JavaScripts, CSS files etc.) is using Reverse Proxy like nginx or CDN services.

But for development it’s very convenient to handle static files by aiohttp server itself.

To do it just register a new static route by UrlDispatcher.add_static() call:

app.router.add_static('/prefix', path_to_static_folder)

When a directory is accessed within a static route then the server responses to client with HTTP/403 Forbidden by default. Displaying folder index instead could be enabled with show_index parameter set to True:

app.router.add_static('/prefix', path_to_static_folder, show_index=True)

When a symlink from the static directory is accessed, the server responses to client with HTTP/404 Not Found by default. To allow the server to follow symlinks, parameter follow_symlinks should be set to True:

app.router.add_static('/prefix', path_to_static_folder, follow_symlinks=True)

When you want to enable cache busting, parameter append_version can be set to True

Cache busting is the process of appending some form of file version hash to the filename of resources like JavaScript and CSS files. The performance advantage of doing this is that we can tell the browser to cache these files indefinitely without worrying about the client not getting the latest version when the file changes:

app.router.add_static('/prefix', path_to_static_folder, append_version=True)

Template Rendering

aiohttp.web does not support template rendering out-of-the-box.

However, there is a third-party library, aiohttp_jinja2, which is supported by the aiohttp authors.

Using it is rather simple. First, setup a jinja2 environment with a call to aiohttp_jinja2.setup():

app = web.Application()

After that you may use the template engine in your handlers. The most convenient way is to simply wrap your handlers with the aiohttp_jinja2.template() decorator:

def handler(request):
    return {'name': 'Andrew', 'surname': 'Svetlov'}

If you prefer the Mako template engine, please take a look at the aiohttp_mako library.

JSON Response

It is a common case to return JSON data in response, aiohttp.web provides a shortcut for returning JSON – aiohttp.web.json_response():

def handler(request):
    data = {'some': 'data'}
    return web.json_response(data)

The shortcut method returns aiohttp.web.Response instance so you can for example set cookies before returning it from handler.

User Sessions

Often you need a container for storing user data across requests. The concept is usually called a session.

aiohttp.web has no built-in concept of a session, however, there is a third-party library, aiohttp_session, that adds session support:

import asyncio
import time
import base64
from cryptography import fernet
from aiohttp import web
from aiohttp_session import setup, get_session, session_middleware
from aiohttp_session.cookie_storage import EncryptedCookieStorage

async def handler(request):
    session = await get_session(request)
    last_visit = session['last_visit'] if 'last_visit' in session else None
    text = 'Last visited: {}'.format(last_visit)
    return web.Response(text=text)

def make_app():
    app = web.Application()
    # secret_key must be 32 url-safe base64-encoded bytes
    fernet_key = fernet.Fernet.generate_key()
    secret_key = base64.urlsafe_b64decode(fernet_key)
    setup(app, EncryptedCookieStorage(secret_key))
    app.router.add_route('GET', '/', handler)
    return app


HTTP Forms

HTTP Forms are supported out of the box.

If form’s method is "GET" (<form method="get">) use Request.query for getting form data.

To access form data with "POST" method use or Request.multipart(). accepts both 'application/x-www-form-urlencoded' and 'multipart/form-data' form’s data encoding (e.g. <form enctype="multipart/form-data">). It stores files data in temporary directory. If client_max_size is specified post raises ValueError exception. For efficiency use Request.multipart(), It is especially effective for uploading large files (File Uploads).

Values submitted by the following form:

<form action="/login" method="post" accept-charset="utf-8"

    <label for="login">Login</label>
    <input id="login" name="login" type="text" value="" autofocus/>
    <label for="password">Password</label>
    <input id="password" name="password" type="password" value=""/>

    <input type="submit" value="login"/>

could be accessed as:

async def do_login(request):
    data = await
    login = data['login']
    password = data['password']

File Uploads

aiohttp.web has built-in support for handling files uploaded from the browser.

First, make sure that the HTML <form> element has its enctype attribute set to enctype="multipart/form-data". As an example, here is a form that accepts an MP3 file:

<form action="/store/mp3" method="post" accept-charset="utf-8"

    <label for="mp3">Mp3</label>
    <input id="mp3" name="mp3" type="file" value=""/>

    <input type="submit" value="submit"/>

Then, in the request handler you can access the file input field as a FileField instance. FileField is simply a container for the file as well as some of its metadata:

async def store_mp3_handler(request):

    # WARNING: don't do that if you plan to receive large files!
    data = await

    mp3 = data['mp3']

    # .filename contains the name of the file in string format.
    filename = mp3.filename

    # .file contains the actual file data that needs to be stored somewhere.
    mp3_file = data['mp3'].file

    content =

    return web.Response(body=content,
                            {'CONTENT-DISPOSITION': mp3_file}))

You might have noticed a big warning in the example above. The general issue is that reads the whole payload in memory, resulting in possible OOM errors. To avoid this, for multipart uploads, you should use Request.multipart() which returns a multipart reader:

async def store_mp3_handler(request):

    reader = await request.multipart()

    # /!\ Don't forget to validate your inputs /!\

    mp3 = await

    filename = mp3.filename

    # You cannot rely on Content-Length if transfer is chunked.
    size = 0
    with open(os.path.join('/spool/yarrr-media/mp3/', filename), 'wb') as f:
        while True:
            chunk = await mp3.read_chunk()  # 8192 bytes by default.
            if not chunk:
            size += len(chunk)

    return web.Response(text='{} sized of {} successfully stored'
                             ''.format(filename, size))


aiohttp.web supports WebSockets out-of-the-box.

To setup a WebSocket, create a WebSocketResponse in a request handler and then use it to communicate with the peer:

async def websocket_handler(request):

    ws = web.WebSocketResponse()
    await ws.prepare(request)

    async for msg in ws:
        if msg.type == aiohttp.WSMsgType.TEXT:
            if == 'close':
                await ws.close()
                await ws.send_str( + '/answer')
        elif msg.type == aiohttp.WSMsgType.ERROR:
            print('ws connection closed with exception %s' %

    print('websocket connection closed')

    return ws

The handler should be registered as HTTP GET processor:

app.router.add_get('/ws', websocket_handler)

Reading from the WebSocket (await ws.receive()) must only be done inside the request handler task; however, writing (ws.send_str(...)) to the WebSocket, closing (await ws.close()) and canceling the handler task may be delegated to other tasks. See also FAQ section.

aiohttp.web creates an implicit asyncio.Task for handling every incoming request.


While aiohttp.web itself only supports WebSockets without downgrading to LONG-POLLING, etc., our team supports SockJS, an aiohttp-based library for implementing SockJS-compatible server code.


Parallel reads from websocket are forbidden, there is no possibility to call aiohttp.web.WebSocketResponse.receive() from two tasks.

See FAQ section for instructions how to solve the problem.


aiohttp.web defines a set of exceptions for every HTTP status code.

Each exception is a subclass of HTTPException and relates to a single HTTP status code.

The exceptions are also a subclass of Response, allowing you to either raise or return them in a request handler for the same effect.

The following snippets are the same:

async def handler(request):
    return aiohttp.web.HTTPFound('/redirect')


async def handler(request):
    raise aiohttp.web.HTTPFound('/redirect')

Each exception class has a status code according to RFC 2068: codes with 100-300 are not really errors; 400s are client errors, and 500s are server errors.

HTTP Exception hierarchy chart:

      * 200 - HTTPOk
      * 201 - HTTPCreated
      * 202 - HTTPAccepted
      * 203 - HTTPNonAuthoritativeInformation
      * 204 - HTTPNoContent
      * 205 - HTTPResetContent
      * 206 - HTTPPartialContent
      * 300 - HTTPMultipleChoices
      * 301 - HTTPMovedPermanently
      * 302 - HTTPFound
      * 303 - HTTPSeeOther
      * 304 - HTTPNotModified
      * 305 - HTTPUseProxy
      * 307 - HTTPTemporaryRedirect
      * 308 - HTTPPermanentRedirect
        * 400 - HTTPBadRequest
        * 401 - HTTPUnauthorized
        * 402 - HTTPPaymentRequired
        * 403 - HTTPForbidden
        * 404 - HTTPNotFound
        * 405 - HTTPMethodNotAllowed
        * 406 - HTTPNotAcceptable
        * 407 - HTTPProxyAuthenticationRequired
        * 408 - HTTPRequestTimeout
        * 409 - HTTPConflict
        * 410 - HTTPGone
        * 411 - HTTPLengthRequired
        * 412 - HTTPPreconditionFailed
        * 413 - HTTPRequestEntityTooLarge
        * 414 - HTTPRequestURITooLong
        * 415 - HTTPUnsupportedMediaType
        * 416 - HTTPRequestRangeNotSatisfiable
        * 417 - HTTPExpectationFailed
        * 421 - HTTPMisdirectedRequest
        * 422 - HTTPUnprocessableEntity
        * 424 - HTTPFailedDependency
        * 426 - HTTPUpgradeRequired
        * 428 - HTTPPreconditionRequired
        * 429 - HTTPTooManyRequests
        * 431 - HTTPRequestHeaderFieldsTooLarge
        * 451 - HTTPUnavailableForLegalReasons
        * 500 - HTTPInternalServerError
        * 501 - HTTPNotImplemented
        * 502 - HTTPBadGateway
        * 503 - HTTPServiceUnavailable
        * 504 - HTTPGatewayTimeout
        * 505 - HTTPVersionNotSupported
        * 506 - HTTPVariantAlsoNegotiates
        * 507 - HTTPInsufficientStorage
        * 510 - HTTPNotExtended
        * 511 - HTTPNetworkAuthenticationRequired

All HTTP exceptions have the same constructor signature:

HTTPNotFound(*, headers=None, reason=None,
             body=None, text=None, content_type=None)

If not directly specified, headers will be added to the default response headers.

Classes HTTPMultipleChoices, HTTPMovedPermanently, HTTPFound, HTTPSeeOther, HTTPUseProxy, HTTPTemporaryRedirect have the following constructor signature:

HTTPFound(location, *, headers=None, reason=None,
          body=None, text=None, content_type=None)

where location is value for Location HTTP header.

HTTPMethodNotAllowed is constructed by providing the incoming unsupported method and list of allowed methods:

HTTPMethodNotAllowed(method, allowed_methods, *,
                     headers=None, reason=None,
                     body=None, text=None, content_type=None)

Data Sharing aka No Singletons Please

aiohttp.web discourages the use of global variables, aka singletons. Every variable should have its own context that is not global.

So, aiohttp.web.Application and aiohttp.web.Request support a interface (i.e. they are dict-like objects), allowing them to be used as data stores.

For storing global-like variables, feel free to save them in an Application instance:

app['my_private_key'] = data

and get it back in the web-handler:

async def handler(request):
    data =['my_private_key']

Variables that are only needed for the lifetime of a Request, can be stored in a Request:

async def handler(request):
  request['my_private_key'] = "data"

This is mostly useful for Middlewares and Signals handlers to store data for further processing by the next handlers in the chain.

To avoid clashing with other aiohttp users and third-party libraries, please choose a unique key name for storing data.

If your code is published on PyPI, then the project name is most likely unique and safe to use as the key. Otherwise, something based on your company name/url would be satisfactory (i.e.


aiohttp.web provides a powerful mechanism for customizing request handlers via middlewares.

Middlewares are setup by providing a sequence of middleware factories to the keyword-only middlewares parameter when creating an Application:

app = web.Application(middlewares=[middleware_factory_1,

A middleware factory is simply a coroutine that implements the logic of a middleware. For example, here’s a trivial middleware factory:

async def middleware_factory(app, handler):
    async def middleware_handler(request):
        return await handler(request)
    return middleware_handler

Every middleware factory should accept two parameters, an app instance and a handler, and return a new handler.

The handler passed in to a middleware factory is the handler returned by the next middleware factory. The last middleware factory always receives the request handler selected by the router itself (by UrlDispatcher.resolve()).


Both the outer middleware_factory coroutine and the inner middleware_handler coroutine are called for every request handled.

Middleware factories should return a new handler that has the same signature as a request handler. That is, it should accept a single Request instance and return a Response, or raise an exception.

Internally, a single request handler is constructed by applying the middleware chain to the original handler in reverse order, and is called by the RequestHandler as a regular handler.

Since middleware factories are themselves coroutines, they may perform extra await calls when creating a new handler, e.g. call database etc.

Middlewares usually call the inner handler, but they may choose to ignore it, e.g. displaying 403 Forbidden page or raising HTTPForbidden exception if user has no permissions to access the underlying resource. They may also render errors raised by the handler, perform some pre- or post-processing like handling CORS and so on.

The following code demonstrates middlewares execution order:

from aiohttp import web
def test(request):
    print('Handler function called')
    return web.Response(text="Hello")

async def middleware1(app, handler):
    async def middleware_handler(request):
        print('Middleware 1 called')
        response = await handler(request)
        print('Middleware 1 finished')

        return response
    return middleware_handler

async def middleware2(app, handler):
    async def middleware_handler(request):
        print('Middleware 2 called')
        response = await handler(request)
        print('Middleware 2 finished')

        return response
    return middleware_handler

app = web.Application(middlewares=[middleware1, middleware2])
app.router.add_get('/', test)

Produced output:

Middleware 1 called
Middleware 2 called
Handler function called
Middleware 2 finished
Middleware 1 finished


A common use of middlewares is to implement custom error pages. The following example will render 404 errors using a JSON response, as might be appropriate a JSON REST service:

import json
from aiohttp import web

def json_error(message):
    return web.Response(
        body=json.dumps({'error': message}).encode('utf-8'),

async def error_middleware(app, handler):
    async def middleware_handler(request):
            response = await handler(request)
            if response.status == 404:
                return json_error(response.message)
            return response
        except web.HTTPException as ex:
            if ex.status == 404:
                return json_error(ex.reason)
    return middleware_handler

app = web.Application(middlewares=[error_middleware])


New in version 0.18.

Although middlewares can customize request handlers before or after a Response has been prepared, they can’t customize a Response while it’s being prepared. For this aiohttp.web provides signals.

For example, a middleware can only change HTTP headers for unprepared responses (see prepare()), but sometimes we need a hook for changing HTTP headers for streamed responses and WebSockets. This can be accomplished by subscribing to the on_response_prepare signal:

async def on_prepare(request, response):
    response.headers['My-Header'] = 'value'


Additionally, the on_startup and on_cleanup signals can be subscribed to for application component setup and tear down accordingly.

The following example will properly initialize and dispose an aiopg connection engine:

from import create_engine

async def create_aiopg(app):
    app['pg_engine'] = await create_engine(

async def dispose_aiopg(app):
    await app['pg_engine'].wait_closed()


Signal handlers should not return a value but may modify incoming mutable parameters.

Signal handlers will be run sequentially, in order they were added. If handler is asynchronous, it will be awaited before calling next one.


Signals API has provisional status, meaning it may be changed in future releases.

Signal subscription and sending will most likely be the same, but signal object creation is subject to change. As long as you are not creating new signals, but simply reusing existing ones, you will not be affected.

Nested applications

Sub applications are designed for solving the problem of the big monolithic code base. Let’s assume we have a project with own business logic and tools like administration panel and debug toolbar.

Administration panel is a separate application by its own nature but all toolbar URLs are served by prefix like /admin.

Thus we’ll create a totally separate application named admin and connect it to main app with prefix by add_subapp():

admin = web.Application()
# setup admin routes, signals and middlewares

app.add_subapp('/admin/', admin)

Middlewares and signals from app and admin are chained.

It means that if URL is '/admin/something' middlewares from app are applied first and admin.middlewares are the next in the call chain.

The same is going for on_response_prepare signal – the signal is delivered to both top level app and admin if processing URL is routed to admin sub-application.

Common signals like on_startup, on_shutdown and on_cleanup are delivered to all registered sub-applications. The passed parameter is sub-application instance, not top-level application.

Third level sub-applications can be nested into second level ones – there are no limitation for nesting level.

Url reversing for sub-applications should generate urls with proper prefix.

But for getting URL sub-application’s router should be used:

admin = web.Application()
admin.router.add_get('/resource', handler, name='name')

app.add_subapp('/admin/', admin)

url = admin.router['name'].url_for()

The generated url from example will have a value URL('/admin/resource').

If main application should do URL reversing for sub-application it could use the following explicit technique:

admin = web.Application()
admin.router.add_get('/resource', handler, name='name')

app.add_subapp('/admin/', admin)
app['admin'] = admin

async def handler(request):  # main application's handler
    admin =['admin']
    url = admin.router['name'].url_for()

Flow control

aiohttp.web has sophisticated flow control for underlying TCP sockets write buffer.

The problem is: by default TCP sockets use Nagle’s algorithm for output buffer which is not optimal for streaming data protocols like HTTP.

Web server response may have one of the following states:

  1. CORK (tcp_cork is True). Don’t send out partial TCP/IP frames. All queued partial frames are sent when the option is cleared again. Optimal for sending big portion of data since data will be sent using minimum frames count.

    If OS does not support CORK mode (neither socket.TCP_CORK nor socket.TCP_NOPUSH exists) the mode is equal to Nagle’s enabled one. The most widespread OS without CORK support is Windows.

  2. NODELAY (tcp_nodelay is True). Disable the Nagle algorithm. This means that small data pieces are always sent as soon as possible, even if there is only a small amount of data. Optimal for transmitting short messages.

  3. Nagle’s algorithm enabled (both tcp_cork and tcp_nodelay are False). Data is buffered until there is a sufficient amount to send out. Avoid using this mode for sending HTTP data until you have no doubts.

By default streaming data (StreamResponse), regular responses (Response and http exceptions derived from it) and websockets (WebSocketResponse) use NODELAY mode, static file handlers work in CORK mode.

To manual mode switch set_tcp_cork() and set_tcp_nodelay() methods can be used. It may be helpful for better streaming control for example.

Expect Header

aiohttp.web supports Expect header. By default it sends HTTP/1.1 100 Continue line to client, or raises HTTPExpectationFailed if header value is not equal to “100-continue”. It is possible to specify custom Expect header handler on per route basis. This handler gets called if Expect header exist in request after receiving all headers and before processing application’s Middlewares and route handler. Handler can return None, in that case the request processing continues as usual. If handler returns an instance of class StreamResponse, request handler uses it as response. Also handler can raise a subclass of HTTPException. In this case all further processing will not happen and client will receive appropriate http response.


A server that does not understand or is unable to comply with any of the expectation values in the Expect field of a request MUST respond with appropriate error status. The server MUST respond with a 417 (Expectation Failed) status if any of the expectations cannot be met or, if there are other problems with the request, some other 4xx status.

If all checks pass, the custom handler must write a HTTP/1.1 100 Continue status code before returning.

The following example shows how to setup a custom handler for the Expect header:

async def check_auth(request):
    if request.version != aiohttp.HttpVersion11:

    if request.headers.get('EXPECT') != '100-continue':
        raise HTTPExpectationFailed(text="Unknown Expect: %s" % expect)

    if request.headers.get('AUTHORIZATION') is None:
        raise HTTPForbidden()

    request.transport.write(b"HTTP/1.1 100 Continue\r\n\r\n")

async def hello(request):
    return web.Response(body=b"Hello, world")

app = web.Application()
app.router.add_get('/', hello, expect_handler=check_auth)

Custom resource implementation

To register custom resource use UrlDispatcher.register_resource(). Resource instance must implement AbstractResource interface.

New in version 1.2.1.

Graceful shutdown

Stopping aiohttp web server by just closing all connections is not always satisfactory.

The problem is: if application supports websockets or data streaming it most likely has open connections at server shutdown time.

The library has no knowledge how to close them gracefully but developer can help by registering Application.on_shutdown signal handler and call the signal on web server closing.

Developer should keep a list of opened connections (Application is a good candidate).

The following websocket snippet shows an example for websocket handler:

app = web.Application()
app['websockets'] = []

async def websocket_handler(request):
    ws = web.WebSocketResponse()
    await ws.prepare(request)['websockets'].append(ws)
        async for msg in ws:

    return ws

Signal handler may look like:

async def on_shutdown(app):
    for ws in app['websockets']:
        await ws.close(code=WSCloseCode.GOING_AWAY,
                       message='Server shutdown')


Proper finalization procedure has three steps:

  1. Stop accepting new client connections by asyncio.Server.close() and asyncio.Server.wait_closed() calls.
  2. Fire Application.shutdown() event.
  3. Close accepted connections from clients by Server.shutdown() call with reasonable small delay.
  4. Call registered application finalizers by Application.cleanup().

The following code snippet performs proper application start, run and finalizing. It’s pretty close to run_app() utility function:

loop = asyncio.get_event_loop()
handler = app.make_handler()
f = loop.create_server(handler, '', 8080)
srv = loop.run_until_complete(f)
print('serving on', srv.sockets[0].getsockname())
except KeyboardInterrupt:

Background tasks

Sometimes there’s a need to perform some asynchronous operations just after application start-up.

Even more, in some sophisticated systems there could be a need to run some background tasks in the event loop along with the application’s request handler. Such as listening to message queue or other network message/event sources (e.g. ZeroMQ, Redis Pub/Sub, AMQP, etc.) to react to received messages within the application.

For example the background task could listen to ZeroMQ on zmq.SUB socket, process and forward retrieved messages to clients connected via WebSocket that are stored somewhere in the application (e.g. in the application['websockets'] list).

To run such short and long running background tasks aiohttp provides an ability to register Application.on_startup signal handler(s) that will run along with the application’s request handler.

For example there’s a need to run one quick task and two long running tasks that will live till the application is alive. The appropriate background tasks could be registered as an Application.on_startup signal handlers as shown in the example below:

async def listen_to_redis(app):
        sub = await aioredis.create_redis(('localhost', 6379), loop=app.loop)
        ch, *_ = await sub.subscribe('news')
        async for msg in ch.iter(encoding='utf-8'):
            # Forward message to all connected websockets:
            for ws in app['websockets']:
                ws.send_str('{}: {}'.format(, msg))
    except asyncio.CancelledError:
        await sub.unsubscribe(
        await sub.quit()

async def start_background_tasks(app):
    app['redis_listener'] = app.loop.create_task(listen_to_redis(app))

async def cleanup_background_tasks(app):
    await app['redis_listener']

app = web.Application()

The task listen_to_redis() will run forever. To shut it down correctly Application.on_cleanup signal handler may be used to send a cancellation to it.

Handling error pages

Pages like 404 Not Found and 500 Internal Error could be handled by custom middleware, see Middlewares for details.

Swagger support

aiohttp-swagger is a library that allow to add Swagger documentation and embed the Swagger-UI into your aiohttp.web project.

CORS support

aiohttp.web itself does not support Cross-Origin Resource Sharing, but there is an aiohttp plugin for it: aiohttp_cors.

Debug Toolbar

aiohttp_debugtoolbar is a very useful library that provides a debugging toolbar while you’re developing an aiohttp.web application.

Install it via pip:

$ pip install aiohttp_debugtoolbar

After that attach the aiohttp_debugtoolbar middleware to your aiohttp.web.Application and call aiohttp_debugtoolbar.setup():

import aiohttp_debugtoolbar
from aiohttp_debugtoolbar import toolbar_middleware_factory

app = web.Application(middlewares=[toolbar_middleware_factory])

The toolbar is ready to use. Enjoy!!!

Dev Tools

aiohttp-devtools provides a couple of tools to simplify development of aiohttp.web applications.

Install via pip:

  $ pip install aiohttp-devtools

 * ``runserver`` provides a development server with auto-reload,
live-reload, static file serving and aiohttp_debugtoolbar_
 * ``start`` is a `cookiecutter command which does the donkey work
of creating new :mod:`aiohttp.web` Applications.

Documentation and a complete tutorial of creating and running an app locally are available at aiohttp-devtools.