WEBSOCKETS(1) websockets WEBSOCKETS(1)

websockets - websockets 12.0

licence version pyversions tests docs openssf

websockets is a library for building WebSocket servers and clients in Python with a focus on correctness, simplicity, robustness, and performance.

It supports several network I/O and control flow paradigms:

1.
The default implementation builds upon asyncio, Python's standard asynchronous I/O framework. It provides an elegant coroutine-based API. It's ideal for servers that handle many clients concurrently.
2.
The threading implementation is a good alternative for clients, especially if you aren't familiar with asyncio. It may also be used for servers that don't need to serve many clients.
3.
The Sans-I/O implementation is designed for integrating in third-party libraries, typically application servers, in addition being used internally by websockets.

Here's an echo server with the asyncio API:

#!/usr/bin/env python
import asyncio
from websockets.server import serve
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def main():
    async with serve(echo, "localhost", 8765):
        await asyncio.Future()  # run forever
asyncio.run(main())

Here's how a client sends and receives messages with the threading API:

#!/usr/bin/env python
import asyncio
from websockets.sync.client import connect
def hello():
    with connect("ws://localhost:8765") as websocket:
        websocket.send("Hello world!")
        message = websocket.recv()
        print(f"Received: {message}")
hello()

Don't worry about the opening and closing handshakes, pings and pongs, or any other behavior described in the WebSocket specification. websockets takes care of this under the hood so you can focus on your application!

Also, websockets provides an interactive client:

$ python -m websockets ws://localhost:8765/
Connected to ws://localhost:8765/.
> Hello world!
< Hello world!
Connection closed: 1000 (OK).

Do you like it? Let's dive in!

websockets requires Python ≥ 3.8.

For each minor version (3.x), only the latest bugfix or security release (3.x.y) is officially supported.

It doesn't have any dependencies.

Install websockets with:

$ pip install websockets

Wheels are available for all platforms.

Learn how to build an real-time web application with websockets.

In this tutorial, you're going to build a web-based Connect Four game.

The web removes the constraint of being in the same room for playing a game. Two players can connect over of the Internet, regardless of where they are, and play in their browsers.

When a player makes a move, it should be reflected immediately on both sides. This is difficult to implement over HTTP due to the request-response style of the protocol.

Indeed, there is no good way to be notified when the other player makes a move. Workarounds such as polling or long-polling introduce significant overhead.

Enter WebSocket.

The WebSocket protocol provides two-way communication between a browser and a server over a persistent connection. That's exactly what you need to exchange moves between players, via a server.

  • In this first part, you will create a server and connect one browser; you can play if you share the same browser.
  • In the second part, you will connect a second browser; you can play from different browsers on a local network.
  • In the third part, you will deploy the game to the web; you can play from any browser connected to the Internet.

This tutorial assumes basic knowledge of Python and JavaScript.

If you're comfortable with virtual environments, you can use one for this tutorial. Else, don't worry: websockets doesn't have any dependencies; it shouldn't create trouble in the default environment.

If you haven't installed websockets yet, do it now:

$ pip install websockets

Confirm that websockets is installed:

$ python -m websockets --version

If you installed another version, you should switch to the corresponding version of the documentation.

Create a directory and download these three files: connect4.js, connect4.css, and connect4.py.

The JavaScript module, along with the CSS file, provides a web-based user interface. Here's its API.

Color of the first player.
Color of the second player.
Draw a board.
•
board -- DOM element containing the board; must be initially empty.
Play a move.
  • board -- DOM element containing the board.
  • player -- PLAYER1 or PLAYER2.
  • column -- between 0 and 6.
  • row -- between 0 and 5.

The Python module provides a class to record moves and tell when a player wins. Here's its API.

Color of the first player.
Color of the second player.
A Connect Four game.
Play a move.
  • player -- PLAYER1 or PLAYER2.
  • column -- between 0 and 6.
Row where the checker lands, between 0 and 5.
RuntimeError -- if the move is illegal.
List of moves played during this game, as (player, column, row) tuples.
PLAYER1 or PLAYER2 if they won; None if the game is still ongoing.

Create an index.html file next to connect4.js and connect4.css with this content:

<!DOCTYPE html>
<html lang="en">
  <head>
    <title>Connect Four</title>
  </head>
  <body>
    <div class="board"></div>
    <script src="main.js" type="module"></script>
  </body>
</html>

This HTML page contains an empty <div> element where you will draw the Connect Four board. It loads a main.js script where you will write all your JavaScript code.

Create a main.js file next to index.html. In this script, when the page loads, draw the board:

import { createBoard, playMove } from "./connect4.js";
window.addEventListener("DOMContentLoaded", () => {
  // Initialize the UI.
  const board = document.querySelector(".board");
  createBoard(board);
});

Open a shell, navigate to the directory containing these files, and start an HTTP server:

$ python -m http.server

Open http://localhost:8000/ in a web browser. The page displays an empty board with seven columns and six rows. You will play moves in this board later.

Create an app.py file next to connect4.py with this content:

#!/usr/bin/env python
import asyncio
import websockets
async def handler(websocket):
    while True:
        message = await websocket.recv()
        print(message)
async def main():
    async with websockets.serve(handler, "", 8001):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

The entry point of this program is asyncio.run(main()). It creates an asyncio event loop, runs the main() coroutine, and shuts down the loop.

The main() coroutine calls serve() to start a websockets server. serve() takes three positional arguments:

  • handler is a coroutine that manages a connection. When a client connects, websockets calls handler with the connection in argument. When handler terminates, websockets closes the connection.
  • The second argument defines the network interfaces where the server can be reached. Here, the server listens on all interfaces, so that other devices on the same local network can connect.
  • The third argument is the port on which the server listens.

Invoking serve() as an asynchronous context manager, in an async with block, ensures that the server shuts down properly when terminating the program.

For each connection, the handler() coroutine runs an infinite loop that receives messages from the browser and prints them.

Open a shell, navigate to the directory containing app.py, and start the server:

$ python app.py

This doesn't display anything. Hopefully the WebSocket server is running. Let's make sure that it works. You cannot test the WebSocket server with a web browser like you tested the HTTP server. However, you can test it with websockets' interactive client.

Open another shell and run this command:

$ python -m websockets ws://localhost:8001/

You get a prompt. Type a message and press "Enter". Switch to the shell where the server is running and check that the server received the message. Good!

Exit the interactive client with Ctrl-C or Ctrl-D.

Now, if you look at the console where you started the server, you can see the stack trace of an exception:

connection handler failed
Traceback (most recent call last):
  ...
  File "app.py", line 22, in handler
    message = await websocket.recv()
  ...
websockets.exceptions.ConnectionClosedOK: received 1000 (OK); then sent 1000 (OK)

Indeed, the server was waiting for the next message with recv() when the client disconnected. When this happens, websockets raises a ConnectionClosedOK exception to let you know that you won't receive another message on this connection.

This exception creates noise in the server logs, making it more difficult to spot real errors when you add functionality to the server. Catch it in the handler() coroutine:

async def handler(websocket):
    while True:
        try:
            message = await websocket.recv()
        except websockets.ConnectionClosedOK:
            break
        print(message)

Stop the server with Ctrl-C and start it again:

$ python app.py

The WebSocket server loads the Python code in app.py then serves every WebSocket request with this version of the code. As a consequence, changes to app.py aren't visible until you restart the server.

This is unlike the HTTP server that you started earlier with python -m http.server. For every request, this HTTP server reads the target file and sends it. That's why changes are immediately visible.

It is possible to restart the WebSocket server automatically but this isn't necessary for this tutorial.

Try connecting and disconnecting the interactive client again. The ConnectionClosedOK exception doesn't appear anymore.

This pattern is so common that websockets provides a shortcut for iterating over messages received on the connection until the client disconnects:

async def handler(websocket):
    async for message in websocket:
        print(message)

Restart the server and check with the interactive client that its behavior didn't change.

At this point, you bootstrapped a web application and a WebSocket server. Let's connect them.

In JavaScript, you open a WebSocket connection as follows:

const websocket = new WebSocket("ws://localhost:8001/");

Before you exchange messages with the server, you need to decide their format. There is no universal convention for this.

Let's use JSON objects with a type key identifying the type of the event and the rest of the object containing properties of the event.

Here's an event describing a move in the middle slot of the board:

const event = {type: "play", column: 3};

Here's how to serialize this event to JSON and send it to the server:

websocket.send(JSON.stringify(event));

Now you have all the building blocks to send moves to the server.

Add this function to main.js:

function sendMoves(board, websocket) {
  // When clicking a column, send a "play" event for a move in that column.
  board.addEventListener("click", ({ target }) => {
    const column = target.dataset.column;
    // Ignore clicks outside a column.
    if (column === undefined) {
      return;
    }
    const event = {
      type: "play",
      column: parseInt(column, 10),
    };
    websocket.send(JSON.stringify(event));
  });
}

sendMoves() registers a listener for click events on the board. The listener figures out which column was clicked, builds a event of type "play", serializes it, and sends it to the server.

Modify the initialization to open the WebSocket connection and call the sendMoves() function:

window.addEventListener("DOMContentLoaded", () => {
  // Initialize the UI.
  const board = document.querySelector(".board");
  createBoard(board);
  // Open the WebSocket connection and register event handlers.
  const websocket = new WebSocket("ws://localhost:8001/");
  sendMoves(board, websocket);
});

Check that the HTTP server and the WebSocket server are still running. If you stopped them, here are the commands to start them again:

$ python -m http.server
$ python app.py

Refresh http://localhost:8000/ in your web browser. Click various columns in the board. The server receives messages with the expected column number.

There isn't any feedback in the board because you haven't implemented that yet. Let's do it.

In JavaScript, you receive WebSocket messages by listening to message events. Here's how to receive a message from the server and deserialize it from JSON:

websocket.addEventListener("message", ({ data }) => {
  const event = JSON.parse(data);
  // do something with event
});

You're going to need three types of messages from the server to the browser:

{type: "play", player: "red", column: 3, row: 0}
{type: "win", player: "red"}
{type: "error", message: "This slot is full."}

The JavaScript code receiving these messages will dispatch events depending on their type and take appropriate action. For example, it will react to an event of type "play" by displaying the move on the board with the playMove() function.

Add this function to main.js:

function showMessage(message) {
  window.setTimeout(() => window.alert(message), 50);
}
function receiveMoves(board, websocket) {
  websocket.addEventListener("message", ({ data }) => {
    const event = JSON.parse(data);
    switch (event.type) {
      case "play":
        // Update the UI with the move.
        playMove(board, event.player, event.column, event.row);
        break;
      case "win":
        showMessage(`Player ${event.player} wins!`);
        // No further messages are expected; close the WebSocket connection.
        websocket.close(1000);
        break;
      case "error":
        showMessage(event.message);
        break;
      default:
        throw new Error(`Unsupported event type: ${event.type}.`);
    }
  });
}

When playMove() modifies the state of the board, the browser renders changes asynchronously. Conversely, window.alert() runs synchronously and blocks rendering while the alert is visible.

If you called window.alert() immediately after playMove(), the browser could display the alert before rendering the move. You could get a "Player red wins!" alert without seeing red's last move.

We're using window.alert() for simplicity in this tutorial. A real application would display these messages in the user interface instead. It wouldn't be vulnerable to this problem.

Modify the initialization to call the receiveMoves() function:

window.addEventListener("DOMContentLoaded", () => {
  // Initialize the UI.
  const board = document.querySelector(".board");
  createBoard(board);
  // Open the WebSocket connection and register event handlers.
  const websocket = new WebSocket("ws://localhost:8001/");
  receiveMoves(board, websocket);
  sendMoves(board, websocket);
});

At this point, the user interface should receive events properly. Let's test it by modifying the server to send some events.

Sending an event from Python is quite similar to JavaScript:

event = {"type": "play", "player": "red", "column": 3, "row": 0}
await websocket.send(json.dumps(event))

Else, websockets raises TypeError: data is a dict-like object.

Modify the handler() coroutine in app.py as follows:

import json
from connect4 import PLAYER1, PLAYER2
async def handler(websocket):
    for player, column, row in [
        (PLAYER1, 3, 0),
        (PLAYER2, 3, 1),
        (PLAYER1, 4, 0),
        (PLAYER2, 4, 1),
        (PLAYER1, 2, 0),
        (PLAYER2, 1, 0),
        (PLAYER1, 5, 0),
    ]:
        event = {
            "type": "play",
            "player": player,
            "column": column,
            "row": row,
        }
        await websocket.send(json.dumps(event))
        await asyncio.sleep(0.5)
    event = {
        "type": "win",
        "player": PLAYER1,
    }
    await websocket.send(json.dumps(event))

Restart the WebSocket server and refresh http://localhost:8000/ in your web browser. Seven moves appear at 0.5 second intervals. Then an alert announces the winner.

Good! Now you know how to communicate both ways.

Once you plug the game engine to process moves, you will have a fully functional game.

Add the game logic

In the handler() coroutine, you're going to initialize a game:

from connect4 import Connect4
async def handler(websocket):
    # Initialize a Connect Four game.
    game = Connect4()
    ...

Then, you're going to iterate over incoming messages and take these steps:

  • parse an event of type "play", the only type of event that the user interface sends;
  • play the move in the board with the play() method, alternating between the two players;
  • if play() raises RuntimeError because the move is illegal, send an event of type "error";
  • else, send an event of type "play" to tell the user interface where the checker lands;
  • if the move won the game, send an event of type "win".

Try to implement this by yourself!

Keep in mind that you must restart the WebSocket server and reload the page in the browser when you make changes.

When it works, you can play the game from a single browser, with players taking alternate turns.

Here's how to enable debug logs:

import logging
logging.basicConfig(format="%(message)s", level=logging.DEBUG)

If you're stuck, a solution is available at the bottom of this document.

In this first part of the tutorial, you learned how to:

  • build and run a WebSocket server in Python with serve();
  • receive a message in a connection handler with recv();
  • send a message in a connection handler with send();
  • iterate over incoming messages with async for message in websocket: ...;
  • open a WebSocket connection in JavaScript with the WebSocket API;
  • send messages in a browser with WebSocket.send();
  • receive messages in a browser by listening to message events;
  • design a set of events to be exchanged between the browser and the server.

You can now play a Connect Four game in a browser, communicating over a WebSocket connection with a server where the game logic resides!

However, the two players share a browser, so the constraint of being in the same room still applies.

Move on to the second part of the tutorial to break this constraint and play from separate browsers.

app.py

#!/usr/bin/env python
import asyncio
import itertools
import json
import websockets
from connect4 import PLAYER1, PLAYER2, Connect4
async def handler(websocket):
    # Initialize a Connect Four game.
    game = Connect4()
    # Players take alternate turns, using the same browser.
    turns = itertools.cycle([PLAYER1, PLAYER2])
    player = next(turns)
    async for message in websocket:
        # Parse a "play" event from the UI.
        event = json.loads(message)
        assert event["type"] == "play"
        column = event["column"]
        try:
            # Play the move.
            row = game.play(player, column)
        except RuntimeError as exc:
            # Send an "error" event if the move was illegal.
            event = {
                "type": "error",
                "message": str(exc),
            }
            await websocket.send(json.dumps(event))
            continue
        # Send a "play" event to update the UI.
        event = {
            "type": "play",
            "player": player,
            "column": column,
            "row": row,
        }
        await websocket.send(json.dumps(event))
        # If move is winning, send a "win" event.
        if game.winner is not None:
            event = {
                "type": "win",
                "player": game.winner,
            }
            await websocket.send(json.dumps(event))
        # Alternate turns.
        player = next(turns)
async def main():
    async with websockets.serve(handler, "", 8001):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

index.html

<!DOCTYPE html>
<html lang="en">
  <head>
    <title>Connect Four</title>
  </head>
  <body>
    <div class="board"></div>
    <script src="main.js" type="module"></script>
  </body>
</html>

main.js

import { createBoard, playMove } from "./connect4.js";
function showMessage(message) {
  window.setTimeout(() => window.alert(message), 50);
}
function receiveMoves(board, websocket) {
  websocket.addEventListener("message", ({ data }) => {
    const event = JSON.parse(data);
    switch (event.type) {
      case "play":
        // Update the UI with the move.
        playMove(board, event.player, event.column, event.row);
        break;
      case "win":
        showMessage(`Player ${event.player} wins!`);
        // No further messages are expected; close the WebSocket connection.
        websocket.close(1000);
        break;
      case "error":
        showMessage(event.message);
        break;
      default:
        throw new Error(`Unsupported event type: ${event.type}.`);
    }
  });
}
function sendMoves(board, websocket) {
  // When clicking a column, send a "play" event for a move in that column.
  board.addEventListener("click", ({ target }) => {
    const column = target.dataset.column;
    // Ignore clicks outside a column.
    if (column === undefined) {
      return;
    }
    const event = {
      type: "play",
      column: parseInt(column, 10),
    };
    websocket.send(JSON.stringify(event));
  });
}
window.addEventListener("DOMContentLoaded", () => {
  // Initialize the UI.
  const board = document.querySelector(".board");
  createBoard(board);
  // Open the WebSocket connection and register event handlers.
  const websocket = new WebSocket("ws://localhost:8001/");
  receiveMoves(board, websocket);
  sendMoves(board, websocket);
});

  • In the first part, you created a server and connected one browser; you could play if you shared the same browser.
  • In this second part, you will connect a second browser; you can play from different browsers on a local network.
  • In the third part, you will deploy the game to the web; you can play from any browser connected to the Internet.

In the first part of the tutorial, you opened a WebSocket connection from a browser to a server and exchanged events to play moves. The state of the game was stored in an instance of the Connect4 class, referenced as a local variable in the connection handler coroutine.

Now you want to open two WebSocket connections from two separate browsers, one for each player, to the same server in order to play the same game. This requires moving the state of the game to a place where both connections can access it.

As long as you're running a single server process, you can share state by storing it in a global variable.

In that case, you must design a way for the process that handles a given connection to be aware of relevant events for that client. This is often achieved with a publish / subscribe mechanism.

How can you make two connection handlers agree on which game they're playing? When the first player starts a game, you give it an identifier. Then, you communicate the identifier to the second player. When the second player joins the game, you look it up with the identifier.

In addition to the game itself, you need to keep track of the WebSocket connections of the two players. Since both players receive the same events, you don't need to treat the two connections differently; you can store both in the same set.

Let's sketch this in code.

A module-level dict enables lookups by identifier:

JOIN = {}

When the first player starts the game, initialize and store it:

import secrets
async def handler(websocket):
    ...
    # Initialize a Connect Four game, the set of WebSocket connections
    # receiving moves from this game, and secret access token.
    game = Connect4()
    connected = {websocket}
    join_key = secrets.token_urlsafe(12)
    JOIN[join_key] = game, connected
    try:
        ...
    finally:
        del JOIN[join_key]

When the second player joins the game, look it up:

async def handler(websocket):
    ...
    join_key = ...  # TODO
    # Find the Connect Four game.
    game, connected = JOIN[join_key]
    # Register to receive moves from this game.
    connected.add(websocket)
    try:
        ...
    finally:
        connected.remove(websocket)

Notice how we're carefully cleaning up global state with try: ... finally: ... blocks. Else, we could leave references to games or connections in global state, which would cause a memory leak.

In both connection handlers, you have a game pointing to the same Connect4 instance, so you can interact with the game, and a connected set of connections, so you can send game events to both players as follows:

async def handler(websocket):
    ...
    for connection in connected:
        await connection.send(json.dumps(event))
    ...

Perhaps you spotted a major piece missing from the puzzle. How does the second player obtain join_key? Let's design new events to carry this information.

To start a game, the first player sends an "init" event:

{type: "init"}

The connection handler for the first player creates a game as shown above and responds with:

{type: "init", join: "<join_key>"}

With this information, the user interface of the first player can create a link to http://localhost:8000/?join=<join_key>. For the sake of simplicity, we will assume that the first player shares this link with the second player outside of the application, for example via an instant messaging service.

To join the game, the second player sends a different "init" event:

{type: "init", join: "<join_key>"}

The connection handler for the second player can look up the game with the join key as shown above. There is no need to respond.

Let's dive into the details of implementing this design.

We'll start with the initialization sequence for the first player.

In main.js, define a function to send an initialization event when the WebSocket connection is established, which triggers an open event:

function initGame(websocket) {
  websocket.addEventListener("open", () => {
    // Send an "init" event for the first player.
    const event = { type: "init" };
    websocket.send(JSON.stringify(event));
  });
}

Update the initialization sequence to call initGame():

window.addEventListener("DOMContentLoaded", () => {
  // Initialize the UI.
  const board = document.querySelector(".board");
  createBoard(board);
  // Open the WebSocket connection and register event handlers.
  const websocket = new WebSocket("ws://localhost:8001/");
  initGame(websocket);
  receiveMoves(board, websocket);
  sendMoves(board, websocket);
});

In app.py, define a new handler coroutine — keep a copy of the previous one to reuse it later:

import secrets
JOIN = {}
async def start(websocket):
    # Initialize a Connect Four game, the set of WebSocket connections
    # receiving moves from this game, and secret access token.
    game = Connect4()
    connected = {websocket}
    join_key = secrets.token_urlsafe(12)
    JOIN[join_key] = game, connected
    try:
        # Send the secret access token to the browser of the first player,
        # where it'll be used for building a "join" link.
        event = {
            "type": "init",
            "join": join_key,
        }
        await websocket.send(json.dumps(event))
        # Temporary - for testing.
        print("first player started game", id(game))
        async for message in websocket:
            print("first player sent", message)
    finally:
        del JOIN[join_key]
async def handler(websocket):
    # Receive and parse the "init" event from the UI.
    message = await websocket.recv()
    event = json.loads(message)
    assert event["type"] == "init"
    # First player starts a new game.
    await start(websocket)

In index.html, add an <a> element to display the link to share with the other player.

<body>
  <div class="actions">
    <a class="action join" href="">Join</a>
  </div>
  <!-- ... -->
</body>

In main.js, modify receiveMoves() to handle the "init" message and set the target of that link:

switch (event.type) {
  case "init":
    // Create link for inviting the second player.
    document.querySelector(".join").href = "?join=" + event.join;
    break;
  // ...
}

Restart the WebSocket server and reload http://localhost:8000/ in the browser. There's a link labeled JOIN below the board with a target that looks like http://localhost:8000/?join=95ftAaU5DJVP1zvb.

The server logs say first player started game .... If you click the board, you see "play" events. There is no feedback in the UI, though, because you haven't restored the game logic yet.

Before we get there, let's handle links with a join query parameter.

We'll now update the initialization sequence to account for the second player.

In main.js, update initGame() to send the join key in the "init" message when it's in the URL:

function initGame(websocket) {
  websocket.addEventListener("open", () => {
    // Send an "init" event according to who is connecting.
    const params = new URLSearchParams(window.location.search);
    let event = { type: "init" };
    if (params.has("join")) {
      // Second player joins an existing game.
      event.join = params.get("join");
    } else {
      // First player starts a new game.
    }
    websocket.send(JSON.stringify(event));
  });
}

In app.py, update the handler coroutine to look for the join key in the "init" message, then load that game:

async def error(websocket, message):
    event = {
        "type": "error",
        "message": message,
    }
    await websocket.send(json.dumps(event))
async def join(websocket, join_key):
    # Find the Connect Four game.
    try:
        game, connected = JOIN[join_key]
    except KeyError:
        await error(websocket, "Game not found.")
        return
    # Register to receive moves from this game.
    connected.add(websocket)
    try:
        # Temporary - for testing.
        print("second player joined game", id(game))
        async for message in websocket:
            print("second player sent", message)
    finally:
        connected.remove(websocket)
async def handler(websocket):
    # Receive and parse the "init" event from the UI.
    message = await websocket.recv()
    event = json.loads(message)
    assert event["type"] == "init"
    if "join" in event:
        # Second player joins an existing game.
        await join(websocket, event["join"])
    else:
        # First player starts a new game.
        await start(websocket)

Restart the WebSocket server and reload http://localhost:8000/ in the browser.

Copy the link labeled JOIN and open it in another browser. You may also open it in another tab or another window of the same browser; however, that makes it a bit tricky to remember which one is the first or second player.

Since games are stored in the memory of the Python process, they're lost when you stop the server.

Whenever you make changes to app.py, you must restart the server, create a new game in a browser, and join it in another browser.

The server logs say first player started game ... and second player joined game .... The numbers match, proving that the game local variable in both connection handlers points to same object in the memory of the Python process.

Click the board in either browser. The server receives "play" events from the corresponding player.

In the initialization sequence, you're routing connections to start() or join() depending on the first message received by the server. This is a common pattern in servers that handle different clients.

Instead of sending an initialization event, you could encode the join key in the WebSocket URI e.g. ws://localhost:8001/join/<join_key>. The WebSocket server would parse websocket.path and route the connection, similar to how HTTP servers route requests.

When you need to send sensitive data like authentication credentials to the server, sending it an event is considered more secure than encoding it in the URI because URIs end up in logs.

For the purposes of this tutorial, both approaches are equivalent because the join key comes from an HTTP URL. There isn't much at risk anyway!

Now you can restore the logic for playing moves and you'll have a fully functional two-player game.

Add the game logic

Once the initialization is done, the game is symmetrical, so you can write a single coroutine to process the moves of both players:

async def play(websocket, game, player, connected):
    ...

With such a coroutine, you can replace the temporary code for testing in start() by:

await play(websocket, game, PLAYER1, connected)

and in join() by:

await play(websocket, game, PLAYER2, connected)

The play() coroutine will reuse much of the code you wrote in the first part of the tutorial.

Try to implement this by yourself!

Keep in mind that you must restart the WebSocket server, reload the page to start a new game with the first player, copy the JOIN link, and join the game with the second player when you make changes.

When play() works, you can play the game from two separate browsers, possibly running on separate computers on the same local network.

A complete solution is available at the bottom of this document.

Let's add one more feature: allow spectators to watch the game.

The process for inviting a spectator can be the same as for inviting the second player. You will have to duplicate all the initialization logic:

  • declare a WATCH global variable similar to JOIN;
  • generate a watch key when creating a game; it must be different from the join key, or else a spectator could hijack a game by tweaking the URL;
  • include the watch key in the "init" event sent to the first player;
  • generate a WATCH link in the UI with a watch query parameter;
  • update the initGame() function to handle such links;
  • update the handler() coroutine to invoke a watch() coroutine for spectators;
  • prevent sendMoves() from sending "play" events for spectators.

Once the initialization sequence is done, watching a game is as simple as registering the WebSocket connection in the connected set in order to receive game events and doing nothing until the spectator disconnects. You can wait for a connection to terminate with wait_closed():

async def watch(websocket, watch_key):
    ...
    connected.add(websocket)
    try:
        await websocket.wait_closed()
    finally:
        connected.remove(websocket)

The connection can terminate because the receiveMoves() function closed it explicitly after receiving a "win" event, because the spectator closed their browser, or because the network failed.

Again, try to implement this by yourself.

When watch() works, you can invite spectators to watch the game from other browsers, as long as they're on the same local network.

As a further improvement, you may support adding spectators while a game is already in progress. This requires replaying moves that were played before the spectator was added to the connected set. Past moves are available in the moves attribute of the game.

This feature is included in the solution proposed below.

When you need to send a message to the two players and to all spectators, you're using this pattern:

async def handler(websocket):
    ...
    for connection in connected:
        await connection.send(json.dumps(event))
    ...

Since this is a very common pattern in WebSocket servers, websockets provides the broadcast() helper for this purpose:

async def handler(websocket):
    ...
    websockets.broadcast(connected, json.dumps(event))
    ...

Calling broadcast() once is more efficient than calling send() in a loop.

However, there's a subtle difference in behavior. Did you notice that there's no await in the second version? Indeed, broadcast() is a function, not a coroutine like send() or recv().

It's quite obvious why recv() is a coroutine. When you want to receive the next message, you have to wait until the client sends it and the network transmits it.

It's less obvious why send() is a coroutine. If you send many messages or large messages, you could write data faster than the network can transmit it or the client can read it. Then, outgoing data will pile up in buffers, which will consume memory and may crash your application.

To avoid this problem, send() waits until the write buffer drains. By slowing down the application as necessary, this ensures that the server doesn't send data too quickly. This is called backpressure and it's useful for building robust systems.

That said, when you're sending the same messages to many clients in a loop, applying backpressure in this way can become counterproductive. When you're broadcasting, you don't want to slow down everyone to the pace of the slowest clients; you want to drop clients that cannot keep up with the data stream. That's why broadcast() doesn't wait until write buffers drain.

For our Connect Four game, there's no difference in practice: the total amount of data sent on a connection for a game of Connect Four is less than 64 KB, so the write buffer never fills up and backpressure never kicks in anyway.

In this second part of the tutorial, you learned how to:

  • configure a connection by exchanging initialization messages;
  • keep track of connections within a single server process;
  • wait until a client disconnects in a connection handler;
  • broadcast a message to many connections efficiently.

You can now play a Connect Four game from separate browser, communicating over WebSocket connections with a server that synchronizes the game logic!

However, the two players have to be on the same local network as the server, so the constraint of being in the same place still mostly applies.

Head over to the third part of the tutorial to deploy the game to the web and remove this constraint.

app.py

#!/usr/bin/env python
import asyncio
import json
import secrets
import websockets
from connect4 import PLAYER1, PLAYER2, Connect4
JOIN = {}
WATCH = {}
async def error(websocket, message):
    """
    Send an error message.
    """
    event = {
        "type": "error",
        "message": message,
    }
    await websocket.send(json.dumps(event))
async def replay(websocket, game):
    """
    Send previous moves.
    """
    # Make a copy to avoid an exception if game.moves changes while iteration
    # is in progress. If a move is played while replay is running, moves will
    # be sent out of order but each move will be sent once and eventually the
    # UI will be consistent.
    for player, column, row in game.moves.copy():
        event = {
            "type": "play",
            "player": player,
            "column": column,
            "row": row,
        }
        await websocket.send(json.dumps(event))
async def play(websocket, game, player, connected):
    """
    Receive and process moves from a player.
    """
    async for message in websocket:
        # Parse a "play" event from the UI.
        event = json.loads(message)
        assert event["type"] == "play"
        column = event["column"]
        try:
            # Play the move.
            row = game.play(player, column)
        except RuntimeError as exc:
            # Send an "error" event if the move was illegal.
            await error(websocket, str(exc))
            continue
        # Send a "play" event to update the UI.
        event = {
            "type": "play",
            "player": player,
            "column": column,
            "row": row,
        }
        websockets.broadcast(connected, json.dumps(event))
        # If move is winning, send a "win" event.
        if game.winner is not None:
            event = {
                "type": "win",
                "player": game.winner,
            }
            websockets.broadcast(connected, json.dumps(event))
async def start(websocket):
    """
    Handle a connection from the first player: start a new game.
    """
    # Initialize a Connect Four game, the set of WebSocket connections
    # receiving moves from this game, and secret access tokens.
    game = Connect4()
    connected = {websocket}
    join_key = secrets.token_urlsafe(12)
    JOIN[join_key] = game, connected
    watch_key = secrets.token_urlsafe(12)
    WATCH[watch_key] = game, connected
    try:
        # Send the secret access tokens to the browser of the first player,
        # where they'll be used for building "join" and "watch" links.
        event = {
            "type": "init",
            "join": join_key,
            "watch": watch_key,
        }
        await websocket.send(json.dumps(event))
        # Receive and process moves from the first player.
        await play(websocket, game, PLAYER1, connected)
    finally:
        del JOIN[join_key]
        del WATCH[watch_key]
async def join(websocket, join_key):
    """
    Handle a connection from the second player: join an existing game.
    """
    # Find the Connect Four game.
    try:
        game, connected = JOIN[join_key]
    except KeyError:
        await error(websocket, "Game not found.")
        return
    # Register to receive moves from this game.
    connected.add(websocket)
    try:
        # Send the first move, in case the first player already played it.
        await replay(websocket, game)
        # Receive and process moves from the second player.
        await play(websocket, game, PLAYER2, connected)
    finally:
        connected.remove(websocket)
async def watch(websocket, watch_key):
    """
    Handle a connection from a spectator: watch an existing game.
    """
    # Find the Connect Four game.
    try:
        game, connected = WATCH[watch_key]
    except KeyError:
        await error(websocket, "Game not found.")
        return
    # Register to receive moves from this game.
    connected.add(websocket)
    try:
        # Send previous moves, in case the game already started.
        await replay(websocket, game)
        # Keep the connection open, but don't receive any messages.
        await websocket.wait_closed()
    finally:
        connected.remove(websocket)
async def handler(websocket):
    """
    Handle a connection and dispatch it according to who is connecting.
    """
    # Receive and parse the "init" event from the UI.
    message = await websocket.recv()
    event = json.loads(message)
    assert event["type"] == "init"
    if "join" in event:
        # Second player joins an existing game.
        await join(websocket, event["join"])
    elif "watch" in event:
        # Spectator watches an existing game.
        await watch(websocket, event["watch"])
    else:
        # First player starts a new game.
        await start(websocket)
async def main():
    async with websockets.serve(handler, "", 8001):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

index.html

<!DOCTYPE html>
<html lang="en">
  <head>
    <title>Connect Four</title>
  </head>
  <body>
    <div class="actions">
      <a class="action new" href="/">New</a>
      <a class="action join" href="">Join</a>
      <a class="action watch" href="">Watch</a>
    </div>
    <div class="board"></div>
    <script src="main.js" type="module"></script>
  </body>
</html>

main.js

import { createBoard, playMove } from "./connect4.js";
function initGame(websocket) {
  websocket.addEventListener("open", () => {
    // Send an "init" event according to who is connecting.
    const params = new URLSearchParams(window.location.search);
    let event = { type: "init" };
    if (params.has("join")) {
      // Second player joins an existing game.
      event.join = params.get("join");
    } else if (params.has("watch")) {
      // Spectator watches an existing game.
      event.watch = params.get("watch");
    } else {
      // First player starts a new game.
    }
    websocket.send(JSON.stringify(event));
  });
}
function showMessage(message) {
  window.setTimeout(() => window.alert(message), 50);
}
function receiveMoves(board, websocket) {
  websocket.addEventListener("message", ({ data }) => {
    const event = JSON.parse(data);
    switch (event.type) {
      case "init":
        // Create links for inviting the second player and spectators.
        document.querySelector(".join").href = "?join=" + event.join;
        document.querySelector(".watch").href = "?watch=" + event.watch;
        break;
      case "play":
        // Update the UI with the move.
        playMove(board, event.player, event.column, event.row);
        break;
      case "win":
        showMessage(`Player ${event.player} wins!`);
        // No further messages are expected; close the WebSocket connection.
        websocket.close(1000);
        break;
      case "error":
        showMessage(event.message);
        break;
      default:
        throw new Error(`Unsupported event type: ${event.type}.`);
    }
  });
}
function sendMoves(board, websocket) {
  // Don't send moves for a spectator watching a game.
  const params = new URLSearchParams(window.location.search);
  if (params.has("watch")) {
    return;
  }
  // When clicking a column, send a "play" event for a move in that column.
  board.addEventListener("click", ({ target }) => {
    const column = target.dataset.column;
    // Ignore clicks outside a column.
    if (column === undefined) {
      return;
    }
    const event = {
      type: "play",
      column: parseInt(column, 10),
    };
    websocket.send(JSON.stringify(event));
  });
}
window.addEventListener("DOMContentLoaded", () => {
  // Initialize the UI.
  const board = document.querySelector(".board");
  createBoard(board);
  // Open the WebSocket connection and register event handlers.
  const websocket = new WebSocket("ws://localhost:8001/");
  initGame(websocket);
  receiveMoves(board, websocket);
  sendMoves(board, websocket);
});

  • In the first part, you created a server and connected one browser; you could play if you shared the same browser.
  • In this second part, you connected a second browser; you could play from different browsers on a local network.
  • In this third part, you will deploy the game to the web; you can play from any browser connected to the Internet.

In the first and second parts of the tutorial, for local development, you ran an HTTP server on http://localhost:8000/ with:

$ python -m http.server

and a WebSocket server on ws://localhost:8001/ with:

$ python app.py

Now you want to deploy these servers on the Internet. There's a vast range of hosting providers to choose from. For the sake of simplicity, we'll rely on:

  • GitHub Pages for the HTTP server;
  • Heroku for the WebSocket server.

Perhaps you committed your work to git while you were progressing through the tutorial. If you didn't, now is a good time, because GitHub and Heroku offer git-based deployment workflows.

Initialize a git repository:

$ git init -b main
Initialized empty Git repository in websockets-tutorial/.git/
$ git commit --allow-empty -m "Initial commit."
[main (root-commit) ...] Initial commit.

Add all files and commit:

$ git add .
$ git commit -m "Initial implementation of Connect Four game."
[main ...] Initial implementation of Connect Four game.
 6 files changed, 500 insertions(+)
 create mode 100644 app.py
 create mode 100644 connect4.css
 create mode 100644 connect4.js
 create mode 100644 connect4.py
 create mode 100644 index.html
 create mode 100644 main.js

Before you deploy the server, you must adapt it to meet requirements of Heroku's runtime. This involves two small changes:

1.
Heroku expects the server to listen on a specific port, provided in the $PORT environment variable.
2.
Heroku sends a SIGTERM signal when shutting down a dyno, which should trigger a clean exit.

Adapt the main() coroutine accordingly:

import os
import signal
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    port = int(os.environ.get("PORT", "8001"))
    async with websockets.serve(handler, "", port):
        await stop

To catch the SIGTERM signal, main() creates a Future called stop and registers a signal handler that sets the result of this future. The value of the future doesn't matter; it's only for waiting for SIGTERM.

Then, by using serve() as a context manager and exiting the context when stop has a result, main() ensures that the server closes connections cleanly and exits on SIGTERM.

The app is now fully compatible with Heroku.

Create a requirements.txt file with this content to install websockets when building the image:

websockets

That's why you don't need to declare that you need a Python runtime.

Create a Procfile file with this content to configure the command for running the server:

web: python app.py

Commit your changes:

$ git add .
$ git commit -m "Deploy to Heroku."
[main ...] Deploy to Heroku.
 3 files changed, 12 insertions(+), 2 deletions(-)
 create mode 100644 Procfile
 create mode 100644 requirements.txt

Follow the set-up instructions to install the Heroku CLI and to log in, if you haven't done that yet.

Create a Heroku app. You must choose a unique name and replace websockets-tutorial by this name in the following command:

$ heroku create websockets-tutorial
Creating ⬢ websockets-tutorial... done
https://websockets-tutorial.herokuapp.com/ | https://git.heroku.com/websockets-tutorial.git

If you reuse a name that someone else already uses, you will receive this error; if this happens, try another name:

$ heroku create websockets-tutorial
Creating ⬢ websockets-tutorial... !
 ▸    Name websockets-tutorial is already taken

Deploy by pushing the code to Heroku:

$ git push heroku
... lots of output...
remote:        Released v1
remote:        https://websockets-tutorial.herokuapp.com/ deployed to Heroku
remote:
remote: Verifying deploy... done.
To https://git.heroku.com/websockets-tutorial.git
 * [new branch]      main -> main

You can test the WebSocket server with the interactive client exactly like you did in the first part of the tutorial. Replace websockets-tutorial by the name of your app in the following command:

$ python -m websockets wss://websockets-tutorial.herokuapp.com/
Connected to wss://websockets-tutorial.herokuapp.com/.
> {"type": "init"}
< {"type": "init", "join": "54ICxFae_Ip7TJE2", "watch": "634w44TblL5Dbd9a"}
Connection closed: 1000 (OK).

It works!

Before you deploy the web application, perhaps you're wondering how it will locate the WebSocket server? Indeed, at this point, its address is hard-coded in main.js:

const websocket = new WebSocket("ws://localhost:8001/");

You can take this strategy one step further by checking the address of the HTTP server and determining the address of the WebSocket server accordingly.

Add this function to main.js; replace python-websockets by your GitHub username and websockets-tutorial by the name of your app on Heroku:

function getWebSocketServer() {
  if (window.location.host === "python-websockets.github.io") {
    return "wss://websockets-tutorial.herokuapp.com/";
  } else if (window.location.host === "localhost:8000") {
    return "ws://localhost:8001/";
  } else {
    throw new Error(`Unsupported host: ${window.location.host}`);
  }
}

Then, update the initialization to connect to this address instead:

const websocket = new WebSocket(getWebSocketServer());

Commit your changes:

$ git add .
$ git commit -m "Configure WebSocket server address."
[main ...] Configure WebSocket server address.
 1 file changed, 11 insertions(+), 1 deletion(-)

Go to GitHub and create a new repository called websockets-tutorial.

Push your code to this repository. You must replace python-websockets by your GitHub username in the following command:

$ git remote add origin git@github.com:python-websockets/websockets-tutorial.git
$ git push -u origin main
Enumerating objects: 11, done.
Counting objects: 100% (11/11), done.
Delta compression using up to 8 threads
Compressing objects: 100% (10/10), done.
Writing objects: 100% (11/11), 5.90 KiB | 2.95 MiB/s, done.
Total 11 (delta 0), reused 0 (delta 0), pack-reused 0
To github.com:<username>/websockets-tutorial.git
 * [new branch]      main -> main
Branch 'main' set up to track remote branch 'main' from 'origin'.

Go back to GitHub, open the Settings tab of the repository and select Pages in the menu. Select the main branch as source and click Save. GitHub tells you that your site is published.

Follow the link and start a game!

In this third part of the tutorial, you learned how to deploy a WebSocket application with Heroku.

You can start a Connect Four game, send the JOIN link to a friend, and play over the Internet!

Congratulations for completing the tutorial. Enjoy building real-time web applications with websockets!

app.py

#!/usr/bin/env python
import asyncio
import json
import os
import secrets
import signal
import websockets
from connect4 import PLAYER1, PLAYER2, Connect4
JOIN = {}
WATCH = {}
async def error(websocket, message):
    """
    Send an error message.
    """
    event = {
        "type": "error",
        "message": message,
    }
    await websocket.send(json.dumps(event))
async def replay(websocket, game):
    """
    Send previous moves.
    """
    # Make a copy to avoid an exception if game.moves changes while iteration
    # is in progress. If a move is played while replay is running, moves will
    # be sent out of order but each move will be sent once and eventually the
    # UI will be consistent.
    for player, column, row in game.moves.copy():
        event = {
            "type": "play",
            "player": player,
            "column": column,
            "row": row,
        }
        await websocket.send(json.dumps(event))
async def play(websocket, game, player, connected):
    """
    Receive and process moves from a player.
    """
    async for message in websocket:
        # Parse a "play" event from the UI.
        event = json.loads(message)
        assert event["type"] == "play"
        column = event["column"]
        try:
            # Play the move.
            row = game.play(player, column)
        except RuntimeError as exc:
            # Send an "error" event if the move was illegal.
            await error(websocket, str(exc))
            continue
        # Send a "play" event to update the UI.
        event = {
            "type": "play",
            "player": player,
            "column": column,
            "row": row,
        }
        websockets.broadcast(connected, json.dumps(event))
        # If move is winning, send a "win" event.
        if game.winner is not None:
            event = {
                "type": "win",
                "player": game.winner,
            }
            websockets.broadcast(connected, json.dumps(event))
async def start(websocket):
    """
    Handle a connection from the first player: start a new game.
    """
    # Initialize a Connect Four game, the set of WebSocket connections
    # receiving moves from this game, and secret access tokens.
    game = Connect4()
    connected = {websocket}
    join_key = secrets.token_urlsafe(12)
    JOIN[join_key] = game, connected
    watch_key = secrets.token_urlsafe(12)
    WATCH[watch_key] = game, connected
    try:
        # Send the secret access tokens to the browser of the first player,
        # where they'll be used for building "join" and "watch" links.
        event = {
            "type": "init",
            "join": join_key,
            "watch": watch_key,
        }
        await websocket.send(json.dumps(event))
        # Receive and process moves from the first player.
        await play(websocket, game, PLAYER1, connected)
    finally:
        del JOIN[join_key]
        del WATCH[watch_key]
async def join(websocket, join_key):
    """
    Handle a connection from the second player: join an existing game.
    """
    # Find the Connect Four game.
    try:
        game, connected = JOIN[join_key]
    except KeyError:
        await error(websocket, "Game not found.")
        return
    # Register to receive moves from this game.
    connected.add(websocket)
    try:
        # Send the first move, in case the first player already played it.
        await replay(websocket, game)
        # Receive and process moves from the second player.
        await play(websocket, game, PLAYER2, connected)
    finally:
        connected.remove(websocket)
async def watch(websocket, watch_key):
    """
    Handle a connection from a spectator: watch an existing game.
    """
    # Find the Connect Four game.
    try:
        game, connected = WATCH[watch_key]
    except KeyError:
        await error(websocket, "Game not found.")
        return
    # Register to receive moves from this game.
    connected.add(websocket)
    try:
        # Send previous moves, in case the game already started.
        await replay(websocket, game)
        # Keep the connection open, but don't receive any messages.
        await websocket.wait_closed()
    finally:
        connected.remove(websocket)
async def handler(websocket):
    """
    Handle a connection and dispatch it according to who is connecting.
    """
    # Receive and parse the "init" event from the UI.
    message = await websocket.recv()
    event = json.loads(message)
    assert event["type"] == "init"
    if "join" in event:
        # Second player joins an existing game.
        await join(websocket, event["join"])
    elif "watch" in event:
        # Spectator watches an existing game.
        await watch(websocket, event["watch"])
    else:
        # First player starts a new game.
        await start(websocket)
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    port = int(os.environ.get("PORT", "8001"))
    async with websockets.serve(handler, "", port):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

index.html

<!DOCTYPE html>
<html lang="en">
  <head>
    <title>Connect Four</title>
  </head>
  <body>
    <div class="actions">
      <a class="action new" href="/">New</a>
      <a class="action join" href="">Join</a>
      <a class="action watch" href="">Watch</a>
    </div>
    <div class="board"></div>
    <script src="main.js" type="module"></script>
  </body>
</html>

main.js

import { createBoard, playMove } from "./connect4.js";
function getWebSocketServer() {
  if (window.location.host === "python-websockets.github.io") {
    return "wss://websockets-tutorial.herokuapp.com/";
  } else if (window.location.host === "localhost:8000") {
    return "ws://localhost:8001/";
  } else {
    throw new Error(`Unsupported host: ${window.location.host}`);
  }
}
function initGame(websocket) {
  websocket.addEventListener("open", () => {
    // Send an "init" event according to who is connecting.
    const params = new URLSearchParams(window.location.search);
    let event = { type: "init" };
    if (params.has("join")) {
      // Second player joins an existing game.
      event.join = params.get("join");
    } else if (params.has("watch")) {
      // Spectator watches an existing game.
      event.watch = params.get("watch");
    } else {
      // First player starts a new game.
    }
    websocket.send(JSON.stringify(event));
  });
}
function showMessage(message) {
  window.setTimeout(() => window.alert(message), 50);
}
function receiveMoves(board, websocket) {
  websocket.addEventListener("message", ({ data }) => {
    const event = JSON.parse(data);
    switch (event.type) {
      case "init":
        // Create links for inviting the second player and spectators.
        document.querySelector(".join").href = "?join=" + event.join;
        document.querySelector(".watch").href = "?watch=" + event.watch;
        break;
      case "play":
        // Update the UI with the move.
        playMove(board, event.player, event.column, event.row);
        break;
      case "win":
        showMessage(`Player ${event.player} wins!`);
        // No further messages are expected; close the WebSocket connection.
        websocket.close(1000);
        break;
      case "error":
        showMessage(event.message);
        break;
      default:
        throw new Error(`Unsupported event type: ${event.type}.`);
    }
  });
}
function sendMoves(board, websocket) {
  // Don't send moves for a spectator watching a game.
  const params = new URLSearchParams(window.location.search);
  if (params.has("watch")) {
    return;
  }
  // When clicking a column, send a "play" event for a move in that column.
  board.addEventListener("click", ({ target }) => {
    const column = target.dataset.column;
    // Ignore clicks outside a column.
    if (column === undefined) {
      return;
    }
    const event = {
      type: "play",
      column: parseInt(column, 10),
    };
    websocket.send(JSON.stringify(event));
  });
}
window.addEventListener("DOMContentLoaded", () => {
  // Initialize the UI.
  const board = document.querySelector(".board");
  createBoard(board);
  // Open the WebSocket connection and register event handlers.
  const websocket = new WebSocket(getWebSocketServer());
  initGame(websocket);
  receiveMoves(board, websocket);
  sendMoves(board, websocket);
});

Procfile

web: python app.py

requirements.txt

websockets

Look at the quick start guide.

In a hurry? Check out these examples.

Here are a few examples to get you started quickly with websockets.

Here's a WebSocket server.

It receives a name from the client, sends a greeting, and closes the connection.

server.py

#!/usr/bin/env python
import asyncio
import websockets
async def hello(websocket):
    name = await websocket.recv()
    print(f"<<< {name}")
    greeting = f"Hello {name}!"
    await websocket.send(greeting)
    print(f">>> {greeting}")
async def main():
    async with websockets.serve(hello, "localhost", 8765):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

serve() executes the connection handler coroutine hello() once for each WebSocket connection. It closes the WebSocket connection when the handler returns.

Here's a corresponding WebSocket client.

It sends a name to the server, receives a greeting, and closes the connection.

client.py

#!/usr/bin/env python
import asyncio
import websockets
async def hello():
    uri = "ws://localhost:8765"
    async with websockets.connect(uri) as websocket:
        name = input("What's your name? ")
        await websocket.send(name)
        print(f">>> {name}")
        greeting = await websocket.recv()
        print(f"<<< {greeting}")
if __name__ == "__main__":
    asyncio.run(hello())

Using connect() as an asynchronous context manager ensures the WebSocket connection is closed.

Secure WebSocket connections improve confidentiality and also reliability because they reduce the risk of interference by bad proxies.

The wss protocol is to ws what https is to http. The connection is encrypted with TLS (Transport Layer Security). wss requires certificates like https.

TLS is sometimes referred to as SSL (Secure Sockets Layer). SSL was an earlier encryption protocol; the name stuck.

Here's how to adapt the server to encrypt connections. You must download localhost.pem and save it in the same directory as server_secure.py.

server_secure.py

#!/usr/bin/env python
import asyncio
import pathlib
import ssl
import websockets
async def hello(websocket):
    name = await websocket.recv()
    print(f"<<< {name}")
    greeting = f"Hello {name}!"
    await websocket.send(greeting)
    print(f">>> {greeting}")
ssl_context = ssl.SSLContext(ssl.PROTOCOL_TLS_SERVER)
localhost_pem = pathlib.Path(__file__).with_name("localhost.pem")
ssl_context.load_cert_chain(localhost_pem)
async def main():
    async with websockets.serve(hello, "localhost", 8765, ssl=ssl_context):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

Here's how to adapt the client similarly.

client_secure.py

#!/usr/bin/env python
import asyncio
import pathlib
import ssl
import websockets
ssl_context = ssl.SSLContext(ssl.PROTOCOL_TLS_CLIENT)
localhost_pem = pathlib.Path(__file__).with_name("localhost.pem")
ssl_context.load_verify_locations(localhost_pem)
async def hello():
    uri = "wss://localhost:8765"
    async with websockets.connect(uri, ssl=ssl_context) as websocket:
        name = input("What's your name? ")
        await websocket.send(name)
        print(f">>> {name}")
        greeting = await websocket.recv()
        print(f"<<< {greeting}")
if __name__ == "__main__":
    asyncio.run(hello())

In this example, the client needs a TLS context because the server uses a self-signed certificate.

When connecting to a secure WebSocket server with a valid certificate — any certificate signed by a CA that your Python installation trusts — you can simply pass ssl=True to connect().

This example demonstrates the ssl argument with a TLS certificate shared between the client and the server. This is a simplistic setup.

Please review the advice and security considerations in the documentation of the ssl module to configure the TLS context securely.

The WebSocket protocol was invented for the web — as the name says!

Here's how to connect to a WebSocket server from a browser.

Run this script in a console:

show_time.py

#!/usr/bin/env python
import asyncio
import datetime
import random
import websockets
async def show_time(websocket):
    while True:
        message = datetime.datetime.utcnow().isoformat() + "Z"
        await websocket.send(message)
        await asyncio.sleep(random.random() * 2 + 1)
async def main():
    async with websockets.serve(show_time, "localhost", 5678):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

Save this file as show_time.html:

show_time.html

<!DOCTYPE html>
<html lang="en">
    <head>
        <title>WebSocket demo</title>
    </head>
    <body>
        <script src="show_time.js"></script>
    </body>
</html>

Save this file as show_time.js:

show_time.js

window.addEventListener("DOMContentLoaded", () => {
  const messages = document.createElement("ul");
  document.body.appendChild(messages);
  const websocket = new WebSocket("ws://localhost:5678/");
  websocket.onmessage = ({ data }) => {
    const message = document.createElement("li");
    const content = document.createTextNode(data);
    message.appendChild(content);
    messages.appendChild(message);
  };
});

Then, open show_time.html in several browsers. Clocks tick irregularly.

Let's change the previous example to send the same timestamps to all browsers, instead of generating independent sequences for each client.

Stop the previous script if it's still running and run this script in a console:

show_time_2.py

#!/usr/bin/env python
import asyncio
import datetime
import random
import websockets
CONNECTIONS = set()
async def register(websocket):
    CONNECTIONS.add(websocket)
    try:
        await websocket.wait_closed()
    finally:
        CONNECTIONS.remove(websocket)
async def show_time():
    while True:
        message = datetime.datetime.utcnow().isoformat() + "Z"
        websockets.broadcast(CONNECTIONS, message)
        await asyncio.sleep(random.random() * 2 + 1)
async def main():
    async with websockets.serve(register, "localhost", 5678):
        await show_time()
if __name__ == "__main__":
    asyncio.run(main())

Refresh show_time.html in all browsers. Clocks tick in sync.

A WebSocket server can receive events from clients, process them to update the application state, and broadcast the updated state to all connected clients.

Here's an example where any client can increment or decrement a counter. The concurrency model of asyncio guarantees that updates are serialized.

Run this script in a console:

counter.py

#!/usr/bin/env python
import asyncio
import json
import logging
import websockets
logging.basicConfig()
USERS = set()
VALUE = 0
def users_event():
    return json.dumps({"type": "users", "count": len(USERS)})
def value_event():
    return json.dumps({"type": "value", "value": VALUE})
async def counter(websocket):
    global USERS, VALUE
    try:
        # Register user
        USERS.add(websocket)
        websockets.broadcast(USERS, users_event())
        # Send current state to user
        await websocket.send(value_event())
        # Manage state changes
        async for message in websocket:
            event = json.loads(message)
            if event["action"] == "minus":
                VALUE -= 1
                websockets.broadcast(USERS, value_event())
            elif event["action"] == "plus":
                VALUE += 1
                websockets.broadcast(USERS, value_event())
            else:
                logging.error("unsupported event: %s", event)
    finally:
        # Unregister user
        USERS.remove(websocket)
        websockets.broadcast(USERS, users_event())
async def main():
    async with websockets.serve(counter, "localhost", 6789):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

Save this file as counter.html:

counter.html

<!DOCTYPE html>
<html lang="en">
    <head>
        <title>WebSocket demo</title>
        <link href="counter.css" rel="stylesheet">
    </head>
    <body>
        <div class="buttons">
            <div class="minus button">-</div>
            <div class="value">?</div>
            <div class="plus button">+</div>
        </div>
        <div class="state">
            <span class="users">?</span> online
        </div>
        <script src="counter.js"></script>
    </body>
</html>

Save this file as counter.css:

counter.css

body {
    font-family: "Courier New", sans-serif;
    text-align: center;
}
.buttons {
    font-size: 4em;
    display: flex;
    justify-content: center;
}
.button, .value {
    line-height: 1;
    padding: 2rem;
    margin: 2rem;
    border: medium solid;
    min-height: 1em;
    min-width: 1em;
}
.button {
    cursor: pointer;
    user-select: none;
}
.minus {
    color: red;
}
.plus {
    color: green;
}
.value {
    min-width: 2em;
}
.state {
    font-size: 2em;
}

Save this file as counter.js:

counter.js

window.addEventListener("DOMContentLoaded", () => {
  const websocket = new WebSocket("ws://localhost:6789/");
  document.querySelector(".minus").addEventListener("click", () => {
    websocket.send(JSON.stringify({ action: "minus" }));
  });
  document.querySelector(".plus").addEventListener("click", () => {
    websocket.send(JSON.stringify({ action: "plus" }));
  });
  websocket.onmessage = ({ data }) => {
    const event = JSON.parse(data);
    switch (event.type) {
      case "value":
        document.querySelector(".value").textContent = event.value;
        break;
      case "users":
        const users = `${event.count} user${event.count == 1 ? "" : "s"}`;
        document.querySelector(".users").textContent = users;
        break;
      default:
        console.error("unsupported event", event);
    }
  };
});

Then open counter.html file in several browsers and play with [+] and [-].

If you're stuck, perhaps you'll find the answer here.

•
Write a coroutine that handles a single connection. It receives a WebSocket protocol instance and the URI path in argument.
  • Call recv() and send() to receive and send messages at any time.
  • When recv() or send() raises ConnectionClosed, clean up and exit. If you started other asyncio.Task, terminate them before exiting.
  • If you aren't awaiting recv(), consider awaiting wait_closed() to detect quickly when the connection is closed.
  • You may ping() or pong() if you wish but it isn't needed in general.
•
Create a server with serve() which is similar to asyncio's create_server(). You can also use it as an asynchronous context manager.
  • The server takes care of establishing connections, then lets the handler execute the application logic, and finally closes the connection after the handler exits normally or with an exception.
  • For advanced customization, you may subclass WebSocketServerProtocol and pass either this subclass or a factory function as the create_protocol argument.

•
Create a client with connect() which is similar to asyncio's create_connection(). You can also use it as an asynchronous context manager.
•
For advanced customization, you may subclass WebSocketClientProtocol and pass either this subclass or a factory function as the create_protocol argument.
  • Call recv() and send() to receive and send messages at any time.
  • You may ping() or pong() if you wish but it isn't needed in general.
  • If you aren't using connect() as a context manager, call close() to terminate the connection.

If you don't understand what websockets is doing, enable logging:

import logging
logger = logging.getLogger('websockets')
logger.setLevel(logging.DEBUG)
logger.addHandler(logging.StreamHandler())

The logs contain:

  • Exceptions in the connection handler at the ERROR level
  • Exceptions in the opening or closing handshake at the INFO level
  • All frames at the DEBUG level — this can be very verbose

If you're new to asyncio, you will certainly encounter issues that are related to asynchronous programming in general rather than to websockets in particular. Fortunately Python's official documentation provides advice to develop with asyncio. Check it out: it's invaluable!

Here are typical patterns for processing messages in a WebSocket server or client. You will certainly implement some of them in your application.

This page gives examples of connection handlers for a server. However, they're also applicable to a client, simply by assuming that websocket is a connection created with connect().

WebSocket connections are long-lived. You will usually write a loop to process several messages during the lifetime of a connection.

To receive messages from the WebSocket connection:

async def consumer_handler(websocket):
    async for message in websocket:
        await consumer(message)

In this example, consumer() is a coroutine implementing your business logic for processing a message received on the WebSocket connection. Each message may be str or bytes.

Iteration terminates when the client disconnects.

To send messages to the WebSocket connection:

async def producer_handler(websocket):
    while True:
        message = await producer()
        await websocket.send(message)

In this example, producer() is a coroutine implementing your business logic for generating the next message to send on the WebSocket connection. Each message must be str or bytes.

Iteration terminates when the client disconnects because send() raises a ConnectionClosed exception, which breaks out of the while True loop.

You can receive and send messages on the same WebSocket connection by combining the consumer and producer patterns. This requires running two tasks in parallel:

async def handler(websocket):
    await asyncio.gather(
        consumer_handler(websocket),
        producer_handler(websocket),
    )

If a task terminates, gather() doesn't cancel the other task. This can result in a situation where the producer keeps running after the consumer finished, which may leak resources.

Here's a way to exit and close the WebSocket connection as soon as a task terminates, after canceling the other task:

async def handler(websocket):
    consumer_task = asyncio.create_task(consumer_handler(websocket))
    producer_task = asyncio.create_task(producer_handler(websocket))
    done, pending = await asyncio.wait(
        [consumer_task, producer_task],
        return_when=asyncio.FIRST_COMPLETED,
    )
    for task in pending:
        task.cancel()

To keep track of currently connected clients, you can register them when they connect and unregister them when they disconnect:

connected = set()
async def handler(websocket):
    # Register.
    connected.add(websocket)
    try:
        # Broadcast a message to all connected clients.
        websockets.broadcast(connected, "Hello!")
        await asyncio.sleep(10)
    finally:
        # Unregister.
        connected.remove(websocket)

This example maintains the set of connected clients in memory. This works as long as you run a single process. It doesn't scale to multiple processes.

If you plan to run multiple processes and you want to communicate updates between processes, then you must deploy a messaging system. You may find publish-subscribe functionality useful.

A complete implementation of this idea with Redis is described in the Django integration guide.

When developing a websockets server, you may run it locally to test changes. Unfortunately, whenever you want to try a new version of the code, you must stop the server and restart it, which slows down your development process.

Web frameworks such as Django or Flask provide a development server that reloads the application automatically when you make code changes. There is no such functionality in websockets because it's designed for production rather than development.

However, you can achieve the same result easily.

Install watchdog with the watchmedo shell utility:

$ pip install 'watchdog[watchmedo]'

Run your server with watchmedo auto-restart:

$ watchmedo auto-restart --pattern "*.py" --recursive --signal SIGTERM \
    python app.py

This example assumes that the server is defined in a script called app.py. Adapt it as necessary.

This guide will help you integrate websockets into a broader system.

If you're looking at adding real-time capabilities to a Django project with WebSocket, you have two main options.

1.
Using Django Channels, a project adding WebSocket to Django, among other features. This approach is fully supported by Django. However, it requires switching to a new deployment architecture.
2.
Deploying a separate WebSocket server next to your Django project. This technique is well suited when you need to add a small set of real-time features — maybe a notification service — to an HTTP application.

This guide shows how to implement the second technique with websockets. It assumes familiarity with Django.

Since the websockets server runs outside of Django, we need to integrate it with django.contrib.auth.

We will generate authentication tokens in the Django project. Then we will send them to the websockets server, where they will authenticate the user.

Generating a token for the current user and making it available in the browser is up to you. You could render the token in a template or fetch it with an API call.

Refer to the topic guide on authentication for details on this design.

We want secure, short-lived tokens containing the user ID. We'll rely on django-sesame, a small library designed exactly for this purpose.

Add django-sesame to the dependencies of your Django project, install it, and configure it in the settings of the project:

AUTHENTICATION_BACKENDS = [
    "django.contrib.auth.backends.ModelBackend",
    "sesame.backends.ModelBackend",
]

(If your project already uses another authentication backend than the default "django.contrib.auth.backends.ModelBackend", adjust accordingly.)

You don't need "sesame.middleware.AuthenticationMiddleware". It is for authenticating users in the Django server, while we're authenticating them in the websockets server.

We'd like our tokens to be valid for 30 seconds. We expect web pages to load and to establish the WebSocket connection within this delay. Configure django-sesame accordingly in the settings of your Django project:

SESAME_MAX_AGE = 30

If you expect your web site to load faster for all clients, a shorter lifespan is possible. However, in the context of this document, it would make manual testing more difficult.

You could also enable single-use tokens. However, this would update the last login date of the user every time a WebSocket connection is established. This doesn't seem like a good idea, both in terms of behavior and in terms of performance.

Now you can generate tokens in a django-admin shell as follows:

>>> from django.contrib.auth import get_user_model
>>> User = get_user_model()
>>> user = User.objects.get(username="<your username>")
>>> from sesame.utils import get_token
>>> get_token(user)
'<your token>'

Keep this console open: since tokens expire after 30 seconds, you'll have to generate a new token every time you want to test connecting to the server.

Let's move on to the websockets server.

Add websockets to the dependencies of your Django project and install it. Indeed, we're going to reuse the environment of the Django project, so we can call its APIs in the websockets server.

Now here's how to implement authentication.

#!/usr/bin/env python
import asyncio
import django
import websockets
django.setup()
from sesame.utils import get_user
from websockets.frames import CloseCode
async def handler(websocket):
    sesame = await websocket.recv()
    user = await asyncio.to_thread(get_user, sesame)
    if user is None:
        await websocket.close(CloseCode.INTERNAL_ERROR, "authentication failed")
        return
    await websocket.send(f"Hello {user}!")
async def main():
    async with websockets.serve(handler, "localhost", 8888):
        await asyncio.Future()  # run forever
if __name__ == "__main__":
    asyncio.run(main())

Let's unpack this code.

We're calling django.setup() before doing anything with Django because we're using Django in a standalone script. This assumes that the DJANGO_SETTINGS_MODULE environment variable is set to the Python path to your settings module.

The connection handler reads the first message received from the client, which is expected to contain a django-sesame token. Then it authenticates the user with get_user(), the API for authentication outside a view. If authentication fails, it closes the connection and exits.

When we call an API that makes a database query such as get_user(), we wrap the call in to_thread(). Indeed, the Django ORM doesn't support asynchronous I/O. It would block the event loop if it didn't run in a separate thread. to_thread() is available since Python 3.9. In earlier versions, use run_in_executor() instead.

Finally, we start a server with serve().

We're ready to test!

Save this code to a file called authentication.py, make sure the DJANGO_SETTINGS_MODULE environment variable is set properly, and start the websockets server:

$ python authentication.py

Generate a new token — remember, they're only valid for 30 seconds — and use it to connect to your server. Paste your token and press Enter when you get a prompt:

$ python -m websockets ws://localhost:8888/
Connected to ws://localhost:8888/
> <your token>
< Hello <your username>!
Connection closed: 1000 (OK).

It works!

If you enter an expired or invalid token, authentication fails and the server closes the connection:

$ python -m websockets ws://localhost:8888/
Connected to ws://localhost:8888.
> not a token
Connection closed: 1011 (internal error) authentication failed.

You can also test from a browser by generating a new token and running the following code in the JavaScript console of the browser:

websocket = new WebSocket("ws://localhost:8888/");
websocket.onopen = (event) => websocket.send("<your token>");
websocket.onmessage = (event) => console.log(event.data);

If you don't want to import your entire Django project into the websockets server, you can build a separate Django project with django.contrib.auth, django-sesame, a suitable User model, and a subset of the settings of the main project.

We can connect and authenticate but our server doesn't do anything useful yet!

Let's send a message every time a user makes an action in the admin. This message will be broadcast to all users who can access the model on which the action was made. This may be used for showing notifications to other users.

Many use cases for WebSocket with Django follow a similar pattern.

We need a event bus to enable communications between Django and websockets. Both sides connect permanently to the bus. Then Django writes events and websockets reads them. For the sake of simplicity, we'll rely on Redis Pub/Sub.

The easiest way to add Redis to a Django project is by configuring a cache backend with django-redis. This library manages connections to Redis efficiently, persisting them between requests, and provides an API to access the Redis connection directly.

Install Redis, add django-redis to the dependencies of your Django project, install it, and configure it in the settings of the project:

CACHES = {
    "default": {
        "BACKEND": "django_redis.cache.RedisCache",
        "LOCATION": "redis://127.0.0.1:6379/1",
    },
}

If you already have a default cache, add a new one with a different name and change get_redis_connection("default") in the code below to the same name.

Now let's write events to the bus.

Add the following code to a module that is imported when your Django project starts. Typically, you would put it in a signals.py module, which you would import in the AppConfig.ready() method of one of your apps:

import json
from django.contrib.admin.models import LogEntry
from django.db.models.signals import post_save
from django.dispatch import receiver
from django_redis import get_redis_connection
@receiver(post_save, sender=LogEntry)
def publish_event(instance, **kwargs):
    event = {
        "model": instance.content_type.name,
        "object": instance.object_repr,
        "message": instance.get_change_message(),
        "timestamp": instance.action_time.isoformat(),
        "user": str(instance.user),
        "content_type_id": instance.content_type_id,
        "object_id": instance.object_id,
    }
    connection = get_redis_connection("default")
    payload = json.dumps(event)
    connection.publish("events", payload)

This code runs every time the admin saves a LogEntry object to keep track of a change. It extracts interesting data, serializes it to JSON, and writes an event to Redis.

Let's check that it works:

$ redis-cli
127.0.0.1:6379> SELECT 1
OK
127.0.0.1:6379[1]> SUBSCRIBE events
Reading messages... (press Ctrl-C to quit)
1) "subscribe"
2) "events"
3) (integer) 1

Leave this command running, start the Django development server and make changes in the admin: add, modify, or delete objects. You should see corresponding events published to the "events" stream.

Now let's turn to reading events and broadcasting them to connected clients. We need to add several features:

  • Keep track of connected clients so we can broadcast messages.
  • Tell which content types the user has permission to view or to change.
  • Connect to the message bus and read events.
  • Broadcast these events to users who have corresponding permissions.

Here's a complete implementation.

#!/usr/bin/env python
import asyncio
import json
import aioredis
import django
import websockets
django.setup()
from django.contrib.contenttypes.models import ContentType
from sesame.utils import get_user
from websockets.frames import CloseCode
CONNECTIONS = {}
def get_content_types(user):
    """Return the set of IDs of content types visible by user."""
    # This does only three database queries because Django caches
    # all permissions on the first call to user.has_perm(...).
    return {
        ct.id
        for ct in ContentType.objects.all()
        if user.has_perm(f"{ct.app_label}.view_{ct.model}")
        or user.has_perm(f"{ct.app_label}.change_{ct.model}")
    }
async def handler(websocket):
    """Authenticate user and register connection in CONNECTIONS."""
    sesame = await websocket.recv()
    user = await asyncio.to_thread(get_user, sesame)
    if user is None:
        await websocket.close(CloseCode.INTERNAL_ERROR, "authentication failed")
        return
    ct_ids = await asyncio.to_thread(get_content_types, user)
    CONNECTIONS[websocket] = {"content_type_ids": ct_ids}
    try:
        await websocket.wait_closed()
    finally:
        del CONNECTIONS[websocket]
async def process_events():
    """Listen to events in Redis and process them."""
    redis = aioredis.from_url("redis://127.0.0.1:6379/1")
    pubsub = redis.pubsub()
    await pubsub.subscribe("events")
    async for message in pubsub.listen():
        if message["type"] != "message":
            continue
        payload = message["data"].decode()
        # Broadcast event to all users who have permissions to see it.
        event = json.loads(payload)
        recipients = (
            websocket
            for websocket, connection in CONNECTIONS.items()
            if event["content_type_id"] in connection["content_type_ids"]
        )
        websockets.broadcast(recipients, payload)
async def main():
    async with websockets.serve(handler, "localhost", 8888):
        await process_events()  # runs forever
if __name__ == "__main__":
    asyncio.run(main())

Since the get_content_types() function makes a database query, it is wrapped inside asyncio.to_thread(). It runs once when each WebSocket connection is open; then its result is cached for the lifetime of the connection. Indeed, running it for each message would trigger database queries for all connected users at the same time, which would hurt the database.

The connection handler merely registers the connection in a global variable, associated to the list of content types for which events should be sent to that connection, and waits until the client disconnects.

The process_events() function reads events from Redis and broadcasts them to all connections that should receive them. We don't care much if a sending a notification fails — this happens when a connection drops between the moment we iterate on connections and the moment the corresponding message is sent — so we start a task with for each message and forget about it. Also, this means we're immediately ready to process the next event, even if it takes time to send a message to a slow client.

Since Redis can publish a message to multiple subscribers, multiple instances of this server can safely run in parallel.

In theory, given enough servers, this design can scale to a hundred million clients, since Redis can handle ten thousand servers and each server can handle ten thousand clients. In practice, you would need a more scalable message bus before reaching that scale, due to the volume of messages.

The WebSocket protocol makes provisions for extending or specializing its features, which websockets supports fully.

During the opening handshake, WebSocket clients and servers negotiate which extensions will be used with which parameters. Then each frame is processed by extensions before being sent or after being received.

As a consequence, writing an extension requires implementing several classes:

  • Extension Factory: it negotiates parameters and instantiates the extension.

    Clients and servers require separate extension factories with distinct APIs.

    Extension factories are the public API of an extension.

  • Extension: it decodes incoming frames and encodes outgoing frames.

    If the extension is symmetrical, clients and servers can use the same class.

    Extensions are initialized by extension factories, so they don't need to be part of the public API of an extension.

websockets provides base classes for extension factories and extensions. See ClientExtensionFactory, ServerExtensionFactory, and Extension for details.

Once your application is ready, learn how to deploy it on various platforms.

This guide describes how to deploy a websockets server to Render.

However, on a free plan, connections are dropped after five minutes, which is quite short for WebSocket application.

We're going to deploy a very simple app. The process would be identical for a more realistic app.

Create repository

Deploying to Render requires a git repository. Let's initialize one:

$ mkdir websockets-echo
$ cd websockets-echo
$ git init -b main
Initialized empty Git repository in websockets-echo/.git/
$ git commit --allow-empty -m "Initial commit."
[main (root-commit) 816c3b1] Initial commit.

Render requires the git repository to be hosted at GitHub or GitLab.

Sign up or log in to GitHub. Create a new repository named websockets-echo. Don't enable any of the initialization options offered by GitHub. Then, follow instructions for pushing an existing repository from the command line.

After pushing, refresh your repository's homepage on GitHub. You should see an empty repository with an empty initial commit.

Create application

Here's the implementation of the app, an echo server. Save it in a file called app.py:

#!/usr/bin/env python
import asyncio
import http
import signal
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def health_check(path, request_headers):
    if path == "/healthz":
        return http.HTTPStatus.OK, [], b"OK\n"
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(
        echo,
        host="",
        port=8080,
        process_request=health_check,
    ):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

This app implements requirements for zero downtime deploys:

  • it provides a health check at /healthz;
  • it closes connections and exits cleanly when it receives a SIGTERM signal.

Create a requirements.txt file containing this line to declare a dependency on websockets:

websockets

Confirm that you created the correct files and commit them to git:

$ ls
app.py           requirements.txt
$ git add .
$ git commit -m "Initial implementation."
[main f26bf7f] Initial implementation.
2 files changed, 37 insertions(+)
create mode 100644 app.py
create mode 100644 requirements.txt

Push the changes to GitHub:

$ git push
...
To github.com:<username>/websockets-echo.git
   816c3b1..f26bf7f  main -> main

The app is ready. Let's deploy it!

Deploy application

Sign up or log in to Render.

Create a new web service. Connect the git repository that you just created.

Then, finalize the configuration of your app as follows:

  • Name: websockets-echo
  • Start Command: python app.py

If you're just experimenting, select the free plan. Create the web service.

To configure the health check, go to Settings, scroll down to Health & Alerts, and set:

•
Health Check Path: /healthz

This triggers a new deployment.

Validate deployment

Let's confirm that your application is running as expected.

Since it's a WebSocket server, you need a WebSocket client, such as the interactive client that comes with websockets.

If you're currently building a websockets server, perhaps you're already in a virtualenv where websockets is installed. If not, you can install it in a new virtualenv as follows:

$ python -m venv websockets-client
$ . websockets-client/bin/activate
$ pip install websockets

Connect the interactive client — you must replace websockets-echo with the name of your Render app in this command:

$ python -m websockets wss://websockets-echo.onrender.com/
Connected to wss://websockets-echo.onrender.com/.
>

Great! Your app is running!

Once you're connected, you can send any message and the server will echo it, or press Ctrl-D to terminate the connection:

> Hello!
< Hello!
Connection closed: 1000 (OK).

You can also confirm that your application shuts down gracefully when you deploy a new version. Due to limitations of Render's free plan, you must upgrade to a paid plan before you perform this test.

Connect an interactive client again — remember to replace websockets-echo with your app:

$ python -m websockets wss://websockets-echo.onrender.com/
Connected to wss://websockets-echo.onrender.com/.
>

Trigger a new deployment with Manual Deploy > Deploy latest commit. When the deployment completes, the connection is closed with code 1001 (going away).

$ python -m websockets wss://websockets-echo.onrender.com/
Connected to wss://websockets-echo.onrender.com/.
Connection closed: 1001 (going away).

If graceful shutdown wasn't working, the server wouldn't perform a closing handshake and the connection would be closed with code 1006 (abnormal closure).

Remember to downgrade to a free plan if you upgraded just for testing this feature.

This guide describes how to deploy a websockets server to Fly.

The free tier include up to three small VMs. This guide uses only one.

We're going to deploy a very simple app. The process would be identical for a more realistic app.

Create application

Here's the implementation of the app, an echo server. Save it in a file called app.py:

#!/usr/bin/env python
import asyncio
import http
import signal
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def health_check(path, request_headers):
    if path == "/healthz":
        return http.HTTPStatus.OK, [], b"OK\n"
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(
        echo,
        host="",
        port=8080,
        process_request=health_check,
    ):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

This app implements typical requirements for running on a Platform as a Service:

  • it provides a health check at /healthz;
  • it closes connections and exits cleanly when it receives a SIGTERM signal.

Create a requirements.txt file containing this line to declare a dependency on websockets:

websockets

The app is ready. Let's deploy it!

Deploy application

Follow the instructions to install the Fly CLI, if you haven't done that yet.

Sign up or log in to Fly.

Launch the app — you'll have to pick a different name because I'm already using websockets-echo:

$ fly launch
Creating app in ...
Scanning source code
Detected a Python app
Using the following build configuration:
    Builder: paketobuildpacks/builder:base
? App Name (leave blank to use an auto-generated name): websockets-echo
? Select organization: ...
? Select region: ...
Created app websockets-echo in organization ...
Wrote config file fly.toml
? Would you like to set up a Postgresql database now? No
We have generated a simple Procfile for you. Modify it to fit your needs and run "fly deploy" to deploy your application.

Fly can build images with a Dockerfile or a buildpack. Here, fly launch configures a generic Paketo buildpack.

If you'd rather package the app with a Dockerfile, check out the guide to containerize an application.

Replace the auto-generated fly.toml with:

app = "websockets-echo"
kill_signal = "SIGTERM"
[build]
  builder = "paketobuildpacks/builder:base"
[[services]]
  internal_port = 8080
  protocol = "tcp"
  [[services.http_checks]]
    path = "/healthz"
  [[services.ports]]
    handlers = ["tls", "http"]
    port = 443

This configuration:

  • listens on port 443, terminates TLS, and forwards to the app on port 8080;
  • declares a health check at /healthz;
  • requests a SIGTERM for terminating the app.

Replace the auto-generated Procfile with:

web: python app.py

This tells Fly how to run the app.

Now you can deploy it:

$ fly deploy
... lots of output...
==> Monitoring deployment
1 desired, 1 placed, 1 healthy, 0 unhealthy [health checks: 1 total, 1 passing]
--> v0 deployed successfully

Validate deployment

Let's confirm that your application is running as expected.

Since it's a WebSocket server, you need a WebSocket client, such as the interactive client that comes with websockets.

If you're currently building a websockets server, perhaps you're already in a virtualenv where websockets is installed. If not, you can install it in a new virtualenv as follows:

$ python -m venv websockets-client
$ . websockets-client/bin/activate
$ pip install websockets

Connect the interactive client — you must replace websockets-echo with the name of your Fly app in this command:

$ python -m websockets wss://websockets-echo.fly.dev/
Connected to wss://websockets-echo.fly.dev/.
>

Great! Your app is running!

Once you're connected, you can send any message and the server will echo it, or press Ctrl-D to terminate the connection:

> Hello!
< Hello!
Connection closed: 1000 (OK).

You can also confirm that your application shuts down gracefully.

Connect an interactive client again — remember to replace websockets-echo with your app:

$ python -m websockets wss://websockets-echo.fly.dev/
Connected to wss://websockets-echo.fly.dev/.
>

In another shell, restart the app — again, replace websockets-echo with your app:

$ fly restart websockets-echo
websockets-echo is being restarted

Go back to the first shell. The connection is closed with code 1001 (going away).

$ python -m websockets wss://websockets-echo.fly.dev/
Connected to wss://websockets-echo.fly.dev/.
Connection closed: 1001 (going away).

If graceful shutdown wasn't working, the server wouldn't perform a closing handshake and the connection would be closed with code 1006 (abnormal closure).

This guide describes how to deploy a websockets server to Heroku. The same principles should apply to other Platform as a Service providers.

When this tutorial was written, in September 2021, Heroku offered a free tier where a websockets app could run at no cost. In November 2022, Heroku removed the free tier, making it impossible to maintain this document. As a consequence, it isn't updated anymore and may be removed in the future.

We're going to deploy a very simple app. The process would be identical for a more realistic app.

Create repository

Deploying to Heroku requires a git repository. Let's initialize one:

$ mkdir websockets-echo
$ cd websockets-echo
$ git init -b main
Initialized empty Git repository in websockets-echo/.git/
$ git commit --allow-empty -m "Initial commit."
[main (root-commit) 1e7947d] Initial commit.

Create application

Here's the implementation of the app, an echo server. Save it in a file called app.py:

#!/usr/bin/env python
import asyncio
import signal
import os
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(
        echo,
        host="",
        port=int(os.environ["PORT"]),
    ):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

Heroku expects the server to listen on a specific port, which is provided in the $PORT environment variable. The app reads it and passes it to serve().

Heroku sends a SIGTERM signal to all processes when shutting down a dyno. When the app receives this signal, it closes connections and exits cleanly.

Create a requirements.txt file containing this line to declare a dependency on websockets:

websockets

Create a Procfile.

web: python app.py

This tells Heroku how to run the app.

Confirm that you created the correct files and commit them to git:

$ ls
Procfile         app.py           requirements.txt
$ git add .
$ git commit -m "Initial implementation."
[main 8418c62] Initial implementation.
 3 files changed, 32 insertions(+)
 create mode 100644 Procfile
 create mode 100644 app.py
 create mode 100644 requirements.txt

The app is ready. Let's deploy it!

Deploy application

Follow the instructions to install the Heroku CLI, if you haven't done that yet.

Sign up or log in to Heroku.

Create a Heroku app — you'll have to pick a different name because I'm already using websockets-echo:

$ heroku create websockets-echo
Creating ⬢ websockets-echo... done
https://websockets-echo.herokuapp.com/ | https://git.heroku.com/websockets-echo.git
$ git push heroku
... lots of output...
remote: -----> Launching...
remote:        Released v1
remote:        https://websockets-echo.herokuapp.com/ deployed to Heroku
remote:
remote: Verifying deploy... done.
To https://git.heroku.com/websockets-echo.git
 * [new branch]      main -> main

Validate deployment

Let's confirm that your application is running as expected.

Since it's a WebSocket server, you need a WebSocket client, such as the interactive client that comes with websockets.

If you're currently building a websockets server, perhaps you're already in a virtualenv where websockets is installed. If not, you can install it in a new virtualenv as follows:

$ python -m venv websockets-client
$ . websockets-client/bin/activate
$ pip install websockets

Connect the interactive client — you must replace websockets-echo with the name of your Heroku app in this command:

$ python -m websockets wss://websockets-echo.herokuapp.com/
Connected to wss://websockets-echo.herokuapp.com/.
>

Great! Your app is running!

Once you're connected, you can send any message and the server will echo it, or press Ctrl-D to terminate the connection:

> Hello!
< Hello!
Connection closed: 1000 (OK).

You can also confirm that your application shuts down gracefully.

Connect an interactive client again — remember to replace websockets-echo with your app:

$ python -m websockets wss://websockets-echo.herokuapp.com/
Connected to wss://websockets-echo.herokuapp.com/.
>

In another shell, restart the app — again, replace websockets-echo with your app:

$ heroku dyno:restart -a websockets-echo
Restarting dynos on ⬢ websockets-echo... done

Go back to the first shell. The connection is closed with code 1001 (going away).

$ python -m websockets wss://websockets-echo.herokuapp.com/
Connected to wss://websockets-echo.herokuapp.com/.
Connection closed: 1001 (going away).

If graceful shutdown wasn't working, the server wouldn't perform a closing handshake and the connection would be closed with code 1006 (abnormal closure).

This guide describes how to deploy a websockets server to Kubernetes. It assumes familiarity with Docker and Kubernetes.

We're going to deploy a simple app to a local Kubernetes cluster and to ensure that it scales as expected.

In a more realistic context, you would follow your organization's practices for deploying to Kubernetes, but you would apply the same principles as far as websockets is concerned.

Here's the app we're going to deploy. Save it in a file called app.py:

#!/usr/bin/env python
import asyncio
import http
import signal
import sys
import time
import websockets
async def slow_echo(websocket):
    async for message in websocket:
        # Block the event loop! This allows saturating a single asyncio
        # process without opening an impractical number of connections.
        time.sleep(0.1)  # 100ms
        await websocket.send(message)
async def health_check(path, request_headers):
    if path == "/healthz":
        return http.HTTPStatus.OK, [], b"OK\n"
    if path == "/inemuri":
        loop = asyncio.get_running_loop()
        loop.call_later(1, time.sleep, 10)
        return http.HTTPStatus.OK, [], b"Sleeping for 10s\n"
    if path == "/seppuku":
        loop = asyncio.get_running_loop()
        loop.call_later(1, sys.exit, 69)
        return http.HTTPStatus.OK, [], b"Terminating\n"
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(
        slow_echo,
        host="",
        port=80,
        process_request=health_check,
    ):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

This is an echo server with one twist: every message blocks the server for 100ms, which creates artificial starvation of CPU time. This makes it easier to saturate the server for load testing.

The app exposes a health check on /healthz. It also provides two other endpoints for testing purposes: /inemuri will make the app unresponsive for 10 seconds and /seppuku will terminate it.

The quest for the perfect Python container image is out of scope of this guide, so we'll go for the simplest possible configuration instead:

FROM python:3.9-alpine
RUN pip install websockets
COPY app.py .
CMD ["python", "app.py"]

After saving this Dockerfile, build the image:

$ docker build -t websockets-test:1.0 .

Test your image by running:

$ docker run --name run-websockets-test --publish 32080:80 --rm \
    websockets-test:1.0

Then, in another shell, in a virtualenv where websockets is installed, connect to the app and check that it echoes anything you send:

$ python -m websockets ws://localhost:32080/
Connected to ws://localhost:32080/.
> Hey there!
< Hey there!
>

Now, in yet another shell, stop the app with:

$ docker kill -s TERM run-websockets-test

Going to the shell where you connected to the app, you can confirm that it shut down gracefully:

$ python -m websockets ws://localhost:32080/
Connected to ws://localhost:32080/.
> Hey there!
< Hey there!
Connection closed: 1001 (going away).

If it didn't, you'd get code 1006 (abnormal closure).

Deploy application

Configuring Kubernetes is even further beyond the scope of this guide, so we'll use a basic configuration for testing, with just one Service and one Deployment:

apiVersion: v1
kind: Service
metadata:
  name: websockets-test
spec:
  type: NodePort
  ports:
    - port: 80
      nodePort: 32080
  selector:
    app: websockets-test
---
apiVersion: apps/v1
kind: Deployment
metadata:
  name: websockets-test
spec:
  selector:
    matchLabels:
      app: websockets-test
  template:
    metadata:
      labels:
        app: websockets-test
    spec:
      containers:
      - name: websockets-test
        image: websockets-test:1.0
        livenessProbe:
          httpGet:
            path: /healthz
            port: 80
          periodSeconds: 1
        ports:
        - containerPort: 80

For local testing, a service of type NodePort is good enough. For deploying to production, you would configure an Ingress.

After saving this to a file called deployment.yaml, you can deploy:

$ kubectl apply -f deployment.yaml
service/websockets-test created
deployment.apps/websockets-test created

Now you have a deployment with one pod running:

$ kubectl get deployment websockets-test
NAME              READY   UP-TO-DATE   AVAILABLE   AGE
websockets-test   1/1     1            1           10s
$ kubectl get pods -l app=websockets-test
NAME                               READY   STATUS    RESTARTS   AGE
websockets-test-86b48f4bb7-nltfh   1/1     Running   0          10s

You can connect to the service — press Ctrl-D to exit:

$ python -m websockets ws://localhost:32080/
Connected to ws://localhost:32080/.
Connection closed: 1000 (OK).

Validate deployment

First, let's ensure the liveness probe works by making the app unresponsive:

$ curl http://localhost:32080/inemuri
Sleeping for 10s

Since we have only one pod, we know that this pod will go to sleep.

The liveness probe is configured to run every second. By default, liveness probes time out after one second and have a threshold of three failures. Therefore Kubernetes should restart the pod after at most 5 seconds.

Indeed, after a few seconds, the pod reports a restart:

$ kubectl get pods -l app=websockets-test
NAME                               READY   STATUS    RESTARTS   AGE
websockets-test-86b48f4bb7-nltfh   1/1     Running   1          42s

Next, let's take it one step further and crash the app:

$ curl http://localhost:32080/seppuku
Terminating

The pod reports a second restart:

$ kubectl get pods -l app=websockets-test
NAME                               READY   STATUS    RESTARTS   AGE
websockets-test-86b48f4bb7-nltfh   1/1     Running   2          72s

All good — Kubernetes delivers on its promise to keep our app alive!

Of course, Kubernetes is for scaling. Let's scale — modestly — to 10 pods:

$ kubectl scale deployment.apps/websockets-test --replicas=10
deployment.apps/websockets-test scaled

After a few seconds, we have 10 pods running:

$ kubectl get deployment websockets-test
NAME              READY   UP-TO-DATE   AVAILABLE   AGE
websockets-test   10/10   10           10          10m

Now let's generate load. We'll use this script:

#!/usr/bin/env python
import asyncio
import sys
import websockets
URI = "ws://localhost:32080"
async def run(client_id, messages):
    async with websockets.connect(URI) as websocket:
        for message_id in range(messages):
            await websocket.send(f"{client_id}:{message_id}")
            await websocket.recv()
async def benchmark(clients, messages):
    await asyncio.wait([
        asyncio.create_task(run(client_id, messages))
        for client_id in range(clients)
    ])
if __name__ == "__main__":
    clients, messages = int(sys.argv[1]), int(sys.argv[2])
    asyncio.run(benchmark(clients, messages))

We'll connect 500 clients in parallel, meaning 50 clients per pod, and have each client send 6 messages. Since the app blocks for 100ms before responding, if connections are perfectly distributed, we expect a total run time slightly over 50 * 6 * 0.1 = 30 seconds.

Let's try it:

$ ulimit -n 512
$ time python benchmark.py 500 6
python benchmark.py 500 6  2.40s user 0.51s system 7% cpu 36.471 total

A total runtime of 36 seconds is in the right ballpark. Repeating this experiment with other parameters shows roughly consistent results, with the high variability you'd expect from a quick benchmark without any effort to stabilize the test setup.

Finally, we can scale back to one pod.

$ kubectl scale deployment.apps/websockets-test --replicas=1
deployment.apps/websockets-test scaled
$ kubectl get deployment websockets-test
NAME              READY   UP-TO-DATE   AVAILABLE   AGE
websockets-test   1/1     1            1           15m

This guide proposes a simple way to deploy a websockets server directly on a Linux or BSD operating system.

We'll configure Supervisor to run several server processes and to restart them if needed.

We'll bind all servers to the same port. The OS will take care of balancing connections.

Create and activate a virtualenv:

$ python -m venv supervisor-websockets
$ . supervisor-websockets/bin/activate

Install websockets and Supervisor:

$ pip install websockets
$ pip install supervisor

Save this app to a file called app.py:

#!/usr/bin/env python
import asyncio
import signal
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(
        echo,
        host="",
        port=8080,
        reuse_port=True,
    ):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

This is an echo server with two features added for the purpose of this guide:

  • It shuts down gracefully when receiving a SIGTERM signal;
  • It enables the reuse_port option of create_server(), which in turns sets SO_REUSEPORT on the accept socket.

Save this Supervisor configuration to supervisord.conf:

[supervisord]
[program:websockets-test]
command = python app.py
process_name = %(program_name)s_%(process_num)02d
numprocs = 4
autorestart = true

This is the minimal configuration required to keep four instances of the app running, restarting them if they exit.

Now start Supervisor in the foreground:

$ supervisord -c supervisord.conf -n
INFO Increased RLIMIT_NOFILE limit to 1024
INFO supervisord started with pid 43596
INFO spawned: 'websockets-test_00' with pid 43597
INFO spawned: 'websockets-test_01' with pid 43598
INFO spawned: 'websockets-test_02' with pid 43599
INFO spawned: 'websockets-test_03' with pid 43600
INFO success: websockets-test_00 entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
INFO success: websockets-test_01 entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
INFO success: websockets-test_02 entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)
INFO success: websockets-test_03 entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)

In another shell, after activating the virtualenv, we can connect to the app — press Ctrl-D to exit:

$ python -m websockets ws://localhost:8080/
Connected to ws://localhost:8080/.
> Hello!
< Hello!
Connection closed: 1000 (OK).

Look at the pid of an instance of the app in the logs and terminate it:

$ kill -TERM 43597

The logs show that Supervisor restarted this instance:

INFO exited: websockets-test_00 (exit status 0; expected)
INFO spawned: 'websockets-test_00' with pid 43629
INFO success: websockets-test_00 entered RUNNING state, process has stayed up for > than 1 seconds (startsecs)

Now let's check what happens when we shut down Supervisor, but first let's establish a connection and leave it open:

$ python -m websockets ws://localhost:8080/
Connected to ws://localhost:8080/.
>

Look at the pid of supervisord itself in the logs and terminate it:

$ kill -TERM 43596

The logs show that Supervisor terminated all instances of the app before exiting:

WARN received SIGTERM indicating exit request
INFO waiting for websockets-test_00, websockets-test_01, websockets-test_02, websockets-test_03 to die
INFO stopped: websockets-test_02 (exit status 0)
INFO stopped: websockets-test_03 (exit status 0)
INFO stopped: websockets-test_01 (exit status 0)
INFO stopped: websockets-test_00 (exit status 0)

And you can see that the connection to the app was closed gracefully:

$ python -m websockets ws://localhost:8080/
Connected to ws://localhost:8080/.
Connection closed: 1001 (going away).

In this example, we've been sharing the same virtualenv for supervisor and websockets.

In a real deployment, you would likely:

  • Install Supervisor with the package manager of the OS.
  • Create a virtualenv dedicated to your application.
  • Add environment=PATH="path/to/your/virtualenv/bin" in the Supervisor configuration. Then python app.py runs in that virtualenv.

This guide demonstrates a way to load balance connections across multiple websockets server processes running on the same machine with nginx.

We'll run server processes with Supervisor as described in this guide.

Run server processes

Save this app to app.py:

#!/usr/bin/env python
import asyncio
import os
import signal
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.unix_serve(
        echo,
        path=f"{os.environ['SUPERVISOR_PROCESS_NAME']}.sock",
    ):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

We'd like to nginx to connect to websockets servers via Unix sockets in order to avoid the overhead of TCP for communicating between processes running in the same OS.

We start the app with unix_serve(). Each server process listens on a different socket thanks to an environment variable set by Supervisor to a different value.

Save this configuration to supervisord.conf:

[supervisord]
[program:websockets-test]
command = python app.py
process_name = %(program_name)s_%(process_num)02d
numprocs = 4
autorestart = true

This configuration runs four instances of the app.

Install Supervisor and run it:

$ supervisord -c supervisord.conf -n

Here's a simple nginx configuration to load balance connections across four processes:

daemon off;
events {
}
http {
    server {
        listen localhost:8080;
        location / {
            proxy_http_version 1.1;
            proxy_pass http://websocket;
            proxy_set_header Connection $http_connection;
            proxy_set_header Upgrade $http_upgrade;
        }
    }
    upstream websocket {
        least_conn;
        server unix:websockets-test_00.sock;
        server unix:websockets-test_01.sock;
        server unix:websockets-test_02.sock;
        server unix:websockets-test_03.sock;
    }
}

We set daemon off so we can run nginx in the foreground for testing.

Then we combine the WebSocket proxying and load balancing guides:

  • The WebSocket protocol requires HTTP/1.1. We must set the HTTP protocol version to 1.1, else nginx defaults to HTTP/1.0 for proxying.
  • The WebSocket handshake involves the Connection and Upgrade HTTP headers. We must pass them to the upstream explicitly, else nginx drops them because they're hop-by-hop headers.

    We deviate from the WebSocket proxying guide because its example adds a Connection: Upgrade header to every upstream request, even if the original request didn't contain that header.

  • In the upstream configuration, we set the load balancing method to least_conn in order to balance the number of active connections across servers. This is best for long running connections.

Save the configuration to nginx.conf, install nginx, and run it:

$ nginx -c nginx.conf -p .

You can confirm that nginx proxies connections properly:

$ PYTHONPATH=src python -m websockets ws://localhost:8080/
Connected to ws://localhost:8080/.
> Hello!
< Hello!
Connection closed: 1000 (OK).

This guide demonstrates a way to load balance connections across multiple websockets server processes running on the same machine with HAProxy.

We'll run server processes with Supervisor as described in this guide.

Run server processes

Save this app to app.py:

#!/usr/bin/env python
import asyncio
import os
import signal
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def main():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(
        echo,
        host="localhost",
        port=8000 + int(os.environ["SUPERVISOR_PROCESS_NAME"][-2:]),
    ):
        await stop
if __name__ == "__main__":
    asyncio.run(main())

Each server process listens on a different port by extracting an incremental index from an environment variable set by Supervisor.

Save this configuration to supervisord.conf:

[supervisord]
[program:websockets-test]
command = python app.py
process_name = %(program_name)s_%(process_num)02d
numprocs = 4
autorestart = true

This configuration runs four instances of the app.

Install Supervisor and run it:

$ supervisord -c supervisord.conf -n

Here's a simple HAProxy configuration to load balance connections across four processes:

defaults
    mode http
    timeout connect 10s
    timeout client 30s
    timeout server 30s
frontend websocket
    bind localhost:8080
    default_backend websocket
backend websocket
    balance leastconn
    server websockets-test_00 localhost:8000
    server websockets-test_01 localhost:8001
    server websockets-test_02 localhost:8002
    server websockets-test_03 localhost:8003

In the backend configuration, we set the load balancing method to leastconn in order to balance the number of active connections across servers. This is best for long running connections.

Save the configuration to haproxy.cfg, install HAProxy, and run it:

$ haproxy -f haproxy.cfg

You can confirm that HAProxy proxies connections properly:

$ PYTHONPATH=src python -m websockets ws://localhost:8080/
Connected to ws://localhost:8080/.
> Hello!
< Hello!
Connection closed: 1000 (OK).

If you're integrating the Sans-I/O layer of websockets into a library, rather than building an application with websockets, follow this guide.

This guide explains how to integrate the Sans-I/O layer of websockets to add support for WebSocket in another library.

As a prerequisite, you should decide how you will handle network I/O and asynchronous control flow.

Your integration layer will provide an API for the application on one side, will talk to the network on the other side, and will rely on websockets to implement the protocol in the middle. [image]

Opening a connection

Client-side

If you're building a client, parse the URI you'd like to connect to:

from websockets.uri import parse_uri
wsuri = parse_uri("ws://example.com/")

Open a TCP connection to (wsuri.host, wsuri.port) and perform a TLS handshake if wsuri.secure is True.

Initialize a ClientProtocol:

from websockets.client import ClientProtocol
protocol = ClientProtocol(wsuri)

Create a WebSocket handshake request with connect() and send it with send_request():

request = protocol.connect()
protocol.send_request(request)

Then, call data_to_send() and send its output to the network, as described in Send data below.

Once you receive enough data, as explained in Receive data below, the first event returned by events_received() is the WebSocket handshake response.

When the handshake fails, the reason is available in handshake_exc:

if protocol.handshake_exc is not None:
    raise protocol.handshake_exc

Else, the WebSocket connection is open.

A WebSocket client API usually performs the handshake then returns a wrapper around the network socket and the ClientProtocol.

Server-side

If you're building a server, accept network connections from clients and perform a TLS handshake if desired.

For each connection, initialize a ServerProtocol:

from websockets.server import ServerProtocol
protocol = ServerProtocol()

Once you receive enough data, as explained in Receive data below, the first event returned by events_received() is the WebSocket handshake request.

Create a WebSocket handshake response with accept() and send it with send_response():

response = protocol.accept(request)
protocol.send_response(response)

Alternatively, you may reject the WebSocket handshake and return an HTTP response with reject():

response = protocol.reject(status, explanation)
protocol.send_response(response)

Then, call data_to_send() and send its output to the network, as described in Send data below.

Even when you call accept(), the WebSocket handshake may fail if the request is incorrect or unsupported.

When the handshake fails, the reason is available in handshake_exc:

if protocol.handshake_exc is not None:
    raise protocol.handshake_exc

Else, the WebSocket connection is open.

A WebSocket server API usually builds a wrapper around the network socket and the ServerProtocol. Then it invokes a connection handler that accepts the wrapper in argument.

It may also provide a way to close all connections and to shut down the server gracefully.

Going forwards, this guide focuses on handling an individual connection.

Go through the five steps below until you reach the end of the data stream.

When receiving data from the network, feed it to the protocol's receive_data() method.

When reaching the end of the data stream, call the protocol's receive_eof() method.

For example, if sock is a socket:

try:
    data = sock.recv(65536)
except OSError:  # socket closed
    data = b""
if data:
    protocol.receive_data(data)
else:
    protocol.receive_eof()

These methods aren't expected to raise exceptions — unless you call them again after calling receive_eof(), which is an error. (If you get an exception, please file a bug!)

Then, call data_to_send() and send its output to the network:

for data in protocol.data_to_send():
    if data:
        sock.sendall(data)
    else:
        sock.shutdown(socket.SHUT_WR)

The empty bytestring signals the end of the data stream. When you see it, you must half-close the TCP connection.

Sending data right after receiving data is necessary because websockets responds to ping frames, close frames, and incorrect inputs automatically.

Closing a WebSocket connection normally involves a two-way WebSocket closing handshake. Then, regardless of whether the closure is normal or abnormal, the server starts the four-way TCP closing handshake. If the network fails at the wrong point, you can end up waiting until the TCP timeout, which is very long.

To prevent dangling TCP connections when you expect the end of the data stream but you never reach it, call close_expected() and, if it returns True, schedule closing the TCP connection after a short timeout:

# start a new execution thread to run this code
sleep(10)
sock.close()  # does nothing if the socket is already closed

If the connection is still open when the timeout elapses, closing the socket makes the execution thread that reads from the socket reach the end of the data stream, possibly with an exception.

If you called receive_eof(), close the TCP connection now. This is a clean closure because the receive buffer is empty.

After receive_eof() signals the end of the read stream, data_to_send() always signals the end of the write stream, unless it already ended. So, at this point, the TCP connection is already half-closed. The only reason for closing it now is to release resources related to the socket.

Now you can exit the loop relaying data from the network to the application.

Finally, call events_received() to obtain events parsed from the data provided to receive_data():

events = connection.events_received()

The first event will be the WebSocket opening handshake request or response. See Opening a connection above for details.

All later events are WebSocket frames. There are two types of frames:

  • Data frames contain messages transferred over the WebSocket connections. You should provide them to the application. See Fragmentation below for how to reassemble messages from frames.
  • Control frames provide information about the connection's state. The main use case is to expose an abstraction over ping and pong to the application. Keep in mind that websockets responds to ping frames and close frames automatically. Don't duplicate this functionality!

The connection object provides one method for each type of WebSocket frame.

For sending a data frame:

  • send_continuation()
  • send_text()
  • send_binary()

These methods raise ProtocolError if you don't set the FIN bit correctly in fragmented messages.

For sending a control frame:

  • send_close()
  • send_ping()
  • send_pong()

send_close() initiates the closing handshake. See Closing a connection below for details.

If you encounter an unrecoverable error and you must fail the WebSocket connection, call fail().

After any of the above, call data_to_send() and send its output to the network, as shown in Send data above.

If you called send_close() or fail(), you expect the end of the data stream. You should follow the process described in Close TCP connection above in order to prevent dangling TCP connections.

Under normal circumstances, when a server wants to close the TCP connection:

  • it closes the write side;
  • it reads until the end of the stream, because it expects the client to close the read side;
  • it closes the socket.

When a client wants to close the TCP connection:

  • it reads until the end of the stream, because it expects the server to close the read side;
  • it closes the write side;
  • it closes the socket.

Applying the rules described earlier in this document gives the intended result. As a reminder, the rules are:

  • When data_to_send() returns the empty bytestring, close the write side of the TCP connection.
  • When you reach the end of the read stream, close the TCP connection.
  • When close_expected() returns True, if you don't reach the end of the read stream quickly, close the TCP connection.

WebSocket messages may be fragmented. Since this is a protocol-level concern, you may choose to reassemble fragmented messages before handing them over to the application.

To reassemble a message, read data frames until you get a frame where the FIN bit is set, then concatenate the payloads of all frames.

You will never receive an inconsistent sequence of frames because websockets raises a ProtocolError and fails the connection when this happens. However, you may receive an incomplete sequence if the connection drops in the middle of a fragmented message.

The Sans-I/O layer expects to run sequentially. If your interact with it from multiple threads or coroutines, you must ensure correct serialization. This should happen automatically in a cooperative multitasking environment.

However, you still have to make sure you don't break this property by accident. For example, serialize writes to the network when data_to_send() returns multiple values to prevent concurrent writes from interleaving incorrectly.

The Sans-I/O layer doesn't do any buffering. It makes events available in events_received() as soon as they're received.

You should make incoming messages available to the application immediately and stop further processing until the application fetches them. This will usually result in the best performance.

about asyncio."

Python's documentation about developing with asyncio is a good complement.

Your connection handler exits prematurely. Wait for the work to be finished before returning.

For example, if your handler has a structure similar to:

async def handler(websocket):
    asyncio.create_task(do_some_work())

change it to:

async def handler(websocket):
    await do_some_work()

Your connection handler exits after processing one message. Write a loop to process multiple messages.

For example, if your handler looks like this:

async def handler(websocket):
    print(websocket.recv())

change it like this:

async def handler(websocket):
    async for message in websocket:
        print(message)

Don't feel bad if this happens to you — it's the most common question in websockets' issue tracker :-)

Your connection handler blocks the event loop. Look for blocking calls.

Any call that may take some time must be asynchronous.

For example, this connection handler prevents the event loop from running during one second:

async def handler(websocket):
    time.sleep(1)
    ...

Change it to:

async def handler(websocket):
    await asyncio.sleep(1)
    ...

In addition, calling a coroutine doesn't guarantee that it will yield control to the event loop.

For example, this connection handler blocks the event loop by sending messages continuously:

async def handler(websocket):
    while True:
        await websocket.send("firehose!")

send() completes synchronously as long as there's space in send buffers. The event loop never runs. (This pattern is uncommon in real-world applications. It occurs mostly in toy programs.)

You can avoid the issue by yielding control to the event loop explicitly:

async def handler(websocket):
    while True:
        await websocket.send("firehose!")
        await asyncio.sleep(0)

All this is part of learning asyncio. It isn't specific to websockets.

See also Python's documentation about running blocking code.

Record all connections in a global variable:

CONNECTIONS = set()
async def handler(websocket):
    CONNECTIONS.add(websocket)
    try:
        await websocket.wait_closed()
    finally:
        CONNECTIONS.remove(websocket)

Then, call broadcast():

import websockets
def message_all(message):
    websockets.broadcast(CONNECTIONS, message)

If you're running multiple server processes, make sure you call message_all in each process.

Record connections in a global variable, keyed by user identifier:

CONNECTIONS = {}
async def handler(websocket):
    user_id = ...  # identify user in your app's context
    CONNECTIONS[user_id] = websocket
    try:
        await websocket.wait_closed()
    finally:
        del CONNECTIONS[user_id]

Then, call send():

async def message_user(user_id, message):
    websocket = CONNECTIONS[user_id]  # raises KeyError if user disconnected
    await websocket.send(message)  # may raise websockets.ConnectionClosed

Add error handling according to the behavior you want if the user disconnected before the message could be sent.

This example supports only one connection per user. To support concurrent connections by the same user, you can change CONNECTIONS to store a set of connections for each user.

If you're running multiple server processes, call message_user in each process. The process managing the user's connection sends the message; other processes do nothing.

When you reach a scale where server processes cannot keep up with the stream of all messages, you need a better architecture. For example, you could deploy an external publish / subscribe system such as Redis. Server processes would subscribe their clients. Then, they would receive messages only for the connections that they're managing.

websockets doesn't provide built-in publish / subscribe functionality.

Record connections in a global variable, keyed by user identifier, as shown in How do I send a message to a single user?

Then, build the set of recipients and broadcast the message to them, as shown in How do I send a message to all users?

Integrate with Django contains a complete implementation of this pattern.

Again, as you scale, you may reach the performance limits of a basic in-process implementation. You may need an external publish / subscribe system like Redis.

You can bind additional arguments to the connection handler with functools.partial():

import asyncio
import functools
import websockets
async def handler(websocket, extra_argument):
    ...
bound_handler = functools.partial(handler, extra_argument=42)
start_server = websockets.serve(bound_handler, ...)

Another way to achieve this result is to define the handler coroutine in a scope where the extra_argument variable exists instead of injecting it through an argument.

It is available in the path attribute.

You may route a connection to different handlers depending on the request path:

async def handler(websocket):
    if websocket.path == "/blue":
        await blue_handler(websocket)
    elif websocket.path == "/green":
        await green_handler(websocket)
    else:
        # No handler for this path; close the connection.
        return

You may also route the connection based on the first message received from the client, as shown in the tutorial. When you want to authenticate the connection before routing it, this is usually more convenient.

Generally speaking, there is far less emphasis on the request path in WebSocket servers than in HTTP servers. When a WebSocket server provides a single endpoint, it may ignore the request path entirely.

How do I access HTTP headers?

To access HTTP headers during the WebSocket handshake, you can override process_request:

async def process_request(self, path, request_headers):
    authorization = request_headers["Authorization"]

Once the connection is established, HTTP headers are available in request_headers and response_headers:

async def handler(websocket):
    authorization = websocket.request_headers["Authorization"]

How do I set HTTP headers?

To set the Sec-WebSocket-Extensions or Sec-WebSocket-Protocol headers in the WebSocket handshake response, use the extensions or subprotocols arguments of serve().

To override the Server header, use the server_header argument. Set it to None to remove the header.

To set other HTTP headers, use the extra_headers argument.

It's available in remote_address:

async def handler(websocket):
    remote_ip = websocket.remote_address[0]

Use the host argument of create_server():

await websockets.serve(handler, host="192.168.0.1", port=8080)

serve() accepts the same arguments as create_server().

What does OSError: [Errno 99] error while attempting to bind on address ('::1', 80, 0, 0): address not available mean?

You are calling serve() without a host argument in a context where IPv6 isn't available.

To listen only on IPv4, specify host="0.0.0.0" or family=socket.AF_INET.

Refer to the documentation of create_server() for details.

How do I close a connection?

websockets takes care of closing the connection when the handler exits.

Exit the serve() context manager.

Here's an example that terminates cleanly when it receives SIGTERM on Unix:

#!/usr/bin/env python
import asyncio
import signal
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def server():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(echo, "localhost", 8765):
        await stop
asyncio.run(server())

Call the server's close() method with close_connections=False.

Here's how to adapt the example just above:

async def server():
    ...
    server = await websockets.serve(echo, "localhost", 8765)
    await stop
    await server.close(close_connections=False)

How do I implement a health check?

Intercept WebSocket handshake requests with the process_request() hook.

When a request is sent to the health check endpoint, treat is as an HTTP request and return a (status, headers, body) tuple, as in this example:

#!/usr/bin/env python
import asyncio
import http
import websockets
async def health_check(path, request_headers):
    if path == "/healthz":
        return http.HTTPStatus.OK, [], b"OK\n"
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def main():
    async with websockets.serve(
        echo, "localhost", 8765,
        process_request=health_check,
    ):
        await asyncio.Future()  # run forever
asyncio.run(main())

You don't.

HTTP and WebSocket have widely different operational characteristics. Running them with the same server becomes inconvenient when you scale.

Providing an HTTP server is out of scope for websockets. It only aims at providing a WebSocket server.

There's limited support for returning HTTP responses with the process_request hook.

If you need more, pick an HTTP server and run it separately.

Alternatively, pick an HTTP framework that builds on top of websockets to support WebSocket connections, like Sanic.

You're exiting the context manager prematurely. Wait for the work to be finished before exiting.

For example, if your code has a structure similar to:

async with connect(...) as websocket:
    asyncio.create_task(do_some_work())

change it to:

async with connect(...) as websocket:
    await do_some_work()

How do I access HTTP headers?

Once the connection is established, HTTP headers are available in request_headers and response_headers.

How do I set HTTP headers?

To set the Origin, Sec-WebSocket-Extensions, or Sec-WebSocket-Protocol headers in the WebSocket handshake request, use the origin, extensions, or subprotocols arguments of connect().

To override the User-Agent header, use the user_agent_header argument. Set it to None to remove the header.

To set other HTTP headers, for example the Authorization header, use the extra_headers argument:

async with connect(..., extra_headers={"Authorization": ...}) as websocket:
    ...

In the threading API, this argument is named additional_headers:

with connect(..., additional_headers={"Authorization": ...}) as websocket:
    ...

Use the host argument of create_connection():

await websockets.connect("ws://example.com", host="192.168.0.1")

connect() accepts the same arguments as create_connection().

How do I close a connection?

The easiest is to use connect() as a context manager:

async with connect(...) as websocket:
    ...

The connection is closed when exiting the context manager.

Use connect() as an asynchronous iterator:

async for websocket in websockets.connect(...):
    try:
        ...
    except websockets.ConnectionClosed:
        continue

Make sure you handle exceptions in the async for loop. Uncaught exceptions will break out of the loop.

You can close the connection.

Here's an example that terminates cleanly when it receives SIGTERM on Unix:

#!/usr/bin/env python
import asyncio
import signal
import websockets
async def client():
    uri = "ws://localhost:8765"
    async with websockets.connect(uri) as websocket:
        # Close the connection when receiving SIGTERM.
        loop = asyncio.get_running_loop()
        loop.add_signal_handler(
            signal.SIGTERM, loop.create_task, websocket.close())
        # Process messages received on the connection.
        async for message in websocket:
            ...
asyncio.run(client())

Look at the ssl argument of create_connection().

connect() accepts the same arguments as create_connection().

Both sides

What does ConnectionClosedError: no close frame received or sent mean?

If you're seeing this traceback in the logs of a server:

connection handler failed
Traceback (most recent call last):
  ...
asyncio.exceptions.IncompleteReadError: 0 bytes read on a total of 2 expected bytes
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
  ...
websockets.exceptions.ConnectionClosedError: no close frame received or sent

or if a client crashes with this traceback:

Traceback (most recent call last):
  ...
ConnectionResetError: [Errno 54] Connection reset by peer
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
  ...
websockets.exceptions.ConnectionClosedError: no close frame received or sent

it means that the TCP connection was lost. As a consequence, the WebSocket connection was closed without receiving and sending a close frame, which is abnormal.

You can catch and handle ConnectionClosed to prevent it from being logged.

There are several reasons why long-lived connections may be lost:

  • End-user devices tend to lose network connectivity often and unpredictably because they can move out of wireless network coverage, get unplugged from a wired network, enter airplane mode, be put to sleep, etc.
  • HTTP load balancers or proxies that aren't configured for long-lived connections may terminate connections after a short amount of time, usually 30 seconds, despite websockets' keepalive mechanism.

If you're facing a reproducible issue, enable debug logs to see when and how connections are closed.

What does ConnectionClosedError: sent 1011 (internal error) keepalive ping timeout; no close frame received mean?

If you're seeing this traceback in the logs of a server:

connection handler failed
Traceback (most recent call last):
  ...
asyncio.exceptions.CancelledError
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
  ...
websockets.exceptions.ConnectionClosedError: sent 1011 (internal error) keepalive ping timeout; no close frame received

or if a client crashes with this traceback:

Traceback (most recent call last):
  ...
asyncio.exceptions.CancelledError
The above exception was the direct cause of the following exception:
Traceback (most recent call last):
  ...
websockets.exceptions.ConnectionClosedError: sent 1011 (internal error) keepalive ping timeout; no close frame received

it means that the WebSocket connection suffered from excessive latency and was closed after reaching the timeout of websockets' keepalive mechanism.

You can catch and handle ConnectionClosed to prevent it from being logged.

There are two main reasons why latency may increase:

  • Poor network connectivity.
  • More traffic than the recipient can handle.

See the discussion of timeouts for details.

If websockets' default timeout of 20 seconds is too short for your use case, you can adjust it with the ping_timeout argument.

On Python ≥ 3.11, use asyncio.timeout():

async with asyncio.timeout(timeout=10):
    message = await websocket.recv()

On older versions of Python, use asyncio.wait_for():

message = await asyncio.wait_for(websocket.recv(), timeout=10)

This technique works for most APIs. When it doesn't, for example with asynchronous context managers, websockets provides an open_timeout argument.

You can bind additional arguments to the protocol factory with functools.partial():

import asyncio
import functools
import websockets
class MyServerProtocol(websockets.WebSocketServerProtocol):
    def __init__(self, *args, extra_argument=None, **kwargs):
        super().__init__(*args, **kwargs)
        # do something with extra_argument
create_protocol = functools.partial(MyServerProtocol, extra_argument=42)
start_server = websockets.serve(..., create_protocol=create_protocol)

This example was for a server. The same pattern applies on a client.

websockets sends pings at 20 seconds intervals to keep the connection open.

It closes the connection if it doesn't get a pong within 20 seconds.

You can adjust this behavior with ping_interval and ping_timeout.

See Timeouts for details.

If you are referring to Ping and Pong frames defined in the WebSocket protocol, don't bother, because websockets handles them for you.

If you are connecting to a server that defines its own heartbeat at the application level, then you need to build that logic into your application.

You must start two tasks, which the event loop will run concurrently. You can achieve this with asyncio.gather() or asyncio.create_task().

Keep track of the tasks and make sure they terminate or you cancel them when the connection terminates.

Your program runs a coroutine that never yields control to the event loop. The coroutine that receives messages never gets a chance to run.

Putting an await statement in a for or a while loop isn't enough to yield control. Awaiting a coroutine may yield control, but there's no guarantee that it will.

For example, send() only yields control when send buffers are full, which never happens in most practical cases.

If you run a loop that contains only synchronous operations and a send() call, you must yield control explicitly with asyncio.sleep():

async def producer(websocket):
    message = generate_next_message()
    await websocket.send(message)
    await asyncio.sleep(0)  # yield control to the event loop

asyncio.sleep() always suspends the current task, allowing other tasks to run. This behavior is documented precisely because it isn't expected from every coroutine.

See issue 867.

If you choose websockets' default implementation based on asyncio, then you shouldn't use threads. Indeed, choosing asyncio to handle concurrency is mutually exclusive with threading.

If you believe that you need to run websockets in a thread and some logic in another thread, you should run that logic in a Task instead. If it blocks the event loop, run_in_executor() will help.

This question is really about asyncio. Please review the advice about Concurrency and Multithreading in the Python documentation.

You are using time.sleep() instead of asyncio.sleep(), which blocks the event loop and prevents asyncio from operating normally.

This may lead to messages getting send but not received, to connection timeouts, and to unexpected results of shotgun debugging e.g. adding an unnecessary call to send() makes the program functional.

Often, this is because you created a script called websockets.py in your current working directory. Then import websockets imports this module instead of the websockets library.

This is an artifact of websockets' history. For its first eight years, only the asyncio implementation existed. Then, the Sans-I/O implementation was added. Moving the code in a legacy submodule eased this refactoring and optimized maintainability.

All public APIs were kept at their original locations. websockets.legacy isn't a public API. It's only visible in the source code and in stack traces. There is no intent to deprecate this implementation — at least until a superior alternative exists.

Not all libraries are as feature-complete as websockets. For a fair benchmark, you should disable features that the other library doesn't provide. Typically, you may need to disable:

  • Compression: set compression=None
  • Keepalive: set ping_interval=None
  • UTF-8 decoding: send bytes rather than str

If websockets is still slower than another Python library, please file a bug.

No, there aren't.

websockets provides high-level, coroutine-based APIs. Compared to callbacks, coroutines make it easier to manage control flow in concurrent code.

If you prefer callback-based APIs, you should use another library.

Check which implementations support which features and known limitations.

Feature support matrices summarize which implementations support which features.

Both sides

asyncio threading Sans-I/O
Perform the opening handshake ✅ ✅ ✅
Send a message ✅ ✅ ✅
Receive a message ✅ ✅ ✅
Iterate over received messages ✅ ✅ ❌
Send a fragmented message ✅ ✅ ✅
Receive a fragmented message after reassembly ✅ ✅ ❌
Receive a fragmented message frame by frame (#479) ❌ ✅ ✅
Send a ping ✅ ✅ ✅
Respond to pings automatically ✅ ✅ ✅
Send a pong ✅ ✅ ✅
Perform the closing handshake ✅ ✅ ✅
Report close codes and reasons from both sides ❌ ✅ ✅
Compress messages (RFC 7692) ✅ ✅ ✅
Tune memory usage for compression ✅ ✅ ✅
Negotiate extensions ✅ ✅ ✅
Implement custom extensions ✅ ✅ ✅
Negotiate a subprotocol ✅ ✅ ✅
Enforce security limits ✅ ✅ ✅
Log events ✅ ✅ ✅
Enforce opening timeout ✅ ✅ —
Enforce closing timeout ✅ ✅ —
Keepalive ✅ ❌ —
Heartbeat ✅ ❌ —

asyncio threading Sans-I/O
Listen on a TCP socket ✅ ✅ —
Listen on a Unix socket ✅ ✅ —
Listen using a preexisting socket ✅ ✅ —
Encrypt connection with TLS ✅ ✅ —
Close server on context exit ✅ ✅ —
Close connection on handler exit ✅ ✅ —
Shut down server gracefully ✅ ✅ —
Check Origin header ✅ ✅ ✅
Customize subprotocol selection ✅ ✅ ✅
Configure Server header ✅ ✅ ✅
Alter opening handshake request ❌ ✅ ✅
Alter opening handshake response ❌ ✅ ✅
Perform HTTP Basic Authentication ✅ ❌ ❌
Perform HTTP Digest Authentication ❌ ❌ ❌
Force HTTP response ✅ ✅ ✅

asyncio threading Sans-I/O
Connect to a TCP socket ✅ ✅ —
Connect to a Unix socket ✅ ✅ —
Connect using a preexisting socket ✅ ✅ —
Encrypt connection with TLS ✅ ✅ —
Close connection on context exit ✅ ✅ —
Reconnect automatically ✅ ❌ —
Configure Origin header ✅ ✅ ✅
Configure User-Agent header ✅ ✅ ✅
Alter opening handshake request ✅ ✅ ✅
Connect to non-ASCII IRIs ✅ ✅ ✅
Perform HTTP Basic Authentication ✅ ✅ ✅
Perform HTTP Digest Authentication (#784) ❌ ❌ ❌
Follow HTTP redirects ✅ ❌ —
Connect via a HTTP proxy (#364) ❌ ❌ —
Connect via a SOCKS5 proxy (#475) ❌ ❌ —

There is no way to control compression of outgoing frames on a per-frame basis (#538). If compression is enabled, all frames are compressed.

The server doesn't check the Host header and respond with a HTTP 400 Bad Request if it is missing or invalid (#1246).

The client API doesn't attempt to guarantee that there is no more than one connection to a given IP address in a CONNECTING state. This behavior is mandated by RFC 6455. However, connect() isn't the right layer for enforcing this constraint. It's the caller's responsibility.

asyncio

This is the default implementation. It's ideal for servers that handle many clients concurrently.

Server (asyncio)

Start a WebSocket server listening on host and port.

Whenever a client connects, the server creates a WebSocketServerProtocol, performs the opening handshake, and delegates to the connection handler, ws_handler.

The handler receives the WebSocketServerProtocol and uses it to send and receive messages.

Once the handler completes, either normally or with an exception, the server performs the closing handshake and closes the connection.

Awaiting serve() yields a WebSocketServer. This object provides a close() method to shut down the server:

stop = asyncio.Future()  # set this future to exit the server
server = await serve(...)
await stop
await server.close()

serve() can be used as an asynchronous context manager. Then, the server is shut down automatically when exiting the context:

stop = asyncio.Future()  # set this future to exit the server
async with serve(...):
    await stop
  • ws_handler (Union[Callable[[WebSocketServerProtocol], Awaitable[Any]], Callable[[WebSocketServerProtocol, str], Awaitable[Any]]]) -- Connection handler. It receives the WebSocket connection, which is a WebSocketServerProtocol, in argument.
  • host (Optional[Union[str, Sequence[str]]]) -- Network interfaces the server binds to. See create_server() for details.
  • port (Optional[int]) -- TCP port the server listens on. See create_server() for details.
  • create_protocol (Optional[Callable[..., WebSocketServerProtocol]]) -- Factory for the asyncio.Protocol managing the connection. It defaults to WebSocketServerProtocol. Set it to a wrapper or a subclass to customize connection handling.
  • logger (Optional[LoggerLike]) -- Logger for this server. It defaults to logging.getLogger("websockets.server"). See the logging guide for details.
  • compression (Optional[str]) -- The "permessage-deflate" extension is enabled by default. Set compression to None to disable it. See the compression guide for details.
  • origins (Optional[Sequence[Optional[Origin]]]) -- Acceptable values of the Origin header, for defending against Cross-Site WebSocket Hijacking attacks. Include None in the list if the lack of an origin is acceptable.
  • extensions (Optional[Sequence[ServerExtensionFactory]]) -- List of supported extensions, in order in which they should be negotiated and run.
  • subprotocols (Optional[Sequence[Subprotocol]]) -- List of supported subprotocols, in order of decreasing preference.
  • extra_headers (Union[HeadersLike, Callable[[str, Headers], HeadersLike]]) -- Arbitrary HTTP headers to add to the response. This can be a HeadersLike or a callable taking the request path and headers in arguments and returning a HeadersLike.
  • server_header (Optional[str]) -- Value of the Server response header. It defaults to "Python/x.y.z websockets/X.Y". Setting it to None removes the header.
  • process_request (Optional[Callable[[str, Headers], Awaitable[Optional[Tuple[StatusLike, HeadersLike, bytes]]]]]) -- Intercept HTTP request before the opening handshake. See process_request() for details.
  • select_subprotocol (Optional[Callable[[Sequence[Subprotocol], Sequence[Subprotocol]], Subprotocol]]) -- Select a subprotocol supported by the client. See select_subprotocol() for details.
  • open_timeout (Optional[float]) -- Timeout for opening connections in seconds. None disables the timeout.

See WebSocketCommonProtocol for the documentation of ping_interval, ping_timeout, close_timeout, max_size, max_queue, read_limit, and write_limit.

Any other keyword arguments are passed the event loop's create_server() method.

For example:

  • You can set ssl to a SSLContext to enable TLS.
  • You can set sock to a socket that you created outside of websockets.
WebSocket server.
WebSocketServer
Start a WebSocket server listening on a Unix socket.

This function is identical to serve(), except the host and port arguments are replaced by path. It is only available on Unix.

Unrecognized keyword arguments are passed the event loop's create_unix_server() method.

It's useful for deploying a server behind a reverse proxy such as nginx.

path (str | None) -- File system path to the Unix socket.

WebSocket server returned by serve().

This class provides the same interface as Server, notably the close() and wait_closed() methods.

It keeps track of WebSocket connections in order to close them properly when shutting down.

logger (Optional[LoggerLike]) -- Logger for this server. It defaults to logging.getLogger("websockets.server"). See the logging guide for details.
Close the server.
  • Close the underlying Server.
  • When close_connections is True, which is the default, close existing connections. Specifically:
  • Reject opening WebSocket connections with an HTTP 503 (service unavailable) error. This happens when the server accepted the TCP connection but didn't complete the opening handshake before closing.
  • Close open WebSocket connections with close code 1001 (going away).
•
Wait until all connection handlers terminate.

close() is idempotent.


Wait until the server is closed.

When wait_closed() returns, all TCP connections are closed and all connection handlers have returned.

To ensure a fast shutdown, a connection handler should always be awaiting at least one of:

  • recv(): when the connection is closed, it raises ConnectionClosedOK;
  • wait_closed(): when the connection is closed, it returns.

Then the connection handler is immediately notified of the shutdown; it can clean up and exit.


See asyncio.Server.get_loop().

See asyncio.Server.is_serving().

See asyncio.Server.start_serving().

Typical use:

server = await serve(..., start_serving=False)
# perform additional setup here...
# ... then start the server
await server.start_serving()

See asyncio.Server.serve_forever().

Typical use:

server = await serve(...)
# this coroutine doesn't return
# canceling it stops the server
await server.serve_forever()

This is an alternative to using serve() as an asynchronous context manager. Shutdown is triggered by canceling serve_forever() instead of exiting a serve() context.


See asyncio.Server.sockets.

Using a connection

WebSocket server connection.

WebSocketServerProtocol provides recv() and send() coroutines for receiving and sending messages.

It supports asynchronous iteration to receive messages:

async for message in websocket:
    await process(message)

The iterator exits normally when the connection is closed with close code 1000 (OK) or 1001 (going away) or without a close code. It raises a ConnectionClosedError when the connection is closed with any other code.

You may customize the opening handshake in a subclass by overriding process_request() or select_subprotocol().

ws_server (WebSocketServer) -- WebSocket server that created this connection.

See serve() for the documentation of ws_handler, logger, origins, extensions, subprotocols, extra_headers, and server_header.

See WebSocketCommonProtocol for the documentation of ping_interval, ping_timeout, close_timeout, max_size, max_queue, read_limit, and write_limit.

Receive the next message.

When the connection is closed, recv() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal connection closure and ConnectionClosedError after a protocol error or a network failure. This is how you detect the end of the message stream.

Canceling recv() is safe. There's no risk of losing the next message. The next invocation of recv() will return it.

This makes it possible to enforce a timeout by wrapping recv() in timeout() or wait_for().

A string (str) for a Text frame. A bytestring (bytes) for a Binary frame.
Data
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If two coroutines call recv() concurrently.
Send a message.

A string (str) is sent as a Text frame. A bytestring or bytes-like object (bytes, bytearray, or memoryview) is sent as a Binary frame.

send() also accepts an iterable or an asynchronous iterable of strings, bytestrings, or bytes-like objects to enable fragmentation. Each item is treated as a message fragment and sent in its own frame. All items must be of the same type, or else send() will raise a TypeError and the connection will be closed.

send() rejects dict-like objects because this is often an error. (If you want to send the keys of a dict-like object as fragments, call its keys() method and pass the result to send().)

Canceling send() is discouraged. Instead, you should close the connection with close(). Indeed, there are only two situations where send() may yield control to the event loop and then get canceled; in both cases, close() has the same effect and is more clear:

1.
The write buffer is full. If you don't want to wait until enough data is sent, your only alternative is to close the connection. close() will likely time out then abort the TCP connection.
2.
message is an asynchronous iterator that yields control. Stopping in the middle of a fragmented message will cause a protocol error and the connection will be closed.

When the connection is closed, send() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal connection closure and ConnectionClosedError after a protocol error or a network failure.

message (Union[Data, Iterable[Data], AsyncIterable[Data]) -- message to send.
  • ConnectionClosed -- When the connection is closed.
  • TypeError -- If message doesn't have a supported type.
Perform the closing handshake.

close() waits for the other end to complete the handshake and for the TCP connection to terminate. As a consequence, there's no need to await wait_closed() after close().

close() is idempotent: it doesn't do anything once the connection is closed.

Wrapping close() in create_task() is safe, given that errors during connection termination aren't particularly useful.

Canceling close() is discouraged. If it takes too long, you can set a shorter close_timeout. If you don't want to wait, let the Python process exit, then the OS will take care of closing the TCP connection.

  • code (int) -- WebSocket close code.
  • reason (str) -- WebSocket close reason.
Wait until the connection is closed.

This coroutine is identical to the closed attribute, except it can be awaited.

This can make it easier to detect connection termination, regardless of its cause, in tasks that interact with the WebSocket connection.


Send a Ping.

A ping may serve as a keepalive, as a check that the remote endpoint received all messages up to this point, or to measure latency.

Canceling ping() is discouraged. If ping() doesn't return immediately, it means the write buffer is full. If you don't want to wait, you should close the connection.

Canceling the Future returned by ping() has no effect.

data (Optional[Data]) -- payload of the ping; a string will be encoded to UTF-8; or None to generate a payload containing four random bytes.
A future that will be completed when the corresponding pong is received. You can ignore it if you don't intend to wait. The result of the future is the latency of the connection in seconds.
pong_waiter = await ws.ping()
# only if you want to wait for the corresponding pong
latency = await pong_waiter
Future[float]
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If another ping was sent with the same data and
    the corresponding pong wasn't received yet.
Send a Pong.

An unsolicited pong may serve as a unidirectional heartbeat.

Canceling pong() is discouraged. If pong() doesn't return immediately, it means the write buffer is full. If you don't want to wait, you should close the connection.

data (Data) -- Payload of the pong. A string will be encoded to UTF-8.
ConnectionClosed -- When the connection is closed.

You can customize the opening handshake in a subclass by overriding these methods:

Intercept the HTTP request and return an HTTP response if appropriate.

You may override this method in a WebSocketServerProtocol subclass, for example:

  • to return an HTTP 200 OK response on a given path; then a load balancer can use this path for a health check;
  • to authenticate the request and return an HTTP 401 Unauthorized or an HTTP 403 Forbidden when authentication fails.

You may also override this method with the process_request argument of serve() and WebSocketServerProtocol. This is equivalent, except process_request won't have access to the protocol instance, so it can't store information for later use.

process_request() is expected to complete quickly. If it may run for a long time, then it should await wait_closed() and exit if wait_closed() completes, or else it could prevent the server from shutting down.

  • path (str) -- request path, including optional query string.
  • request_headers (Headers) -- request headers.
None to continue the WebSocket handshake normally.

An HTTP response, represented by a 3-uple of the response status, headers, and body, to abort the WebSocket handshake and return that HTTP response instead.

Optional[Tuple[StatusLike, HeadersLike, bytes]]
Pick a subprotocol among those supported by the client and the server.

If several subprotocols are available, select the preferred subprotocol by giving equal weight to the preferences of the client and the server.

If no subprotocol is available, proceed without a subprotocol.

You may provide a select_subprotocol argument to serve() or WebSocketServerProtocol to override this logic. For example, you could reject the handshake if the client doesn't support a particular subprotocol, rather than accept the handshake without that subprotocol.

  • client_subprotocols (Sequence[Subprotocol]) -- list of subprotocols offered by the client.
  • server_subprotocols (Sequence[Subprotocol]) -- list of subprotocols available on the server.
Selected subprotocol, if a common subprotocol was found.

None to continue without a subprotocol.

Optional[Subprotocol]

WebSocket connection objects also provide these attributes:

Unique identifier of the connection. Useful in logs.
Logger for this connection.
Local address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family; see getsockname().

None if the TCP connection isn't established yet.

Remote address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family; see getpeername().

None if the TCP connection isn't established yet.

True when the connection is open; False otherwise.

This attribute may be used to detect disconnections. However, this approach is discouraged per the EAFP principle. Instead, you should handle ConnectionClosed exceptions.

True when the connection is closed; False otherwise.

Be aware that both open and closed are False during the opening and closing sequences.

Latency of the connection, in seconds.

This value is updated after sending a ping frame and receiving a matching pong frame. Before the first ping, latency is 0.

By default, websockets enables a keepalive mechanism that sends ping frames automatically at regular intervals. You can also send ping frames and measure latency with ping().

The following attributes are available after the opening handshake, once the WebSocket connection is open:

Path of the opening handshake request.
Opening handshake request headers.
Opening handshake response headers.
Subprotocol, if one was negotiated.

The following attributes are available after the closing handshake, once the WebSocket connection is closed:

WebSocket close code, defined in section 7.1.5 of RFC 6455.

None if the connection isn't closed yet.

WebSocket close reason, defined in section 7.1.6 of RFC 6455.

None if the connection isn't closed yet.

websockets supports HTTP Basic Authentication according to RFC 7235 and RFC 7617.

Protocol factory that enforces HTTP Basic Auth.

basic_auth_protocol_factory() is designed to integrate with serve() like this:

websockets.serve(
    ...,
    create_protocol=websockets.basic_auth_protocol_factory(
        realm="my dev server",
        credentials=("hello", "iloveyou"),
    )
)
  • realm (str | None) -- Scope of protection. It should contain only ASCII characters because the encoding of non-ASCII characters is undefined. Refer to section 2.2 of RFC 7235 for details.
  • credentials (Tuple[str, str] | Iterable[Tuple[str, str]] | None) -- Hard coded authorized credentials. It can be a (username, password) pair or a list of such pairs.
  • check_credentials (Callable[[str, str], Awaitable[bool]] | None) -- Coroutine that verifies credentials. It receives username and password arguments and returns a bool. One of credentials or check_credentials must be provided but not both.
  • create_protocol (Callable[[...], BasicAuthWebSocketServerProtocol] | None) -- Factory that creates the protocol. By default, this is BasicAuthWebSocketServerProtocol. It can be replaced by a subclass.
TypeError -- If the credentials or check_credentials argument is
wrong.
WebSocket server protocol that enforces HTTP Basic Auth.

Scope of protection.

If provided, it should contain only ASCII characters because the encoding of non-ASCII characters is undefined.

Username of the authenticated user.
Check whether credentials are authorized.

This coroutine may be overridden in a subclass, for example to authenticate against a database or an external service.

  • username (str) -- HTTP Basic Auth username.
  • password (str) -- HTTP Basic Auth password.
True if the handshake should continue; False if it should fail with an HTTP 401 error.
bool

Broadcast a message to several WebSocket connections.

A string (str) is sent as a Text frame. A bytestring or bytes-like object (bytes, bytearray, or memoryview) is sent as a Binary frame.

broadcast() pushes the message synchronously to all connections even if their write buffers are overflowing. There's no backpressure.

If you broadcast messages faster than a connection can handle them, messages will pile up in its write buffer until the connection times out. Keep ping_interval and ping_timeout low to prevent excessive memory usage from slow connections.

Unlike send(), broadcast() doesn't support sending fragmented messages. Indeed, fragmentation is useful for sending large messages without buffering them in memory, while broadcast() buffers one copy per connection as fast as possible.

broadcast() skips connections that aren't open in order to avoid errors on connections where the closing handshake is in progress.

broadcast() ignores failures to write the message on some connections. It continues writing to other connections. On Python 3.11 and above, you may set raise_exceptions to True to record failures and raise all exceptions in a PEP 654 ExceptionGroup.

  • websockets (Iterable[WebSocketCommonProtocol]) -- WebSocket connections to which the message will be sent.
  • message (str | bytes) -- Message to send.
  • raise_exceptions (bool) -- Whether to raise an exception in case of failures.
TypeError -- If message doesn't have a supported type.

Client (asyncio)

Opening a connection

Connect to the WebSocket server at uri.

Awaiting connect() yields a WebSocketClientProtocol which can then be used to send and receive messages.

connect() can be used as a asynchronous context manager:

async with websockets.connect(...) as websocket:
    ...

The connection is closed automatically when exiting the context.

connect() can be used as an infinite asynchronous iterator to reconnect automatically on errors:

async for websocket in websockets.connect(...):
    try:
        ...
    except websockets.ConnectionClosed:
        continue

The connection is closed automatically after each iteration of the loop.

If an error occurs while establishing the connection, connect() retries with exponential backoff. The backoff delay starts at three seconds and increases up to one minute.

If an error occurs in the body of the loop, you can handle the exception and connect() will reconnect with the next iteration; or you can let the exception bubble up and break out of the loop. This lets you decide which errors trigger a reconnection and which errors are fatal.

  • uri (str) -- URI of the WebSocket server.
  • create_protocol (Optional[Callable[..., WebSocketClientProtocol]]) -- Factory for the asyncio.Protocol managing the connection. It defaults to WebSocketClientProtocol. Set it to a wrapper or a subclass to customize connection handling.
  • logger (Optional[LoggerLike]) -- Logger for this client. It defaults to logging.getLogger("websockets.client"). See the logging guide for details.
  • compression (Optional[str]) -- The "permessage-deflate" extension is enabled by default. Set compression to None to disable it. See the compression guide for details.
  • origin (Optional[Origin]) -- Value of the Origin header, for servers that require it.
  • extensions (Optional[Sequence[ClientExtensionFactory]]) -- List of supported extensions, in order in which they should be negotiated and run.
  • subprotocols (Optional[Sequence[Subprotocol]]) -- List of supported subprotocols, in order of decreasing preference.
  • extra_headers (Optional[HeadersLike]) -- Arbitrary HTTP headers to add to the handshake request.
  • user_agent_header (Optional[str]) -- Value of the User-Agent request header. It defaults to "Python/x.y.z websockets/X.Y". Setting it to None removes the header.
  • open_timeout (Optional[float]) -- Timeout for opening the connection in seconds. None disables the timeout.

See WebSocketCommonProtocol for the documentation of ping_interval, ping_timeout, close_timeout, max_size, max_queue, read_limit, and write_limit.

Any other keyword arguments are passed the event loop's create_connection() method.

For example:

  • You can set ssl to a SSLContext to enforce TLS settings. When connecting to a wss:// URI, if ssl isn't provided, a TLS context is created with create_default_context().
  • You can set host and port to connect to a different host and port from those found in uri. This only changes the destination of the TCP connection. The host name from uri is still used in the TLS handshake for secure connections and in the Host header.
  • InvalidURI -- If uri isn't a valid WebSocket URI.
  • OSError -- If the TCP connection fails.
  • InvalidHandshake -- If the opening handshake fails.
  • TimeoutError -- If the opening handshake times out.
Similar to connect(), but for connecting to a Unix socket.

This function builds upon the event loop's create_unix_connection() method.

It is only available on Unix.

It's mainly useful for debugging servers listening on Unix sockets.

  • path (str | None) -- File system path to the Unix socket.
  • uri (str) -- URI of the WebSocket server; the host is used in the TLS handshake for secure connections and in the Host header.

Using a connection

WebSocket client connection.

WebSocketClientProtocol provides recv() and send() coroutines for receiving and sending messages.

It supports asynchronous iteration to receive incoming messages:

async for message in websocket:
    await process(message)

The iterator exits normally when the connection is closed with close code 1000 (OK) or 1001 (going away) or without a close code. It raises a ConnectionClosedError when the connection is closed with any other code.

See connect() for the documentation of logger, origin, extensions, subprotocols, extra_headers, and user_agent_header.

See WebSocketCommonProtocol for the documentation of ping_interval, ping_timeout, close_timeout, max_size, max_queue, read_limit, and write_limit.


Receive the next message.

When the connection is closed, recv() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal connection closure and ConnectionClosedError after a protocol error or a network failure. This is how you detect the end of the message stream.

Canceling recv() is safe. There's no risk of losing the next message. The next invocation of recv() will return it.

This makes it possible to enforce a timeout by wrapping recv() in timeout() or wait_for().

A string (str) for a Text frame. A bytestring (bytes) for a Binary frame.
Data
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If two coroutines call recv() concurrently.
Send a message.

A string (str) is sent as a Text frame. A bytestring or bytes-like object (bytes, bytearray, or memoryview) is sent as a Binary frame.

send() also accepts an iterable or an asynchronous iterable of strings, bytestrings, or bytes-like objects to enable fragmentation. Each item is treated as a message fragment and sent in its own frame. All items must be of the same type, or else send() will raise a TypeError and the connection will be closed.

send() rejects dict-like objects because this is often an error. (If you want to send the keys of a dict-like object as fragments, call its keys() method and pass the result to send().)

Canceling send() is discouraged. Instead, you should close the connection with close(). Indeed, there are only two situations where send() may yield control to the event loop and then get canceled; in both cases, close() has the same effect and is more clear:

1.
The write buffer is full. If you don't want to wait until enough data is sent, your only alternative is to close the connection. close() will likely time out then abort the TCP connection.
2.
message is an asynchronous iterator that yields control. Stopping in the middle of a fragmented message will cause a protocol error and the connection will be closed.

When the connection is closed, send() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal connection closure and ConnectionClosedError after a protocol error or a network failure.

message (Union[Data, Iterable[Data], AsyncIterable[Data]) -- message to send.
  • ConnectionClosed -- When the connection is closed.
  • TypeError -- If message doesn't have a supported type.
Perform the closing handshake.

close() waits for the other end to complete the handshake and for the TCP connection to terminate. As a consequence, there's no need to await wait_closed() after close().

close() is idempotent: it doesn't do anything once the connection is closed.

Wrapping close() in create_task() is safe, given that errors during connection termination aren't particularly useful.

Canceling close() is discouraged. If it takes too long, you can set a shorter close_timeout. If you don't want to wait, let the Python process exit, then the OS will take care of closing the TCP connection.

  • code (int) -- WebSocket close code.
  • reason (str) -- WebSocket close reason.
Wait until the connection is closed.

This coroutine is identical to the closed attribute, except it can be awaited.

This can make it easier to detect connection termination, regardless of its cause, in tasks that interact with the WebSocket connection.


Send a Ping.

A ping may serve as a keepalive, as a check that the remote endpoint received all messages up to this point, or to measure latency.

Canceling ping() is discouraged. If ping() doesn't return immediately, it means the write buffer is full. If you don't want to wait, you should close the connection.

Canceling the Future returned by ping() has no effect.

data (Optional[Data]) -- payload of the ping; a string will be encoded to UTF-8; or None to generate a payload containing four random bytes.
A future that will be completed when the corresponding pong is received. You can ignore it if you don't intend to wait. The result of the future is the latency of the connection in seconds.
pong_waiter = await ws.ping()
# only if you want to wait for the corresponding pong
latency = await pong_waiter
Future[float]
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If another ping was sent with the same data and
    the corresponding pong wasn't received yet.
Send a Pong.

An unsolicited pong may serve as a unidirectional heartbeat.

Canceling pong() is discouraged. If pong() doesn't return immediately, it means the write buffer is full. If you don't want to wait, you should close the connection.

data (Data) -- Payload of the pong. A string will be encoded to UTF-8.
ConnectionClosed -- When the connection is closed.

WebSocket connection objects also provide these attributes:

Unique identifier of the connection. Useful in logs.
Logger for this connection.
Local address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family; see getsockname().

None if the TCP connection isn't established yet.

Remote address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family; see getpeername().

None if the TCP connection isn't established yet.

True when the connection is open; False otherwise.

This attribute may be used to detect disconnections. However, this approach is discouraged per the EAFP principle. Instead, you should handle ConnectionClosed exceptions.

True when the connection is closed; False otherwise.

Be aware that both open and closed are False during the opening and closing sequences.

Latency of the connection, in seconds.

This value is updated after sending a ping frame and receiving a matching pong frame. Before the first ping, latency is 0.

By default, websockets enables a keepalive mechanism that sends ping frames automatically at regular intervals. You can also send ping frames and measure latency with ping().

The following attributes are available after the opening handshake, once the WebSocket connection is open:

Path of the opening handshake request.
Opening handshake request headers.
Opening handshake response headers.
Subprotocol, if one was negotiated.

The following attributes are available after the closing handshake, once the WebSocket connection is closed:

WebSocket close code, defined in section 7.1.5 of RFC 6455.

None if the connection isn't closed yet.

WebSocket close reason, defined in section 7.1.6 of RFC 6455.

None if the connection isn't closed yet.

threading

This alternative implementation can be a good choice for clients.

Server (threading)

Create a WebSocket server listening on host and port.

Whenever a client connects, the server creates a ServerConnection, performs the opening handshake, and delegates to the handler.

The handler receives a ServerConnection instance, which you can use to send and receive messages.

Once the handler completes, either normally or with an exception, the server performs the closing handshake and closes the connection.

WebSocketServer mirrors the API of BaseServer. Treat it as a context manager to ensure that it will be closed and call the serve_forever() method to serve requests:

def handler(websocket):
    ...
with websockets.sync.server.serve(handler, ...) as server:
    server.serve_forever()
  • handler (Callable[[ServerConnection], None]) -- Connection handler. It receives the WebSocket connection, which is a ServerConnection, in argument.
  • host (str | None) -- Network interfaces the server binds to. See create_server() for details.
  • port (int | None) -- TCP port the server listens on. See create_server() for details.
  • sock (socket | None) -- Preexisting TCP socket. sock replaces host and port. You may call socket.create_server() to create a suitable TCP socket.
  • ssl_context (SSLContext | None) -- Configuration for enabling TLS on the connection.
  • origins (Sequence[Origin | None] | None) -- Acceptable values of the Origin header, for defending against Cross-Site WebSocket Hijacking attacks. Include None in the list if the lack of an origin is acceptable.
  • extensions (Sequence[ServerExtensionFactory] | None) -- List of supported extensions, in order in which they should be negotiated and run.
  • subprotocols (Sequence[Subprotocol] | None) -- List of supported subprotocols, in order of decreasing preference.
  • select_subprotocol (Callable[[ServerConnection, Sequence[Subprotocol]], Subprotocol | None] | None) -- Callback for selecting a subprotocol among those supported by the client and the server. It receives a ServerConnection (not a ServerProtocol!) instance and a list of subprotocols offered by the client. Other than the first argument, it has the same behavior as the ServerProtocol.select_subprotocol method.
  • process_request (Callable[[ServerConnection, Request], Response | None] | None) -- Intercept the request during the opening handshake. Return an HTTP response to force the response or None to continue normally. When you force an HTTP 101 Continue response, the handshake is successful. Else, the connection is aborted.
  • process_response (Callable[[ServerConnection, Request, Response], Response | None] | None) -- Intercept the response during the opening handshake. Return an HTTP response to force the response or None to continue normally. When you force an HTTP 101 Continue response, the handshake is successful. Else, the connection is aborted.
  • server_header (str | None) -- Value of the Server response header. It defaults to "Python/x.y.z websockets/X.Y". Setting it to None removes the header.
  • compression (str | None) -- The "permessage-deflate" extension is enabled by default. Set compression to None to disable it. See the compression guide for details.
  • open_timeout (float | None) -- Timeout for opening connections in seconds. None disables the timeout.
  • close_timeout (float | None) -- Timeout for closing connections in seconds. None disables the timeout.
  • max_size (int | None) -- Maximum size of incoming messages in bytes. None disables the limit.
  • logger (Logger | LoggerAdapter | None) -- Logger for this server. It defaults to logging.getLogger("websockets.server"). See the logging guide for details.
  • create_connection (Type[ServerConnection] | None) -- Factory for the ServerConnection managing the connection. Set it to a wrapper or a subclass to customize connection handling.
Create a WebSocket server listening on a Unix socket.

This function is identical to serve(), except the host and port arguments are replaced by path. It's only available on Unix.

It's useful for deploying a server behind a reverse proxy such as nginx.

  • handler (Callable[[ServerConnection], Any]) -- Connection handler. It receives the WebSocket connection, which is a ServerConnection, in argument.
  • path (str | None) -- File system path to the Unix socket.

WebSocket server returned by serve().

This class mirrors the API of BaseServer, notably the serve_forever() and shutdown() methods, as well as the context manager protocol.

  • socket (socket.socket) -- Server socket listening for new connections.
  • handler (Callable[[socket.socket, Any], None]) -- Handler for one connection. Receives the socket and address returned by accept().
  • logger (Optional[LoggerLike]) -- Logger for this server.
See socketserver.BaseServer.serve_forever().

This method doesn't return. Calling shutdown() from another thread stops the server.

Typical use:

with serve(...) as server:
    server.serve_forever()

See socketserver.BaseServer.shutdown().

See socketserver.BaseServer.fileno().

Using a connection

Threaded implementation of a WebSocket server connection.

ServerConnection provides recv() and send() methods for receiving and sending messages.

It supports iteration to receive messages:

for message in websocket:
    process(message)

The iterator exits normally when the connection is closed with close code 1000 (OK) or 1001 (going away) or without a close code. It raises a ConnectionClosedError when the connection is closed with any other code.

  • socket (socket.socket) -- Socket connected to a WebSocket client.
  • protocol (ServerProtocol) -- Sans-I/O connection.
  • close_timeout (Optional[float]) -- Timeout for closing the connection in seconds.
Iterate on incoming messages.

The iterator calls recv() and yields messages in an infinite loop.

It exits when the connection is closed normally. It raises a ConnectionClosedError exception after a protocol error or a network failure.


Receive the next message.

When the connection is closed, recv() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal closure and ConnectionClosedError after a protocol error or a network failure. This is how you detect the end of the message stream.

If timeout is None, block until a message is received. If timeout is set and no message is received within timeout seconds, raise TimeoutError. Set timeout to 0 to check if a message was already received.

If the message is fragmented, wait until all fragments are received, reassemble them, and return the whole message.

A string (str) for a Text frame or a bytestring (bytes) for a Binary frame.
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If two threads call recv() or
    recv_streaming() concurrently.
str | bytes
Receive the next message frame by frame.

If the message is fragmented, yield each fragment as it is received. The iterator must be fully consumed, or else the connection will become unusable.

recv_streaming() raises the same exceptions as recv().

An iterator of strings (str) for a Text frame or bytestrings (bytes) for a Binary frame.
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If two threads call recv() or
    recv_streaming() concurrently.
Iterator[str | bytes]
Send a message.

A string (str) is sent as a Text frame. A bytestring or bytes-like object (bytes, bytearray, or memoryview) is sent as a Binary frame.

send() also accepts an iterable of strings, bytestrings, or bytes-like objects to enable fragmentation. Each item is treated as a message fragment and sent in its own frame. All items must be of the same type, or else send() will raise a TypeError and the connection will be closed.

send() rejects dict-like objects because this is often an error. (If you really want to send the keys of a dict-like object as fragments, call its keys() method and pass the result to send().)

When the connection is closed, send() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal connection closure and ConnectionClosedError after a protocol error or a network failure.

message (str | bytes | Iterable[str | bytes]) -- Message to send.
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If a connection is busy sending a fragmented message.
  • TypeError -- If message doesn't have a supported type.
Perform the closing handshake.

close() waits for the other end to complete the handshake, for the TCP connection to terminate, and for all incoming messages to be read with recv().

close() is idempotent: it doesn't do anything once the connection is closed.

  • code (int) -- WebSocket close code.
  • reason (str) -- WebSocket close reason.
Send a Ping.

A ping may serve as a keepalive or as a check that the remote endpoint received all messages up to this point

data (str | bytes | None) -- Payload of the ping. A str will be encoded to UTF-8. If data is None, the payload is four random bytes.
An event that will be set when the corresponding pong is received. You can ignore it if you don't intend to wait.
pong_event = ws.ping()
pong_event.wait()  # only if you want to wait for the pong
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If another ping was sent with the same data and
    the corresponding pong wasn't received yet.
Event
Send a Pong.

An unsolicited pong may serve as a unidirectional heartbeat.

data (str | bytes) -- Payload of the pong. A str will be encoded to UTF-8.
ConnectionClosed -- When the connection is closed.

WebSocket connection objects also provide these attributes:

Unique identifier of the connection. Useful in logs.
Logger for this connection.
Local address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family. See getsockname().

Remote address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family. See getpeername().

The following attributes are available after the opening handshake, once the WebSocket connection is open:

Opening handshake request.
Opening handshake response.
Subprotocol negotiated during the opening handshake.

None if no subprotocol was negotiated.

Client (threading)

Opening a connection

Connect to the WebSocket server at uri.

This function returns a ClientConnection instance, which you can use to send and receive messages.

connect() may be used as a context manager:

async with websockets.sync.client.connect(...) as websocket:
    ...

The connection is closed automatically when exiting the context.

  • uri (str) -- URI of the WebSocket server.
  • sock (socket | None) -- Preexisting TCP socket. sock overrides the host and port from uri. You may call socket.create_connection() to create a suitable TCP socket.
  • ssl_context (SSLContext | None) -- Configuration for enabling TLS on the connection.
  • server_hostname (str | None) -- Host name for the TLS handshake. server_hostname overrides the host name from uri.
  • origin (Origin | None) -- Value of the Origin header, for servers that require it.
  • extensions (Sequence[ClientExtensionFactory] | None) -- List of supported extensions, in order in which they should be negotiated and run.
  • subprotocols (Sequence[Subprotocol] | None) -- List of supported subprotocols, in order of decreasing preference.
  • additional_headers (HeadersLike | None) -- Arbitrary HTTP headers to add to the handshake request.
  • user_agent_header (str | None) -- Value of the User-Agent request header. It defaults to "Python/x.y.z websockets/X.Y". Setting it to None removes the header.
  • compression (str | None) -- The "permessage-deflate" extension is enabled by default. Set compression to None to disable it. See the compression guide for details.
  • open_timeout (float | None) -- Timeout for opening the connection in seconds. None disables the timeout.
  • close_timeout (float | None) -- Timeout for closing the connection in seconds. None disables the timeout.
  • max_size (int | None) -- Maximum size of incoming messages in bytes. None disables the limit.
  • logger (Logger | LoggerAdapter | None) -- Logger for this client. It defaults to logging.getLogger("websockets.client"). See the logging guide for details.
  • create_connection (Type[ClientConnection] | None) -- Factory for the ClientConnection managing the connection. Set it to a wrapper or a subclass to customize connection handling.
  • InvalidURI -- If uri isn't a valid WebSocket URI.
  • OSError -- If the TCP connection fails.
  • InvalidHandshake -- If the opening handshake fails.
  • TimeoutError -- If the opening handshake times out.
Connect to a WebSocket server listening on a Unix socket.

This function is identical to connect(), except for the additional path argument. It's only available on Unix.

It's mainly useful for debugging servers listening on Unix sockets.

  • path (str | None) -- File system path to the Unix socket.
  • uri (str | None) -- URI of the WebSocket server. uri defaults to ws://localhost/ or, when a ssl_context is provided, to wss://localhost/.

Using a connection

Threaded implementation of a WebSocket client connection.

ClientConnection provides recv() and send() methods for receiving and sending messages.

It supports iteration to receive messages:

for message in websocket:
    process(message)

The iterator exits normally when the connection is closed with close code 1000 (OK) or 1001 (going away) or without a close code. It raises a ConnectionClosedError when the connection is closed with any other code.

  • socket (socket.socket) -- Socket connected to a WebSocket server.
  • protocol (ClientProtocol) -- Sans-I/O connection.
  • close_timeout (Optional[float]) -- Timeout for closing the connection in seconds.
Iterate on incoming messages.

The iterator calls recv() and yields messages in an infinite loop.

It exits when the connection is closed normally. It raises a ConnectionClosedError exception after a protocol error or a network failure.


Receive the next message.

When the connection is closed, recv() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal closure and ConnectionClosedError after a protocol error or a network failure. This is how you detect the end of the message stream.

If timeout is None, block until a message is received. If timeout is set and no message is received within timeout seconds, raise TimeoutError. Set timeout to 0 to check if a message was already received.

If the message is fragmented, wait until all fragments are received, reassemble them, and return the whole message.

A string (str) for a Text frame or a bytestring (bytes) for a Binary frame.
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If two threads call recv() or
    recv_streaming() concurrently.
str | bytes
Receive the next message frame by frame.

If the message is fragmented, yield each fragment as it is received. The iterator must be fully consumed, or else the connection will become unusable.

recv_streaming() raises the same exceptions as recv().

An iterator of strings (str) for a Text frame or bytestrings (bytes) for a Binary frame.
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If two threads call recv() or
    recv_streaming() concurrently.
Iterator[str | bytes]
Send a message.

A string (str) is sent as a Text frame. A bytestring or bytes-like object (bytes, bytearray, or memoryview) is sent as a Binary frame.

send() also accepts an iterable of strings, bytestrings, or bytes-like objects to enable fragmentation. Each item is treated as a message fragment and sent in its own frame. All items must be of the same type, or else send() will raise a TypeError and the connection will be closed.

send() rejects dict-like objects because this is often an error. (If you really want to send the keys of a dict-like object as fragments, call its keys() method and pass the result to send().)

When the connection is closed, send() raises ConnectionClosed. Specifically, it raises ConnectionClosedOK after a normal connection closure and ConnectionClosedError after a protocol error or a network failure.

message (str | bytes | Iterable[str | bytes]) -- Message to send.
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If a connection is busy sending a fragmented message.
  • TypeError -- If message doesn't have a supported type.
Perform the closing handshake.

close() waits for the other end to complete the handshake, for the TCP connection to terminate, and for all incoming messages to be read with recv().

close() is idempotent: it doesn't do anything once the connection is closed.

  • code (int) -- WebSocket close code.
  • reason (str) -- WebSocket close reason.
Send a Ping.

A ping may serve as a keepalive or as a check that the remote endpoint received all messages up to this point

data (str | bytes | None) -- Payload of the ping. A str will be encoded to UTF-8. If data is None, the payload is four random bytes.
An event that will be set when the corresponding pong is received. You can ignore it if you don't intend to wait.
pong_event = ws.ping()
pong_event.wait()  # only if you want to wait for the pong
  • ConnectionClosed -- When the connection is closed.
  • RuntimeError -- If another ping was sent with the same data and
    the corresponding pong wasn't received yet.
Event
Send a Pong.

An unsolicited pong may serve as a unidirectional heartbeat.

data (str | bytes) -- Payload of the pong. A str will be encoded to UTF-8.
ConnectionClosed -- When the connection is closed.

WebSocket connection objects also provide these attributes:

Unique identifier of the connection. Useful in logs.
Logger for this connection.
Local address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family. See getsockname().

Remote address of the connection.

For IPv4 connections, this is a (host, port) tuple.

The format of the address depends on the address family. See getpeername().

The following attributes are available after the opening handshake, once the WebSocket connection is open:

Opening handshake request.
Opening handshake response.
Subprotocol negotiated during the opening handshake.

None if no subprotocol was negotiated.

Sans-I/O

This layer is designed for integrating in third-party libraries, typically application servers.

Server (Sans-I/O)

Sans-I/O implementation of a WebSocket server connection.
  • origins (Optional[Sequence[Optional[Origin]]]) -- acceptable values of the Origin header; include None in the list if the lack of an origin is acceptable. This is useful for defending against Cross-Site WebSocket Hijacking attacks.
  • extensions (List[Extension]) -- list of supported extensions, in order in which they should be tried.
  • subprotocols (Optional[Sequence[Subprotocol]]) -- list of supported subprotocols, in order of decreasing preference.
  • select_subprotocol (Optional[Callable[[ServerProtocol, Sequence[Subprotocol]], Optional[Subprotocol]]]) -- Callback for selecting a subprotocol among those supported by the client and the server. It has the same signature as the select_subprotocol() method, including a ServerProtocol instance as first argument.
  • state (State) -- initial state of the WebSocket connection.
  • max_size (Optional[int]) -- maximum size of incoming messages in bytes; None disables the limit.
  • logger (Logger | LoggerAdapter) -- logger for this connection; defaults to logging.getLogger("websockets.client"); see the logging guide for details.
Receive data from the network.

After calling this method:

  • You must call data_to_send() and send this data to the network.
  • You should call events_received() and process resulting events.
EOFError -- if receive_eof() was called earlier.
Receive the end of the data stream from the network.

After calling this method:

  • You must call data_to_send() and send this data to the network; it will return [b""], signaling the end of the stream, or [].
  • You aren't expected to call events_received(); it won't return any new events.
EOFError -- if receive_eof() was called earlier.
Create a handshake response to accept the connection.

If the connection cannot be established, the handshake response actually rejects the handshake.

You must send the handshake response with send_response().

You may modify it before sending it, for example to add HTTP headers.

request (Request) -- WebSocket handshake request event received from the client.
WebSocket handshake response event to send to the client.
Response
Pick a subprotocol among those offered by the client.

If several subprotocols are supported by both the client and the server, pick the first one in the list declared the server.

If the server doesn't support any subprotocols, continue without a subprotocol, regardless of what the client offers.

If the server supports at least one subprotocol and the client doesn't offer any, abort the handshake with an HTTP 400 error.

You provide a select_subprotocol argument to ServerProtocol to override this logic. For example, you could accept the connection even if client doesn't offer a subprotocol, rather than reject it.

Here's how to negotiate the chat subprotocol if the client supports it and continue without a subprotocol otherwise:

def select_subprotocol(protocol, subprotocols):
    if "chat" in subprotocols:
        return "chat"
subprotocols (Sequence[Subprotocol]) -- list of subprotocols offered by the client.
Selected subprotocol, if a common subprotocol was found.

None to continue without a subprotocol.

Optional[Subprotocol]
NegotiationError -- custom implementations may raise this exception
to abort the handshake with an HTTP 400 error.
Create a handshake response to reject the connection.

A short plain text response is the best fallback when failing to establish a WebSocket connection.

You must send the handshake response with send_response().

You can modify it before sending it, for example to alter HTTP headers.

  • status (HTTPStatus | int) -- HTTP status code.
  • text (str) -- HTTP response body; will be encoded to UTF-8.
WebSocket handshake response event to send to the client.
Response
Send a handshake response to the client.
response (Response) -- WebSocket handshake response event to send.
Send a Continuation frame.
  • data (bytes) -- payload containing the same kind of data as the initial frame.
  • fin (bool) -- FIN bit; set it to True if this is the last frame of a fragmented message and to False otherwise.
ProtocolError -- if a fragmented message isn't in progress.
Send a Text frame.
  • data (bytes) -- payload containing text encoded with UTF-8.
  • fin (bool) -- FIN bit; set it to False if this is the first frame of a fragmented message.
ProtocolError -- if a fragmented message is in progress.
Send a Binary frame.
  • data (bytes) -- payload containing arbitrary binary data.
  • fin (bool) -- FIN bit; set it to False if this is the first frame of a fragmented message.
ProtocolError -- if a fragmented message is in progress.
Send a Close frame.
  • code (int | None) -- close code.
  • reason (str) -- close reason.
ProtocolError -- if a fragmented message is being sent, if the code
isn't valid, or if a reason is provided without a code
Send a Ping frame.
data (bytes) -- payload containing arbitrary binary data.
Send a Pong frame.
data (bytes) -- payload containing arbitrary binary data.
Fail the WebSocket connection.
  • code (int) -- close code
  • reason (str) -- close reason
ProtocolError -- if the code isn't valid.
Fetch events generated from data received from the network.

Call this method immediately after any of the receive_*() methods.

Process resulting events, likely by passing them to the application.

Events read from the connection.
List[Event]
Obtain data to send to the network.

Call this method immediately after any of the receive_*(), send_*(), or fail() methods.

Write resulting data to the connection.

The empty bytestring SEND_EOF signals the end of the data stream. When you receive it, half-close the TCP connection.

Data to write to the connection.
List[bytes]
Tell if the TCP connection is expected to close soon.

Call this method immediately after any of the receive_*(), send_close(), or fail() methods.

If it returns True, schedule closing the TCP connection after a short timeout if the other side hasn't already closed it.

Whether the TCP connection is expected to close soon.
bool

WebSocket protocol objects also provide these attributes:

Unique identifier of the connection. Useful in logs.
Logger for this connection.
WebSocket connection state.

Defined in 4.1, 4.2, 7.1.3, and 7.1.4 of RFC 6455.

The following attributes are available after the opening handshake, once the WebSocket connection is open:

Exception to raise if the opening handshake failed.

None if the opening handshake succeeded.

The following attributes are available after the closing handshake, once the WebSocket connection is closed:

WebSocket close code.

None if the connection isn't closed yet.

WebSocket close reason.

None if the connection isn't closed yet.

Exception to raise when trying to interact with a closed connection.

Don't raise this exception while the connection state is CLOSING; wait until it's CLOSED.

Indeed, the exception includes the close code and reason, which are known only once the connection is closed.

AssertionError -- if the connection isn't closed yet.

Client (Sans-I/O)

Sans-I/O implementation of a WebSocket client connection.
  • wsuri (WebSocketURI) -- URI of the WebSocket server, parsed with parse_uri().
  • origin (Optional[Origin]) -- value of the Origin header. This is useful when connecting to a server that validates the Origin header to defend against Cross-Site WebSocket Hijacking attacks.
  • extensions (Optional[Sequence[ClientExtensionFactory]]) -- list of supported extensions, in order in which they should be tried.
  • subprotocols (Optional[Sequence[Subprotocol]]) -- list of supported subprotocols, in order of decreasing preference.
  • state (State) -- initial state of the WebSocket connection.
  • max_size (Optional[int]) -- maximum size of incoming messages in bytes; None disables the limit.
  • logger (Optional[LoggerLike]) -- logger for this connection; defaults to logging.getLogger("websockets.client"); see the logging guide for details.
Receive data from the network.

After calling this method:

  • You must call data_to_send() and send this data to the network.
  • You should call events_received() and process resulting events.
EOFError -- if receive_eof() was called earlier.
Receive the end of the data stream from the network.

After calling this method:

  • You must call data_to_send() and send this data to the network; it will return [b""], signaling the end of the stream, or [].
  • You aren't expected to call events_received(); it won't return any new events.
EOFError -- if receive_eof() was called earlier.
Create a handshake request to open a connection.

You must send the handshake request with send_request().

You can modify it before sending it, for example to add HTTP headers.

WebSocket handshake request event to send to the server.
Request
Send a handshake request to the server.
request (Request) -- WebSocket handshake request event.
Send a Continuation frame.
  • data (bytes) -- payload containing the same kind of data as the initial frame.
  • fin (bool) -- FIN bit; set it to True if this is the last frame of a fragmented message and to False otherwise.
ProtocolError -- if a fragmented message isn't in progress.
Send a Text frame.
  • data (bytes) -- payload containing text encoded with UTF-8.
  • fin (bool) -- FIN bit; set it to False if this is the first frame of a fragmented message.
ProtocolError -- if a fragmented message is in progress.
Send a Binary frame.
  • data (bytes) -- payload containing arbitrary binary data.
  • fin (bool) -- FIN bit; set it to False if this is the first frame of a fragmented message.
ProtocolError -- if a fragmented message is in progress.
Send a Close frame.
  • code (int | None) -- close code.
  • reason (str) -- close reason.
ProtocolError -- if a fragmented message is being sent, if the code
isn't valid, or if a reason is provided without a code
Send a Ping frame.
data (bytes) -- payload containing arbitrary binary data.
Send a Pong frame.
data (bytes) -- payload containing arbitrary binary data.
Fail the WebSocket connection.
  • code (int) -- close code
  • reason (str) -- close reason
ProtocolError -- if the code isn't valid.
Fetch events generated from data received from the network.

Call this method immediately after any of the receive_*() methods.

Process resulting events, likely by passing them to the application.

Events read from the connection.
List[Event]
Obtain data to send to the network.

Call this method immediately after any of the receive_*(), send_*(), or fail() methods.

Write resulting data to the connection.

The empty bytestring SEND_EOF signals the end of the data stream. When you receive it, half-close the TCP connection.

Data to write to the connection.
List[bytes]
Tell if the TCP connection is expected to close soon.

Call this method immediately after any of the receive_*(), send_close(), or fail() methods.

If it returns True, schedule closing the TCP connection after a short timeout if the other side hasn't already closed it.

Whether the TCP connection is expected to close soon.
bool

WebSocket protocol objects also provide these attributes:

Unique identifier of the connection. Useful in logs.
Logger for this connection.
WebSocket connection state.

Defined in 4.1, 4.2, 7.1.3, and 7.1.4 of RFC 6455.

The following attributes are available after the opening handshake, once the WebSocket connection is open:

Exception to raise if the opening handshake failed.

None if the opening handshake succeeded.

The following attributes are available after the closing handshake, once the WebSocket connection is closed:

WebSocket close code.

None if the connection isn't closed yet.

WebSocket close reason.

None if the connection isn't closed yet.

Exception to raise when trying to interact with a closed connection.

Don't raise this exception while the connection state is CLOSING; wait until it's CLOSED.

Indeed, the exception includes the close code and reason, which are known only once the connection is closed.

AssertionError -- if the connection isn't closed yet.

The Per-Message Deflate extension is built in. You may also define custom extensions.

The WebSocket protocol supports extensions.

At the time of writing, there's only one registered extension with a public specification, WebSocket Per-Message Deflate.

Per-Message Deflate

websockets.extensions.permessage_deflate implements WebSocket Per-Message Deflate.

This extension is specified in RFC 7692.

Refer to the topic guide on compression to learn more about tuning compression settings.

Client-side extension factory for the Per-Message Deflate extension.

Parameters behave as described in section 7.1 of RFC 7692.

Set them to True to include them in the negotiation offer without a value or to an integer value to include them with this value.

  • server_no_context_takeover (bool) -- prevent server from using context takeover.
  • client_no_context_takeover (bool) -- prevent client from using context takeover.
  • server_max_window_bits (Optional[int]) -- maximum size of the server's LZ77 sliding window in bits, between 8 and 15.
  • client_max_window_bits (Optional[Union[int, bool]]) -- maximum size of the client's LZ77 sliding window in bits, between 8 and 15, or True to indicate support without setting a limit.
  • compress_settings (Optional[Dict[str, Any]]) -- additional keyword arguments for zlib.compressobj(), excluding wbits.
Server-side extension factory for the Per-Message Deflate extension.

Parameters behave as described in section 7.1 of RFC 7692.

Set them to True to include them in the negotiation offer without a value or to an integer value to include them with this value.

  • server_no_context_takeover (bool) -- prevent server from using context takeover.
  • client_no_context_takeover (bool) -- prevent client from using context takeover.
  • server_max_window_bits (Optional[int]) -- maximum size of the server's LZ77 sliding window in bits, between 8 and 15.
  • client_max_window_bits (Optional[int]) -- maximum size of the client's LZ77 sliding window in bits, between 8 and 15.
  • compress_settings (Optional[Dict[str, Any]]) -- additional keyword arguments for zlib.compressobj(), excluding wbits.
  • require_client_max_window_bits (bool) -- do not enable compression at all if client doesn't advertise support for client_max_window_bits; the default behavior is to enable compression without enforcing client_max_window_bits.

extensions.

Refer to the how-to guide on extensions to learn more about writing an extension.

Base class for extensions.
Extension identifier.
Decode an incoming frame.
  • frame (Frame) -- incoming frame.
  • max_size (int | None) -- maximum payload size in bytes.
Decoded frame.
Frame
PayloadTooBig -- if decoding the payload exceeds max_size.
Encode an outgoing frame.
frame (Frame) -- outgoing frame.
Encoded frame.
Frame
Base class for client-side extension factories.
Extension identifier.
Build parameters to send to the server for this extension.
Parameters to send to the server.
List[ExtensionParameter]
Process parameters received from the server.
  • params (Sequence[ExtensionParameter]) -- parameters received from the server for this extension.
  • accepted_extensions (Sequence[Extension]) -- list of previously accepted extensions.
An extension instance.
Extension
NegotiationError -- if parameters aren't acceptable.
Base class for server-side extension factories.
Process parameters received from the client.
  • params (Sequence[ExtensionParameter]) -- parameters received from the client for this extension.
  • accepted_extensions (Sequence[Extension]) -- list of previously accepted extensions.
To accept the offer, parameters to send to the client for this extension and an extension instance.
Tuple[List[ExtensionParameter], Extension]
NegotiationError -- to reject the offer, if parameters received from
the client aren't acceptable.

These low-level APIs are shared by all implementations.

WebSocket frame.

Opcode.
websockets.frames.Opcode
Payload data.
bytes
FIN bit.
bool
RSV1 bit.
bool
RSV2 bit.
bool
RSV3 bit.
bool

Only these fields are needed. The MASK bit, payload length and masking-key are handled on the fly when parsing and serializing frames.

Opcode values for WebSocket frames.
Code and reason for WebSocket close frames.

Close code.
int
Close reason.
str
Close code values for WebSocket close frames.

WebSocket handshake request.

Request path, including optional query.
str
Request headers.
websockets.datastructures.Headers
WebSocket handshake response.

Response code.
int
Response reason.
str
Response headers.
websockets.datastructures.Headers
Response body, if any.
bytes | None
Efficient data structure for manipulating HTTP headers.

A list of (name, values) is inefficient for lookups.

A dict doesn't suffice because header names are case-insensitive and multiple occurrences of headers with the same name are possible.

Headers stores HTTP headers in a hybrid data structure to provide efficient insertions and lookups while preserving the original data.

In order to account for multiple values with minimal hassle, Headers follows this logic:

•
  • if there's no value, KeyError is raised;
  • if there's exactly one value, it's returned;
  • if there's more than one value, MultipleValuesError is raised.
  • When setting a header with headers[name] = value, the value is appended to the list of values for that header.
  • When deleting a header with del headers[name], all values for that header are removed (this is slow).

Other methods for manipulating headers are consistent with this logic.

As long as no header occurs multiple times, Headers behaves like dict, except keys are lower-cased to provide case-insensitivity.

Two methods support manipulating multiple values explicitly:

  • get_all() returns a list of all values for a header;
  • raw_items() returns an iterator of (name, values) pairs.

Return the (possibly empty) list of all values for a header.
key (str) -- header name.
Return an iterator of all values as (name, value) pairs.

Exception raised when Headers has more than one value for a key.

Parse and validate a WebSocket URI.
uri (str) -- WebSocket URI.
Parsed WebSocket URI.
WebSocketURI
InvalidURI -- if uri isn't a valid WebSocket URI.
WebSocket URI.

True for a wss URI, False for a ws URI.
bool
Normalized to lower case.
str
Always set even if it's the default.
int
May be empty.
str
May be empty if the URI doesn't include a query component.
str
Available when the URI contains User Information.
str | None
Available when the URI contains User Information.
str | None

websockets.exceptions defines the following exception hierarchy:

•
•
  • ConnectionClosedError
  • ConnectionClosedOK
•
  • SecurityError
  • InvalidMessage
  • InvalidHeaderFormat
  • InvalidHeaderValue
  • InvalidOrigin
  • InvalidUpgrade
  • InvalidStatus
  • InvalidStatusCode (legacy)
  • DuplicateParameter
  • InvalidParameterName
  • InvalidParameterValue
  • AbortHandshake
  • RedirectHandshake
  • InvalidState
  • InvalidURI
  • PayloadTooBig
  • ProtocolError
Raised to abort the handshake on purpose and return an HTTP response.

This exception is an implementation detail.

The public API is process_request().


HTTP status code.
HTTPStatus
HTTP response headers.
Headers
HTTP response body.
bytes
Raised when trying to interact with a closed connection.

if a close frame was received, its code and reason are available in rcvd.code and rcvd.reason.
Optional[Close]
if a close frame was sent, its code and reason are available in sent.code and sent.reason.
Optional[Close]
if close frames were received and sent, this attribute tells in which order this happened, from the perspective of this side of the connection.
Optional[bool]
Like ConnectionClosed, when the connection terminated with an error.

A close frame with a code other than 1000 (OK) or 1001 (going away) was received or sent, or the closing handshake didn't complete properly.


Like ConnectionClosed, when the connection terminated properly.

A close code with code 1000 (OK) or 1001 (going away) or without a code was received and sent.


Raised when a parameter name is repeated in an extension header.

Raised during the handshake when the WebSocket connection fails.
Raised when an HTTP header doesn't have a valid format or value.

Raised when an HTTP header cannot be parsed.

The format of the header doesn't match the grammar for that header.


Raised when an HTTP header has a wrong value.

The format of the header is correct but a value isn't acceptable.


Raised when a handshake request or response is malformed.
Raised when the Origin header in a request isn't allowed.

Raised when a parameter name in an extension header is invalid.

Raised when a parameter value in an extension header is invalid.

Raised when an operation is forbidden in the current state.

This exception is an implementation detail.

It should never be raised in normal circumstances.

Raised when a handshake response rejects the WebSocket upgrade.

Raised when a handshake response status code is invalid.

Raised when connecting to a URI that isn't a valid WebSocket URI.

Raised when the Upgrade or Connection header isn't correct.

Raised when negotiating an extension fails.
Raised when receiving a frame with a payload exceeding the maximum size.
Raised when a frame breaks the protocol.
Raised when a handshake gets redirected.

This exception is an implementation detail.


Raised when a handshake request or response breaks a security rule.

Security limits are hard coded.

Base class for all exceptions defined by websockets.
alias of ProtocolError

Types supported in a WebSocket message: str for a Text frame, bytes for a Binary.

alias of Union[str, bytes]

Types accepted where a Logger is expected.

alias of Union[Logger, LoggerAdapter]

Types accepted where an HTTPStatus is expected.

alias of Union[HTTPStatus, int]

Value of a Origin header.
Subprotocol in a Sec-WebSocket-Protocol header.
Name of a WebSocket extension.
Parameter of a WebSocket extension.

alias of Tuple[str, Optional[str]]

Events that events_received() may return.

alias of Union[Request, Response, Frame]

Types accepted where Headers is expected.

In addition to Headers itself, this includes dict-like types where both keys and values are str.

alias of Union[Headers, Mapping[str, str], Iterable[Tuple[str, str]], SupportsKeysAndGetItem]

Dict-like types with keys() -> str and __getitem__(key: str) -> str methods.

Public APIs documented in this API reference are subject to the backwards-compatibility policy.

Anything that isn't listed in the API reference is a private API. There's no guarantees of behavior or backwards-compatibility for private APIs.

For convenience, many public APIs can be imported directly from the websockets package.

Get a deeper understanding of how websockets is built and why.

When you deploy your websockets server to production, at a high level, your architecture will almost certainly look like the following diagram: [image]

The basic unit for scaling a websockets server is "one server process". Each blue box in the diagram represents one server process.

There's more variation in routing. While the routing layer is shown as one big box, it is likely to involve several subsystems.

When you design a deployment, your should consider two questions:

1.
How will I run the appropriate number of server processes?
2.
How will I route incoming connections to these processes?

These questions are strongly related. There's a wide range of acceptable answers, depending on your goals and your constraints.

You can find a few concrete examples in the deployment how-to guides.

Typically, one server process will manage a few hundreds or thousands connections, depending on the frequency of messages and the amount of work they require.

CPU and memory usage increase with the number of connections to the server.

Often CPU is the limiting factor. If a server process goes to 100% CPU, then you reached the limit. How much headroom you want to keep is up to you.

Once you know how many connections a server process can manage and how many connections you need to handle, you can calculate how many processes to run.

You can also automate this calculation by configuring an autoscaler to keep CPU usage or connection count within acceptable limits.

Don't scale with threads. Threads doesn't make sense for a server built with asyncio.

Most solutions for running multiple instances of a server process fall into one of these three buckets:

1.
Running N processes on a platform:
  • a Kubernetes Deployment
  • its equivalent on a Platform as a Service provider
2.
Running N servers:
  • an AWS Auto Scaling group, a GCP Managed instance group, etc.
  • a fixed set of long-lived servers
3.
Running N processes on a server:
•
preferably via a process manager or supervisor

Option 1 is easiest of you have access to such a platform.

Option 2 almost always combines with option 3.

Run a Python program that invokes serve(). That's it.

Don't run an ASGI server such as Uvicorn, Hypercorn, or Daphne. They're alternatives to websockets, not complements.

Don't run a WSGI server such as Gunicorn, Waitress, or mod_wsgi. They aren't designed to run WebSocket applications.

Applications servers handle network connections and expose a Python API. You don't need one because websockets handles network connections directly.

Process managers send the SIGTERM signal to terminate processes. Catch this signal and exit the server to ensure a graceful shutdown.

Here's an example:

#!/usr/bin/env python
import asyncio
import signal
import websockets
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def server():
    # Set the stop condition when receiving SIGTERM.
    loop = asyncio.get_running_loop()
    stop = loop.create_future()
    loop.add_signal_handler(signal.SIGTERM, stop.set_result, None)
    async with websockets.serve(echo, "localhost", 8765):
        await stop
asyncio.run(server())

When exiting the context manager, serve() closes all connections with code 1001 (going away). As a consequence:

  • If the connection handler is awaiting recv(), it receives a ConnectionClosedOK exception. It can catch the exception and clean up before exiting.
  • Otherwise, it should be waiting on wait_closed(), so it can receive the ConnectionClosedOK exception and exit.

This example is easily adapted to handle other signals.

If you override the default signal handler for SIGINT, which raises KeyboardInterrupt, be aware that you won't be able to interrupt a program with Ctrl-C anymore when it's stuck in a loop.

Since the routing layer is directly exposed to the Internet, it should provide appropriate protection against threats ranging from Internet background noise to targeted attacks.

You should always secure WebSocket connections with TLS. Since the routing layer carries the public domain name, it should terminate TLS connections.

Finally, it must route connections to the server processes, balancing new connections across them.

Here are typical solutions for load balancing, matched to ways of running processes:

1.
If you're running on a platform, it comes with a routing layer:
  • a Kubernetes Ingress and Service
  • a service mesh: Istio, Consul, Linkerd, etc.
  • the routing mesh of a Platform as a Service
2.
If you're running N servers, you may load balance with:
  • a cloud load balancer: AWS Elastic Load Balancing, GCP Cloud Load Balancing, etc.
  • A software load balancer: HAProxy, NGINX, etc.
3.
If you're running N processes on a server, you may load balance with:
  • A software load balancer: HAProxy, NGINX, etc.
  • The operating system — all processes listen on the same port

You may trust the load balancer to handle encryption and to provide security. You may add another layer in front of the load balancer for these purposes.

There are many possibilities. Don't add layers that you don't need, though.

How do I implement a health check?

Load balancers need a way to check whether server processes are up and running to avoid routing connections to a non-functional backend.

websockets provide minimal support for responding to HTTP requests with the process_request() hook.

Here's an example:

#!/usr/bin/env python
import asyncio
import http
import websockets
async def health_check(path, request_headers):
    if path == "/healthz":
        return http.HTTPStatus.OK, [], b"OK\n"
async def echo(websocket):
    async for message in websocket:
        await websocket.send(message)
async def main():
    async with websockets.serve(
        echo, "localhost", 8765,
        process_request=health_check,
    ):
        await asyncio.Future()  # run forever
asyncio.run(main())

When you run a WebSocket client, your code calls coroutines provided by websockets.

If an error occurs, websockets tells you by raising an exception. For example, it raises a ConnectionClosed exception if the other side closes the connection.

When you run a WebSocket server, websockets accepts connections, performs the opening handshake, runs the connection handler coroutine that you provided, and performs the closing handshake.

Given this inversion of control, if an error happens in the opening handshake or if the connection handler crashes, there is no way to raise an exception that you can handle.

Logs tell you about these errors.

Besides errors, you may want to record the activity of the server.

In a request/response protocol such as HTTP, there's an obvious way to record activity: log one event per request/response. Unfortunately, this solution doesn't work well for a bidirectional protocol such as WebSocket.

Instead, when running as a server, websockets logs one event when a connection is established and another event when a connection is closed.

By default, websockets doesn't log an event for every message. That would be excessive for many applications exchanging small messages at a fast rate. If you need this level of detail, you could add logging in your own code.

Finally, you can enable debug logs to get details about everything websockets is doing. This can be useful when developing clients as well as servers.

See log levels below for a list of events logged by websockets logs at each log level.

websockets relies on the logging module from the standard library in order to maximize compatibility and integrate nicely with other libraries:

import logging

websockets logs to the "websockets.client" and "websockets.server" loggers.

websockets doesn't provide a default logging configuration because requirements vary a lot depending on the environment.

Here's a basic configuration for a server in production:

logging.basicConfig(
    format="%(asctime)s %(message)s",
    level=logging.INFO,
)

Here's how to enable debug logs for development:

logging.basicConfig(
    format="%(message)s",
    level=logging.DEBUG,
)

Furthermore, websockets adds a websocket attribute to log records, so you can include additional information about the current connection in logs.

You could attempt to add information with a formatter:

# this doesn't work!
logging.basicConfig(
    format="{asctime} {websocket.id} {websocket.remote_address[0]} {message}",
    level=logging.INFO,
    style="{",
)

However, this technique runs into two problems:

  • The formatter applies to all records. It will crash if it receives a record without a websocket attribute. For example, this happens when logging that the server starts because there is no current connection.
  • Even with str.format() style, you're restricted to attribute and index lookups, which isn't enough to implement some fairly simple requirements.

There's a better way. connect() and serve() accept a logger argument to override the default Logger. You can set logger to a LoggerAdapter that enriches logs.

For example, if the server is behind a reverse proxy, remote_address gives the IP address of the proxy, which isn't useful. IP addresses of clients are provided in an HTTP header set by the proxy.

Here's how to include them in logs, assuming they're in the X-Forwarded-For header:

logging.basicConfig(
    format="%(asctime)s %(message)s",
    level=logging.INFO,
)
class LoggerAdapter(logging.LoggerAdapter):
    """Add connection ID and client IP address to websockets logs."""
    def process(self, msg, kwargs):
        try:
            websocket = kwargs["extra"]["websocket"]
        except KeyError:
            return msg, kwargs
        xff = websocket.request_headers.get("X-Forwarded-For")
        return f"{websocket.id} {xff} {msg}", kwargs
async with websockets.serve(
    ...,
    # Python < 3.10 requires passing None as the second argument.
    logger=LoggerAdapter(logging.getLogger("websockets.server"), None),
):
    ...

Even though logging predates structured logging, it's still possible to output logs as JSON with a bit of effort.

First, we need a Formatter that renders JSON:

import json
import logging
import datetime
class JSONFormatter(logging.Formatter):
    """
    Render logs as JSON.
    To add details to a log record, store them in a ``event_data``
    custom attribute. This dict is merged into the event.
    """
    def __init__(self):
        pass  # override logging.Formatter constructor
    def format(self, record):
        event = {
            "timestamp": self.getTimestamp(record.created),
            "message": record.getMessage(),
            "level": record.levelname,
            "logger": record.name,
        }
        event_data = getattr(record, "event_data", None)
        if event_data:
            event.update(event_data)
        if record.exc_info:
            event["exc_info"] = self.formatException(record.exc_info)
        if record.stack_info:
            event["stack_info"] = self.formatStack(record.stack_info)
        return json.dumps(event)
    def getTimestamp(self, created):
        return datetime.datetime.utcfromtimestamp(created).isoformat()

Then, we configure logging to apply this formatter:

handler = logging.StreamHandler()
handler.setFormatter(formatter)
logger = logging.getLogger()
logger.addHandler(handler)
logger.setLevel(logging.INFO)

Finally, we populate the event_data custom attribute in log records with a LoggerAdapter:

class LoggerAdapter(logging.LoggerAdapter):
    """Add connection ID and client IP address to websockets logs."""
    def process(self, msg, kwargs):
        try:
            websocket = kwargs["extra"]["websocket"]
        except KeyError:
            return msg, kwargs
        kwargs["extra"]["event_data"] = {
            "connection_id": str(websocket.id),
            "remote_addr": websocket.request_headers.get("X-Forwarded-For"),
        }
        return msg, kwargs
async with websockets.serve(
    ...,
    # Python < 3.10 requires passing None as the second argument.
    logger=LoggerAdapter(logging.getLogger("websockets.server"), None),
):
    ...

If your application doesn't configure logging, Python outputs messages of severity WARNING and higher to stderr. As a consequence, you will see a message and a stack trace if a connection handler coroutine crashes or if you hit a bug in websockets.

If you want to disable this behavior for websockets, you can add a NullHandler:

logging.getLogger("websockets").addHandler(logging.NullHandler())

Additionally, if your application configures logging, you must disable propagation to the root logger, or else its handlers could output logs:

logging.getLogger("websockets").propagate = False

Alternatively, you could set the log level to CRITICAL for the "websockets" logger, as the highest level currently used is ERROR:

logging.getLogger("websockets").setLevel(logging.CRITICAL)

Or you could configure a filter to drop all messages:

logging.getLogger("websockets").addFilter(lambda record: None)

Here's what websockets logs at each level.

ERROR

  • Exceptions raised by connection handler coroutines in servers
  • Exceptions resulting from bugs in websockets

WARNING

•
Failures in broadcast()

INFO

  • Server starting and stopping
  • Server establishing and closing connections
  • Client reconnecting automatically

DEBUG

  • Changes to the state of connections
  • Handshake requests and responses
  • All frames sent and received
  • Steps to close a connection
  • Keepalive pings and pongs
  • Errors handled transparently

Debug messages have cute prefixes that make logs easier to scan:

  • > - send something
  • < - receive something
  • = - set connection state
  • x - shut down connection
  • % - manage pings and pongs
  • ! - handle errors and timeouts

The WebSocket protocol was designed for creating web applications that need bidirectional communication between clients running in browsers and servers.

In most practical use cases, WebSocket servers need to authenticate clients in order to route communications appropriately and securely.

RFC 6455 stays elusive when it comes to authentication:

This protocol doesn't prescribe any particular way that servers can authenticate clients during the WebSocket handshake. The WebSocket server can use any client authentication mechanism available to a generic HTTP server, such as cookies, HTTP authentication, or TLS authentication.

None of these three mechanisms works well in practice. Using cookies is cumbersome, HTTP authentication isn't supported by all mainstream browsers, and TLS authentication in a browser is an esoteric user experience.

Fortunately, there are better alternatives! Let's discuss them.

Consider a setup where the WebSocket server is separate from the HTTP server.

Most servers built with websockets to complement a web application adopt this design because websockets doesn't aim at supporting HTTP.

The following diagram illustrates the authentication flow. [image]

Assuming the current user is authenticated with the HTTP server (1), the application needs to obtain credentials from the HTTP server (2) in order to send them to the WebSocket server (3), who can check them against the database of user accounts (4).

Usernames and passwords aren't a good choice of credentials here, if only because passwords aren't available in clear text in the database.

Tokens linked to user accounts are a better choice. These tokens must be impossible to forge by an attacker. For additional security, they can be short-lived or even single-use.

Assume the web application obtained authentication credentials, likely a token, from the HTTP server. There's four options for passing them to the WebSocket server.

1.
Sending credentials as the first message in the WebSocket connection.

This is fully reliable and the most secure mechanism in this discussion. It has two minor downsides:

  • Authentication is performed at the application layer. Ideally, it would be managed at the protocol layer.
  • Authentication is performed after the WebSocket handshake, making it impossible to monitor authentication failures with HTTP response codes.
2.
Adding credentials to the WebSocket URI in a query parameter.

This is also fully reliable but less secure. Indeed, it has a major downside:

•
URIs end up in logs, which leaks credentials. Even if that risk could be lowered with single-use tokens, it is usually considered unacceptable.

Authentication is still performed at the application layer but it can happen before the WebSocket handshake, which improves separation of concerns and enables responding to authentication failures with HTTP 401.

3.
Setting a cookie on the domain of the WebSocket URI.

Cookies are undoubtedly the most common and hardened mechanism for sending credentials from a web application to a server. In an HTTP application, credentials would be a session identifier or a serialized, signed session.

Unfortunately, when the WebSocket server runs on a different domain from the web application, this idea bumps into the Same-Origin Policy. For security reasons, setting a cookie on a different origin is impossible.

The proper workaround consists in:

  • creating a hidden iframe served from the domain of the WebSocket server
  • sending the token to the iframe with postMessage
  • setting the cookie in the iframe

before opening the WebSocket connection.

Sharing a parent domain (e.g. example.com) between the HTTP server (e.g. www.example.com) and the WebSocket server (e.g. ws.example.com) and setting the cookie on that parent domain would work too.

However, the cookie would be shared with all subdomains of the parent domain. For a cookie containing credentials, this is unacceptable.

4.
Adding credentials to the WebSocket URI in user information.

Letting the browser perform HTTP Basic Auth is a nice idea in theory.

In practice it doesn't work due to poor support in browsers.

As of May 2021:

  • Chrome 90 behaves as expected.
  • Firefox 88 caches credentials too aggressively.

    When connecting again to the same server with new credentials, it reuses the old credentials, which may be expired, resulting in an HTTP 401. Then the next connection succeeds. Perhaps errors clear the cache.

    When tokens are short-lived or single-use, this bug produces an interesting effect: every other WebSocket connection fails.

  • Safari 14 ignores credentials entirely.

Two other options are off the table:

1.
Setting a custom HTTP header

This would be the most elegant mechanism, solving all issues with the options discussed above.

Unfortunately, it doesn't work because the WebSocket API doesn't support setting custom headers.

2.
Authenticating with a TLS certificate

While this is suggested by the RFC, installing a TLS certificate is too far from the mainstream experience of browser users. This could make sense in high security contexts. I hope developers working on such projects don't take security advice from the documentation of random open source projects.

The experiments/authentication directory demonstrates these techniques.

Run the experiment in an environment where websockets is installed:

$ python experiments/authentication/app.py
Running on http://localhost:8000/

When you browse to the HTTP server at http://localhost:8000/ and you submit a username, the server creates a token and returns a testing web page.

This page opens WebSocket connections to four WebSocket servers running on four different origins. It attempts to authenticate with the token in four different ways.

As soon as the connection is open, the client sends a message containing the token:

const websocket = new WebSocket("ws://.../");
websocket.onopen = () => websocket.send(token);
// ...

At the beginning of the connection handler, the server receives this message and authenticates the user. If authentication fails, the server closes the connection:

async def first_message_handler(websocket):
    token = await websocket.recv()
    user = get_user(token)
    if user is None:
        await websocket.close(CloseCode.INTERNAL_ERROR, "authentication failed")
        return
    ...

The client adds the token to the WebSocket URI in a query parameter before opening the connection:

const uri = `ws://.../?token=${token}`;
const websocket = new WebSocket(uri);
// ...

The server intercepts the HTTP request, extracts the token and authenticates the user. If authentication fails, it returns an HTTP 401:

class QueryParamProtocol(websockets.WebSocketServerProtocol):
    async def process_request(self, path, headers):
        token = get_query_parameter(path, "token")
        if token is None:
            return http.HTTPStatus.UNAUTHORIZED, [], b"Missing token\n"
        user = get_user(token)
        if user is None:
            return http.HTTPStatus.UNAUTHORIZED, [], b"Invalid token\n"
        self.user = user
async def query_param_handler(websocket):
    user = websocket.user
    ...

The client sets a cookie containing the token before opening the connection.

The cookie must be set by an iframe loaded from the same origin as the WebSocket server. This requires passing the token to this iframe.

// in main window
iframe.contentWindow.postMessage(token, "http://...");
// in iframe
document.cookie = `token=${data}; SameSite=Strict`;
// in main window
const websocket = new WebSocket("ws://.../");
// ...

This sequence must be synchronized between the main window and the iframe. This involves several events. Look at the full implementation for details.

The server intercepts the HTTP request, extracts the token and authenticates the user. If authentication fails, it returns an HTTP 401:

class CookieProtocol(websockets.WebSocketServerProtocol):
    async def process_request(self, path, headers):
        # Serve iframe on non-WebSocket requests
        ...
        token = get_cookie(headers.get("Cookie", ""), "token")
        if token is None:
            return http.HTTPStatus.UNAUTHORIZED, [], b"Missing token\n"
        user = get_user(token)
        if user is None:
            return http.HTTPStatus.UNAUTHORIZED, [], b"Invalid token\n"
        self.user = user
async def cookie_handler(websocket):
    user = websocket.user
    ...

The client adds the token to the WebSocket URI in user information before opening the connection:

const uri = `ws://token:${token}@.../`;
const websocket = new WebSocket(uri);
// ...

Since HTTP Basic Auth is designed to accept a username and a password rather than a token, we send token as username and the token as password.

The server intercepts the HTTP request, extracts the token and authenticates the user. If authentication fails, it returns an HTTP 401:

class UserInfoProtocol(websockets.BasicAuthWebSocketServerProtocol):
    async def check_credentials(self, username, password):
        if username != "token":
            return False
        user = get_user(password)
        if user is None:
            return False
        self.user = user
        return True
async def user_info_handler(websocket):
    user = websocket.user
    ...

When the WebSocket client is a standalone program rather than a script running in a browser, there are far fewer constraints. HTTP Authentication is the best solution in this scenario.

To authenticate a websockets client with HTTP Basic Authentication (RFC 7617), include the credentials in the URI:

async with websockets.connect(
    f"wss://{username}:{password}@example.com",
) as websocket:
    ...

(You must quote() username and password if they contain unsafe characters.)

To authenticate a websockets client with HTTP Bearer Authentication (RFC 6750), add a suitable Authorization header:

async with websockets.connect(
    "wss://example.com",
    extra_headers={"Authorization": f"Bearer {token}"}
) as websocket:
    ...

use broadcast()."

If you want to learn about its design in depth, continue reading this document.

WebSocket servers often send the same message to all connected clients or to a subset of clients for which the message is relevant.

Let's explore options for broadcasting a message, explain the design of broadcast(), and discuss alternatives.

For each option, we'll provide a connection handler called handler() and a function or coroutine called broadcast() that sends a message to all connected clients.

Integrating them is left as an exercise for the reader. You could start with:

import asyncio
import websockets
async def handler(websocket):
    ...
async def broadcast(message):
    ...
async def broadcast_messages():
    while True:
        await asyncio.sleep(1)
        message = ...  # your application logic goes here
        await broadcast(message)
async def main():
    async with websockets.serve(handler, "localhost", 8765):
        await broadcast_messages()  # runs forever
if __name__ == "__main__":
    asyncio.run(main())

broadcast_messages() must yield control to the event loop between each message, or else it will never let the server run. That's why it includes await asyncio.sleep(1).

A complete example is available in the experiments/broadcast directory.

The most obvious way to send a message to all connected clients consists in keeping track of them and sending the message to each of them.

Here's a connection handler that registers clients in a global variable:

CLIENTS = set()
async def handler(websocket):
    CLIENTS.add(websocket)
    try:
        await websocket.wait_closed()
    finally:
        CLIENTS.remove(websocket)

This implementation assumes that the client will never send any messages. If you'd rather not make this assumption, you can change:

await websocket.wait_closed()

to:

async for _ in websocket:
    pass

Here's a coroutine that broadcasts a message to all clients:

async def broadcast(message):
    for websocket in CLIENTS.copy():
        try:
            await websocket.send(message)
        except websockets.ConnectionClosed:
            pass

There are two tricks in this version of broadcast().

First, it makes a copy of CLIENTS before iterating it. Else, if a client connects or disconnects while broadcast() is running, the loop would fail with:

RuntimeError: Set changed size during iteration

Second, it ignores ConnectionClosed exceptions because a client could disconnect between the moment broadcast() makes a copy of CLIENTS and the moment it sends a message to this client. This is fine: a client that disconnected doesn't belongs to "all connected clients" anymore.

The naive way can be very fast. Indeed, if all connections have enough free space in their write buffers, await websocket.send(message) writes the message and returns immediately, as it doesn't need to wait for the buffer to drain. In this case, broadcast() doesn't yield control to the event loop, which minimizes overhead.

The naive way can also fail badly. If the write buffer of a connection reaches write_limit, broadcast() waits for the buffer to drain before sending the message to other clients. This can cause a massive drop in performance.

As a consequence, this pattern works only when write buffers never fill up, which is usually outside of the control of the server.

If you know for sure that you will never write more than write_limit bytes within ping_interval + ping_timeout, then websockets will terminate slow connections before the write buffer has time to fill up.

Don't set extreme write_limit, ping_interval, and ping_timeout values to ensure that this condition holds. Set reasonable values and use the built-in broadcast() function instead.

The naive way didn't work well because it serialized writes, while the whole point of asynchronous I/O is to perform I/O concurrently.

Let's modify broadcast() to send messages concurrently:

async def send(websocket, message):
    try:
        await websocket.send(message)
    except websockets.ConnectionClosed:
        pass
def broadcast(message):
    for websocket in CLIENTS:
        asyncio.create_task(send(websocket, message))

We move the error handling logic in a new coroutine and we schedule a Task to run it instead of executing it immediately.

Since broadcast() no longer awaits coroutines, we can make it a function rather than a coroutine and do away with the copy of CLIENTS.

This version of broadcast() makes clients independent from one another: a slow client won't block others. As a side effect, it makes messages independent from one another.

If you broadcast several messages, there is no strong guarantee that they will be sent in the expected order. Fortunately, the event loop runs tasks in the order in which they are created, so the order is correct in practice.

Technically, this is an implementation detail of the event loop. However, it seems unlikely for an event loop to run tasks in an order other than FIFO.

If you wanted to enforce the order without relying this implementation detail, you could be tempted to wait until all clients have received the message:

async def broadcast(message):
    if CLIENTS:  # asyncio.wait doesn't accept an empty list
        await asyncio.wait([
            asyncio.create_task(send(websocket, message))
            for websocket in CLIENTS
        ])

However, this doesn't really work in practice. Quite often, it will block until the slowest client times out.

At this point, it becomes apparent that backpressure, usually a good practice, doesn't work well when broadcasting a message to thousands of clients.

When you're sending messages to a single client, you don't want to send them faster than the network can transfer them and the client accept them. This is why send() checks if the write buffer is full and, if it is, waits until it drain, giving the network and the client time to catch up. This provides backpressure.

Without backpressure, you could pile up data in the write buffer until the server process runs out of memory and the operating system kills it.

The send() API is designed to enforce backpressure by default. This helps users of websockets write robust programs even if they never heard about backpressure.

For comparison, asyncio.StreamWriter requires users to understand backpressure and to await drain() explicitly after each write().

When broadcasting messages, backpressure consists in slowing down all clients in an attempt to let the slowest client catch up. With thousands of clients, the slowest one is probably timing out and isn't going to receive the message anyway. So it doesn't make sense to synchronize with the slowest client.

How do we avoid running out of memory when slow clients can't keep up with the broadcast rate, then? The most straightforward option is to disconnect them.

If a client gets too far behind, eventually it reaches the limit defined by ping_timeout and websockets terminates the connection. You can read the discussion of keepalive and timeouts for details.

The built-in broadcast() function is similar to the naive way. The main difference is that it doesn't apply backpressure.

This provides the best performance by avoiding the overhead of scheduling and running one task per client.

Also, when sending text messages, encoding to UTF-8 happens only once rather than once per client, providing a small performance gain.

Per-client queues

At this point, we deal with slow clients rather brutally: we disconnect then.

Can we do better? For example, we could decide to skip or to batch messages, depending on how far behind a client is.

To implement this logic, we can create a queue of messages for each client and run a task that gets messages from the queue and sends them to the client:

import asyncio
CLIENTS = set()
async def relay(queue, websocket):
    while True:
        # Implement custom logic based on queue.qsize() and
        # websocket.transport.get_write_buffer_size() here.
        message = await queue.get()
        await websocket.send(message)
async def handler(websocket):
    queue = asyncio.Queue()
    relay_task = asyncio.create_task(relay(queue, websocket))
    CLIENTS.add(queue)
    try:
        await websocket.wait_closed()
    finally:
        CLIENTS.remove(queue)
        relay_task.cancel()

Then we can broadcast a message by pushing it to all queues:

def broadcast(message):
    for queue in CLIENTS:
        queue.put_nowait(message)

The queues provide an additional buffer between the broadcast() function and clients. This makes it easier to support slow clients without excessive memory usage because queued messages aren't duplicated to write buffers until relay() processes them.

Can we avoid centralizing the list of connected clients in a global variable?

If each client subscribes to a stream a messages, then broadcasting becomes as simple as publishing a message to the stream.

Here's a message stream that supports multiple consumers:

class PubSub:
    def __init__(self):
        self.waiter = asyncio.Future()
    def publish(self, value):
        waiter, self.waiter = self.waiter, asyncio.Future()
        waiter.set_result((value, self.waiter))
    async def subscribe(self):
        waiter = self.waiter
        while True:
            value, waiter = await waiter
            yield value
    __aiter__ = subscribe
PUBSUB = PubSub()

The stream is implemented as a linked list of futures. It isn't necessary to synchronize consumers. They can read the stream at their own pace, independently from one another. Once all consumers read a message, there are no references left, therefore the garbage collector deletes it.

The connection handler subscribes to the stream and sends messages:

async def handler(websocket):
    async for message in PUBSUB:
        await websocket.send(message)

The broadcast function publishes to the stream:

def broadcast(message):
    PUBSUB.publish(message)

Like per-client queues, this version supports slow clients with limited memory usage. Unlike per-client queues, it makes it difficult to tell how far behind a client is. The PubSub class could be extended or refactored to provide this information.

The for loop is gone from this version of the broadcast() function. However, there's still a for loop iterating on all clients hidden deep inside asyncio. When publish() sets the result of the waiter future, asyncio loops on callbacks registered with this future and schedules them. This is how connection handlers receive the next value from the asynchronous iterator returned by subscribe().

The built-in broadcast() function sends all messages without yielding control to the event loop. So does the naive way when the network and clients are fast and reliable.

For each client, a WebSocket frame is prepared and sent to the network. This is the minimum amount of work required to broadcast a message.

It would be tempting to prepare a frame and reuse it for all connections. However, this isn't possible in general for two reasons:

  • Clients can negotiate different extensions. You would have to enforce the same extensions with the same parameters. For example, you would have to select some compression settings and reject clients that cannot support these settings.
  • Extensions can be stateful, producing different encodings of the same message depending on previous messages. For example, you would have to disable context takeover to make compression stateless, resulting in poor compression rates.

All other patterns discussed above yield control to the event loop once per client because messages are sent by different tasks. This makes them slower than the built-in broadcast() function.

There is no major difference between the performance of per-client queues and publish–subscribe.

Most WebSocket servers exchange JSON messages because they're convenient to parse and serialize in a browser. These messages contain text data and tend to be repetitive.

This makes the stream of messages highly compressible. Enabling compression can reduce network traffic by more than 80%.

There's a standard for compressing messages. RFC 7692 defines WebSocket Per-Message Deflate, a compression extension based on the Deflate algorithm.

connect() and serve() enable compression by default because the reduction in network bandwidth is usually worth the additional memory and CPU cost.

If you want to disable compression, set compression=None:

import websockets
websockets.connect(..., compression=None)
websockets.serve(..., compression=None)

If you want to customize compression settings, you can enable the Per-Message Deflate extension explicitly with ClientPerMessageDeflateFactory or ServerPerMessageDeflateFactory:

import websockets
from websockets.extensions import permessage_deflate
websockets.connect(
    ...,
    extensions=[
        permessage_deflate.ClientPerMessageDeflateFactory(
            server_max_window_bits=11,
            client_max_window_bits=11,
            compress_settings={"memLevel": 4},
        ),
    ],
)
websockets.serve(
    ...,
    extensions=[
        permessage_deflate.ServerPerMessageDeflateFactory(
            server_max_window_bits=11,
            client_max_window_bits=11,
            compress_settings={"memLevel": 4},
        ),
    ],
)

The Window Bits and Memory Level values in these examples reduce memory usage at the expense of compression rate.

When a client and a server enable the Per-Message Deflate extension, they negotiate two parameters to guarantee compatibility between compression and decompression. These parameters affect the trade-off between compression rate and memory usage for both sides.

  • Context Takeover means that the compression context is retained between messages. In other words, compression is applied to the stream of messages rather than to each message individually.

    Context takeover should remain enabled to get good performance on applications that send a stream of messages with similar structure, that is, most applications.

    This requires retaining the compression context and state between messages, which increases the memory footprint of a connection.

  • Window Bits controls the size of the compression context. It must be an integer between 9 (lowest memory usage) and 15 (best compression). Setting it to 8 is possible but rejected by some versions of zlib.

    On the server side, websockets defaults to 12. Specifically, the compression window size (server to client) is always 12 while the decompression window (client to server) size may be 12 or 15 depending on whether the client supports configuring it.

    On the client side, websockets lets the server pick a suitable value, which has the same effect as defaulting to 15.

zlib offers additional parameters for tuning compression. They control the trade-off between compression rate, memory usage, and CPU usage only for compressing. They're transparent for decompressing. Unless mentioned otherwise, websockets inherits defaults of compressobj().

  • Memory Level controls the size of the compression state. It must be an integer between 1 (lowest memory usage) and 9 (best compression).

    websockets defaults to 5. This is lower than zlib's default of 8. Not only does a lower memory level reduce memory usage, but it can also increase speed thanks to memory locality.

  • Compression Level controls the effort to optimize compression. It must be an integer between 1 (lowest CPU usage) and 9 (best compression).
  • Strategy selects the compression strategy. The best choice depends on the type of data being compressed.

By default, websockets enables compression with conservative settings that optimize memory usage at the cost of a slightly worse compression rate: Window Bits = 12 and Memory Level = 5. This strikes a good balance for small messages that are typical of WebSocket servers.

Here's how various compression settings affect memory usage of a single connection on a 64-bit system, as well a benchmark of compressed size and compression time for a corpus of small JSON documents.

Window Bits Memory Level Memory usage Size vs. default Time vs. default
15 8 322 KiB -4.0% +15%
14 7 178 KiB -2.6% +10%
13 6 106 KiB -1.4% +5%
12 5 70 KiB = =
11 4 52 KiB +3.7% -5%
10 3 43 KiB +90% +50%
9 2 39 KiB +160% +100%
— — 19 KiB +452% —

Window Bits and Memory Level don't have to move in lockstep. However, other combinations don't yield significantly better results than those shown above.

Compressed size and compression time depend heavily on the kind of messages exchanged by the application so this example may not apply to your use case.

You can adapt compression/benchmark.py by creating a list of typical messages and passing it to the _run function.

Window Bits = 11 and Memory Level = 4 looks like the sweet spot in this table.

websockets defaults to Window Bits = 12 and Memory Level = 5 to stay away from Window Bits = 10 or Memory Level = 3 where performance craters, raising doubts on what could happen at Window Bits = 11 and Memory Level = 4 on a different corpus.

Defaults must be safe for all applications, hence a more conservative choice.

The benchmark focuses on compression because it's more expensive than decompression. Indeed, leaving aside small allocations, theoretical memory usage is:

  • (1 << (windowBits + 2)) + (1 << (memLevel + 9)) for compression;
  • 1 << windowBits for decompression.

CPU usage is also higher for compression than decompression.

While it's always possible for a server to use a smaller window size for compressing outgoing messages, using a smaller window size for decompressing incoming messages requires collaboration from clients.

When a client doesn't support configuring the size of its compression window, websockets enables compression with the largest possible decompression window. In most use cases, this is more efficient than disabling compression both ways.

If you are very sensitive to memory usage, you can reverse this behavior by setting the require_client_max_window_bits parameter of ServerPerMessageDeflateFactory to True.

By default, websockets enables compression with Memory Level = 5 but leaves the Window Bits setting up to the server.

There's two good reasons and one bad reason for not optimizing the client side like the server side:

1.
If the maintainers of a server configured some optimized settings, we don't want to override them with more restrictive settings.
2.
Optimizing memory usage doesn't matter very much for clients because it's uncommon to open thousands of client connections in a program.
3.
On a more pragmatic note, some servers misbehave badly when a client configures compression settings. AWS API Gateway is the worst offender.

Unfortunately, even though websockets is right and AWS is wrong, many users jump to the conclusion that websockets doesn't work.

Until the ecosystem levels up, interoperability with buggy servers seems more valuable than optimizing memory usage.

This blog post by Ilya Grigorik provides more details about how compression settings affect memory usage and how to optimize them.

This experiment by Peter Thorson recommends Window Bits = 11 and Memory Level = 4 for optimizing memory usage.

Since the WebSocket protocol is intended for real-time communications over long-lived connections, it is desirable to ensure that connections don't break, and if they do, to report the problem quickly.

Connections can drop as a consequence of temporary network connectivity issues, which are very common, even within data centers.

Furthermore, WebSocket builds on top of HTTP/1.1 where connections are short-lived, even with Connection: keep-alive. Typically, HTTP/1.1 infrastructure closes idle connections after 30 to 120 seconds.

As a consequence, proxies may terminate WebSocket connections prematurely when no message was exchanged in 30 seconds.

To avoid these problems, websockets runs a keepalive and heartbeat mechanism based on WebSocket Ping and Pong frames, which are designed for this purpose.

It loops through these steps:

1.
Wait 20 seconds.
2.
Send a Ping frame.
3.
Receive a corresponding Pong frame within 20 seconds.

If the Pong frame isn't received, websockets considers the connection broken and closes it.

This mechanism serves two purposes:

1.
It creates a trickle of traffic so that the TCP connection isn't idle and network infrastructure along the path keeps it open ("keepalive").
2.
It detects if the connection drops or becomes so slow that it's unusable in practice ("heartbeat"). In that case, it terminates the connection and your application gets a ConnectionClosed exception.

Timings are configurable with the ping_interval and ping_timeout arguments of connect() and serve(). Shorter values will detect connection drops faster but they will increase network traffic and they will be more sensitive to latency.

Setting ping_interval to None disables the whole keepalive and heartbeat mechanism.

Setting ping_timeout to None disables only timeouts. This enables keepalive, to keep idle connections open, and disables heartbeat, to support large latency spikes.

TCP keepalive is disabled by default on most operating systems. When enabled, the default interval is two hours or more, which is far too much.

Browsers don't enable a keepalive mechanism like websockets by default. As a consequence, they can fail to notice that a WebSocket connection is broken for an extended period of time, until the TCP connection times out.

In this scenario, the WebSocket object in the browser doesn't fire a close event. If you have a reconnection mechanism, it doesn't kick in because it believes that the connection is still working.

If your browser-based app mysteriously and randomly fails to receive events, this is a likely cause. You need a keepalive mechanism in the browser to avoid this scenario.

Unfortunately, the WebSocket API in browsers doesn't expose the native Ping and Pong functionality in the WebSocket protocol. You have to roll your own in the application layer.

Latency between a client and a server may increase for two reasons:

  • Network connectivity is poor. When network packets are lost, TCP attempts to retransmit them, which manifests as latency. Excessive packet loss makes the connection unusable in practice. At some point, timing out is a reasonable choice.
  • Traffic is high. For example, if a client sends messages on the connection faster than a server can process them, this manifests as latency as well, because data is waiting in flight, mostly in OS buffers.

    If the server is more than 20 seconds behind, it doesn't see the Pong before the default timeout elapses. As a consequence, it closes the connection. This is a reasonable choice to prevent overload.

    If traffic spikes cause unwanted timeouts and you're confident that the server will catch up eventually, you can increase ping_timeout or you can set it to None to disable heartbeat entirely.

    The same reasoning applies to situations where the server sends more traffic than the client can accept.

The latency measured during the last exchange of Ping and Pong frames is available in the latency attribute. Alternatively, you can measure the latency at any time with the ping method.

This document describes the design of websockets. It assumes familiarity with the specification of the WebSocket protocol in RFC 6455.

It's primarily intended at maintainers. It may also be useful for users who wish to understand what happens under the hood.

WARNING:

Internals described in this document may change at any time.

Backwards compatibility is only guaranteed for public APIs.

WebSocket connections go through a trivial state machine:

  • CONNECTING: initial state,
  • OPEN: when the opening handshake is complete,
  • CLOSING: when the closing handshake is started,
  • CLOSED: when the TCP connection is closed.

Transitions happen in the following places:

  • CONNECTING -> OPEN: in connection_open() which runs when the opening handshake completes and the WebSocket connection is established — not to be confused with connection_made() which runs when the TCP connection is established;
  • OPEN -> CLOSING: in write_frame() immediately before sending a close frame; since receiving a close frame triggers sending a close frame, this does the right thing regardless of which side started the closing handshake; also in fail_connection() which duplicates a few lines of code from write_close_frame() and write_frame();
  • * -> CLOSED: in connection_lost() which is always called exactly once when the TCP connection is closed.

The following diagram shows which coroutines are running at each stage of the connection lifecycle on the client side.

The lifecycle is identical on the server side, except inversion of control makes the equivalent of connect() implicit.

Coroutines shown in green are called by the application. Multiple coroutines may interact with the WebSocket connection concurrently.

Coroutines shown in gray manage the connection. When the opening handshake succeeds, connection_open() starts two tasks:

  • transfer_data_task runs transfer_data() which handles incoming data and lets recv() consume it. It may be canceled to terminate the connection. It never exits with an exception other than CancelledError. See data transfer below.
  • keepalive_ping_task runs keepalive_ping() which sends Ping frames at regular intervals and ensures that corresponding Pong frames are received. It is canceled when the connection terminates. It never exits with an exception other than CancelledError.
  • close_connection_task runs close_connection() which waits for the data transfer to terminate, then takes care of closing the TCP connection. It must not be canceled. It never exits with an exception. See connection termination below.

Besides, fail_connection() starts the same close_connection_task when the opening handshake fails, in order to close the TCP connection.

Splitting the responsibilities between two tasks makes it easier to guarantee that websockets can terminate connections:

  • within a fixed timeout,
  • without leaking pending tasks,
  • without leaking open TCP connections,

regardless of whether the connection terminates normally or abnormally.

transfer_data_task completes when no more data will be received on the connection. Under normal circumstances, it exits after exchanging close frames.

close_connection_task completes when the TCP connection is closed.

websockets performs the opening handshake when establishing a WebSocket connection. On the client side, connect() executes it before returning the protocol to the caller. On the server side, it's executed before passing the protocol to the ws_handler coroutine handling the connection.

While the opening handshake is asymmetrical — the client sends an HTTP Upgrade request and the server replies with an HTTP Switching Protocols response — websockets aims at keeping the implementation of both sides consistent with one another.

On the client side, handshake():

  • builds an HTTP request based on the uri and parameters passed to connect();
  • writes the HTTP request to the network;
  • reads an HTTP response from the network;
  • checks the HTTP response, validates extensions and subprotocol, and configures the protocol accordingly;
  • moves to the OPEN state.

On the server side, handshake():

  • reads an HTTP request from the network;
  • calls process_request() which may abort the WebSocket handshake and return an HTTP response instead; this hook only makes sense on the server side;
  • checks the HTTP request, negotiates extensions and subprotocol, and configures the protocol accordingly;
  • builds an HTTP response based on the above and parameters passed to serve();
  • writes the HTTP response to the network;
  • moves to the OPEN state;
  • returns the path part of the uri.

The most significant asymmetry between the two sides of the opening handshake lies in the negotiation of extensions and, to a lesser extent, of the subprotocol. The server knows everything about both sides and decides what the parameters should be for the connection. The client merely applies them.

If anything goes wrong during the opening handshake, websockets fails the connection.

Once the opening handshake has completed, the WebSocket protocol enters the data transfer phase. This part is almost symmetrical. There are only two differences between a server and a client:

  • client-to-server masking: the client masks outgoing frames; the server unmasks incoming frames;
  • closing the TCP connection: the server closes the connection immediately; the client waits for the server to do it.

These differences are so minor that all the logic for data framing, for sending and receiving data and for closing the connection is implemented in the same class, WebSocketCommonProtocol.

The is_client attribute tells which side a protocol instance is managing. This attribute is defined on the WebSocketServerProtocol and WebSocketClientProtocol classes.

The following diagram shows how data flows between an application built on top of websockets and a remote endpoint. It applies regardless of which side is the server or the client.

Public methods are shown in green, private methods in yellow, and buffers in orange. Methods related to connection termination are omitted; connection termination is discussed in another section below.

The left side of the diagram shows how websockets receives data.

Incoming data is written to a StreamReader in order to implement flow control and provide backpressure on the TCP connection.

transfer_data_task, which is started when the WebSocket connection is established, processes this data.

When it receives data frames, it reassembles fragments and puts the resulting messages in the messages queue.

When it encounters a control frame:

  • if it's a close frame, it starts the closing handshake;
  • if it's a ping frame, it answers with a pong frame;
  • if it's a pong frame, it acknowledges the corresponding ping (unless it's an unsolicited pong).

Running this process in a task guarantees that control frames are processed promptly. Without such a task, websockets would depend on the application to drive the connection by having exactly one coroutine awaiting recv() at any time. While this happens naturally in many use cases, it cannot be relied upon.

Then recv() fetches the next message from the messages queue, with some complexity added for handling backpressure and termination correctly.

The right side of the diagram shows how websockets sends data.

send() writes one or several data frames containing the message. While sending a fragmented message, concurrent calls to send() are put on hold until all fragments are sent. This makes concurrent calls safe.

ping() writes a ping frame and yields a Future which will be completed when a matching pong frame is received.

pong() writes a pong frame.

close() writes a close frame and waits for the TCP connection to terminate.

Outgoing data is written to a StreamWriter in order to implement flow control and provide backpressure from the TCP connection.

When the other side of the connection initiates the closing handshake, read_message() receives a close frame while in the OPEN state. It moves to the CLOSING state, sends a close frame, and returns None, causing transfer_data_task to terminate.

When this side of the connection initiates the closing handshake with close(), it moves to the CLOSING state and sends a close frame. When the other side sends a close frame, read_message() receives it in the CLOSING state and returns None, also causing transfer_data_task to terminate.

If the other side doesn't send a close frame within the connection's close timeout, websockets fails the connection.

The closing handshake can take up to 2 * close_timeout: one close_timeout to write a close frame and one close_timeout to receive a close frame.

Then websockets terminates the TCP connection.

close_connection_task, which is started when the WebSocket connection is established, is responsible for eventually closing the TCP connection.

First close_connection_task waits for transfer_data_task to terminate, which may happen as a result of:

  • a successful closing handshake: as explained above, this exits the infinite loop in transfer_data_task;
  • a timeout while waiting for the closing handshake to complete: this cancels transfer_data_task;
  • a protocol error, including connection errors: depending on the exception, transfer_data_task fails the connection with a suitable code and exits.

close_connection_task is separate from transfer_data_task to make it easier to implement the timeout on the closing handshake. Canceling transfer_data_task creates no risk of canceling close_connection_task and failing to close the TCP connection, thus leaking resources.

Then close_connection_task cancels keepalive_ping(). This task has no protocol compliance responsibilities. Terminating it to avoid leaking it is the only concern.

Terminating the TCP connection can take up to 2 * close_timeout on the server side and 3 * close_timeout on the client side. Clients start by waiting for the server to close the connection, hence the extra close_timeout. Then both sides go through the following steps until the TCP connection is lost: half-closing the connection (only for non-TLS connections), closing the connection, aborting the connection. At this point the connection drops regardless of what happens on the network.

If the opening handshake doesn't complete successfully, websockets fails the connection by closing the TCP connection.

Once the opening handshake has completed, websockets fails the connection by canceling transfer_data_task and sending a close frame if appropriate.

transfer_data_task exits, unblocking close_connection_task, which closes the TCP connection.

WebSocketServer closes asynchronously like asyncio.Server. The shutdown happen in two steps:

1.
Stop listening and accepting new connections;
2.
Close established connections with close code 1001 (going away) or, if the opening handshake is still in progress, with HTTP status code 503 (Service Unavailable).

The first call to close starts a task that performs this sequence. Further calls are ignored. This is the easiest way to make close and wait_closed idempotent.

websockets provides a WebSocket application server. It manages connections and passes them to user-provided connection handlers. This is an inversion of control scenario: library code calls user code.

If a connection drops, the corresponding handler should terminate. If the server shuts down, all connection handlers must terminate. Canceling connection handlers would terminate them.

However, using cancellation for this purpose would require all connection handlers to handle it properly. For example, if a connection handler starts some tasks, it should catch CancelledError, terminate or cancel these tasks, and then re-raise the exception.

Cancellation is tricky in asyncio applications, especially when it interacts with finalization logic. In the example above, what if a handler gets interrupted with CancelledError while it's finalizing the tasks it started, after detecting that the connection dropped?

websockets considers that cancellation may only be triggered by the caller of a coroutine when it doesn't care about the results of that coroutine anymore. (Source: Guido van Rossum). Since connection handlers run arbitrary user code, websockets has no way of deciding whether that code is still doing something worth caring about.

For these reasons, websockets never cancels connection handlers. Instead it expects them to detect when the connection is closed, execute finalization logic if needed, and exit.

Conversely, cancellation isn't a concern for WebSocket clients because they don't involve inversion of control.

Most public APIs of websockets are coroutines. They may be canceled, for example if the user starts a task that calls these coroutines and cancels the task later. websockets must handle this situation.

Cancellation during the opening handshake is handled like any other exception: the TCP connection is closed and the exception is re-raised. This can only happen on the client side. On the server side, the opening handshake is managed by websockets and nothing results in a cancellation.

Once the WebSocket connection is established, internal tasks transfer_data_task and close_connection_task mustn't get accidentally canceled if a coroutine that awaits them is canceled. In other words, they must be shielded from cancellation.

recv() waits for the next message in the queue or for transfer_data_task to terminate, whichever comes first. It relies on wait() for waiting on two futures in parallel. As a consequence, even though it's waiting on a Future signaling the next message and on transfer_data_task, it doesn't propagate cancellation to them.

ensure_open() is called by send(), ping(), and pong(). When the connection state is CLOSING, it waits for transfer_data_task but shields it to prevent cancellation.

close() waits for the data transfer task to terminate with timeout(). If it's canceled or if the timeout elapses, transfer_data_task is canceled, which is correct at this point. close() then waits for close_connection_task but shields it to prevent cancellation.

close() and fail_connection() are the only places where transfer_data_task may be canceled.

close_connection_task starts by waiting for transfer_data_task. It catches CancelledError to prevent a cancellation of transfer_data_task from propagating to close_connection_task.

NOTE:

This section discusses backpressure from the perspective of a server but the concept applies to clients symmetrically.

With a naive implementation, if a server receives inputs faster than it can process them, or if it generates outputs faster than it can send them, data accumulates in buffers, eventually causing the server to run out of memory and crash.

The solution to this problem is backpressure. Any part of the server that receives inputs faster than it can process them and send the outputs must propagate that information back to the previous part in the chain.

websockets is designed to make it easy to get backpressure right.

For incoming data, websockets builds upon StreamReader which propagates backpressure to its own buffer and to the TCP stream. Frames are parsed from the input stream and added to a bounded queue. If the queue fills up, parsing halts until the application reads a frame.

For outgoing data, websockets builds upon StreamWriter which implements flow control. If the output buffers grow too large, it waits until they're drained. That's why all APIs that write frames are asynchronous.

Of course, it's still possible for an application to create its own unbounded buffers and break the backpressure. Be careful with queues.

NOTE:

This section discusses buffers from the perspective of a server but it applies to clients as well.

An asynchronous systems works best when its buffers are almost always empty.

For example, if a client sends data too fast for a server, the queue of incoming messages will be constantly full. The server will always be 32 messages (by default) behind the client. This consumes memory and increases latency for no good reason. The problem is called bufferbloat.

If buffers are almost always full and that problem cannot be solved by adding capacity — typically because the system is bottlenecked by the output and constantly regulated by backpressure — reducing the size of buffers minimizes negative consequences.

By default websockets has rather high limits. You can decrease them according to your application's characteristics.

Bufferbloat can happen at every level in the stack where there is a buffer. For each connection, the receiving side contains these buffers:

  • OS buffers: tuning them is an advanced optimization.
  • StreamReader bytes buffer: the default limit is 64 KiB. You can set another limit by passing a read_limit keyword argument to connect() or serve().
  • Incoming messages deque: its size depends both on the size and the number of messages it contains. By default the maximum UTF-8 encoded size is 1 MiB and the maximum number is 32. In the worst case, after UTF-8 decoding, a single message could take up to 4 MiB of memory and the overall memory consumption could reach 128 MiB. You should adjust these limits by setting the max_size and max_queue keyword arguments of connect() or serve() according to your application's requirements.

For each connection, the sending side contains these buffers:

  • StreamWriter bytes buffer: the default size is 64 KiB. You can set another limit by passing a write_limit keyword argument to connect() or serve().
  • OS buffers: tuning them is an advanced optimization.

Awaiting any combination of recv(), send(), close() ping(), or pong() concurrently is safe, including multiple calls to the same method, with one exception and one limitation.

  • Only one coroutine can receive messages at a time. This constraint avoids non-deterministic behavior (and simplifies the implementation). If a coroutine is awaiting recv(), awaiting it again in another coroutine raises RuntimeError.
  • Sending a fragmented message forces serialization. Indeed, the WebSocket protocol doesn't support multiplexing messages. If a coroutine is awaiting send() to send a fragmented message, awaiting it again in another coroutine waits until the first call completes. This will be transparent in many cases. It may be a concern if the fragmented message is generated slowly by an asynchronous iterator.

Receiving frames is independent from sending frames. This isolates recv(), which receives frames, from the other methods, which send frames.

While the connection is open, each frame is sent with a single write. Combined with the concurrency model of asyncio, this enforces serialization. The only other requirement is to prevent interleaving other data frames in the middle of a fragmented message.

After the connection is closed, sending a frame raises ConnectionClosed, which is safe.

Memory usage

In most cases, memory usage of a WebSocket server is proportional to the number of open connections. When a server handles thousands of connections, memory usage can become a bottleneck.

Memory usage of a single connection is the sum of:

1.
the baseline amount of memory websockets requires for each connection,
2.
the amount of data held in buffers before the application processes it,
3.
any additional memory allocated by the application itself.

Compression settings are the main factor affecting the baseline amount of memory used by each connection.

With websockets' defaults, on the server side, a single connections uses 70 KiB of memory.

Refer to the topic guide on compression to learn more about tuning compression settings.

Under normal circumstances, buffers are almost always empty.

Under high load, if a server receives more messages than it can process, bufferbloat can result in excessive memory usage.

By default websockets has generous limits. It is strongly recommended to adapt them to your application. When you call serve():

  • Set max_size (default: 1 MiB, UTF-8 encoded) to the maximum size of messages your application generates.
  • Set max_queue (default: 32) to the maximum number of messages your application expects to receive faster than it can process them. The queue provides burst tolerance without slowing down the TCP connection.

Furthermore, you can lower read_limit and write_limit (default: 64 KiB) to reduce the size of buffers for incoming and outgoing data.

The design document provides more details about buffers.

For production use, a server should require encrypted connections.

See this example of encrypting connections with TLS.

Memory usage

WARNING:

An attacker who can open an arbitrary number of connections will be able to perform a denial of service by memory exhaustion. If you're concerned by denial of service attacks, you must reject suspicious connections before they reach websockets, typically in a reverse proxy.

With the default settings, opening a connection uses 70 KiB of memory.

Sending some highly compressed messages could use up to 128 MiB of memory with an amplification factor of 1000 between network traffic and memory usage.

Configuring a server to optimize memory usage will improve security in addition to improving performance.

websockets implements additional limits on the amount of data it accepts in order to minimize exposure to security vulnerabilities.

In the opening handshake, websockets limits the number of HTTP headers to 256 and the size of an individual header to 4096 bytes. These limits are 10 to 20 times larger than what's expected in standard use cases. They're hard-coded.

If you need to change these limits, you can monkey-patch the constants in websockets.http11.

Here are tips to optimize performance.

You can make a websockets application faster by running it with uvloop.

(This advice isn't specific to websockets. It applies to any asyncio application.)

broadcast() is the most efficient way to send a message to many clients.

This is about websockets-the-project rather than websockets-the-software.

Backwards-compatibility policy

websockets is intended for production use. Therefore, stability is a goal.

websockets also aims at providing the best API for WebSocket in Python.

While we value stability, we value progress more. When an improvement requires changing a public API, we make the change and document it in this changelog.

When possible with reasonable effort, we preserve backwards-compatibility for five years after the release that introduced the change.

When a release contains backwards-incompatible API changes, the major version is increased, else the minor version is increased. Patch versions are only for fixing regressions shortly after a release.

Only documented APIs are public. Undocumented, private APIs may change without notice.

October 21, 2023

Backwards-incompatible changes

websockets 11.0 is the last version supporting Python 3.7.

  • Made convenience imports from websockets compatible with static code analysis tools such as auto-completion in an IDE or type checking with mypy.
  • Accepted a plain int where an HTTPStatus is expected.
  • Added CloseCode.

May 7, 2023

Bug fixes

•
Fixed the threading implementation of servers on Windows.

April 18, 2023

Bug fixes

•
Fixed a deadlock in the threading implementation when closing a connection without reading all messages.

April 6, 2023

Bug fixes

•
Restored the C extension in the source distribution.

April 2, 2023

Backwards-incompatible changes

Aliases provide compatibility for all previously public APIs according to the backwards-compatibility policy.

  • The connection module was renamed to protocol.
  • The connection.Connection, server.ServerConnection, and client.ClientConnection classes were renamed to protocol.Protocol, server.ServerProtocol, and client.ClientProtocol.

If you instantiate ServerProtocol or ClientProtocol directly, make sure you are using keyword arguments.

Receiving an empty close frame now results in ConnectionClosedOK instead of ConnectionClosedError.

As a consequence, calling WebSocket.close() without arguments in a browser isn't reported as an error anymore.

You can adjust the timeout with the open_timeout parameter. Set it to None to disable the timeout entirely.

New features

It may be more convenient if you don't need to manage many connections and you're more comfortable with threading than asyncio.

It is particularly suited to client applications that establish only one connection. It may be used for servers handling few connections.

See connect() and serve() for details.

  • Added open_timeout to serve().
  • Made it possible to close a server without closing existing connections.
  • Added select_subprotocol to customize negotiation of subprotocols in the Sans-I/O layer.

  • Added platform-independent wheels.
  • Improved error handling in broadcast().
  • Set server_hostname automatically on TLS connections when providing a sock argument to connect().

October 25, 2022

New features

  • Validated compatibility with Python 3.11.
  • Added the latency property to protocols.
  • Changed ping to return the latency of the connection.
  • Supported overriding or removing the User-Agent header in clients and the Server header in servers.
  • Added deployment guides for more Platform as a Service providers.

•
Improved FAQ.

April 17, 2022

Backwards-incompatible changes

Use the handshake_exc attribute of ServerProtocol and ClientProtocol instead.

See Integrate the Sans-I/O layer for details.

•
Reduced noise in logs when ssl or zlib raise exceptions.

February 21, 2022

  • Made compression negotiation more lax for compatibility with Firefox.
  • Improved FAQ and quick start guide.

Bug fixes

  • Fixed backwards-incompatibility in 10.1 for connection handlers created with functools.partial().
  • Avoided leaking open sockets when connect() is canceled.

November 14, 2021

New features

  • Added a tutorial.
  • Made the second parameter of connection handlers optional. It will be deprecated in the next major release. The request path is available in the path attribute of the first argument.

    If you implemented the connection handler of a server as:

async def handler(request, path):
    ...

You should replace it by:

async def handler(request):
    path = request.path  # if handler() uses the path argument
    ...
•
Added python -m websockets --version.

  • Added wheels for Python 3.10, PyPy 3.7, and for more platforms.
  • Reverted optimization of default compression settings for clients, mainly to avoid triggering bugs in poorly implemented servers like AWS API Gateway.
  • Mirrored the entire Server API in WebSocketServer.
  • Improved performance for large messages on ARM processors.
  • Documented how to auto-reload on code changes in development.

Bug fixes

•
Avoided half-closing TCP connections that are already closed.

September 9, 2021

Backwards-incompatible changes

websockets 9.1 is the last version supporting Python 3.6.

This reflects a decision made in Python 3.8. See the release notes of Python 3.10 for details.

The loop parameter is also removed from WebSocketServer. This should be transparent.

connect() times out after 10 seconds by default.

You can adjust the timeout with the open_timeout parameter. Set it to None to disable the timeout entirely.

See the release notes of websockets 3.0 for details.

If you raise ConnectionClosed or a subclass, rather than catch them when websockets raises them, you must change your code.

If you raise InvalidURI, rather than catch it when websockets raises it, you must change your code.

New features

in third-party libraries."

If you're integrating websockets in a library, rather than just using it, look at the Sans-I/O integration guide.

  • Added compatibility with Python 3.10.
  • Added broadcast() to send a message to many clients.
  • Added support for reconnecting automatically by using connect() as an asynchronous iterator.
  • Added open_timeout to connect().
  • Documented how to integrate with Django.
  • Documented how to deploy websockets in production, with several options.
  • Documented how to authenticate connections.
  • Documented how to broadcast messages to many connections.

  • Improved logging. See the logging guide.
  • Optimized default compression settings to reduce memory usage.
  • Optimized processing of client-to-server messages when the C extension isn't available.
  • Supported relative redirects in connect().
  • Handled TCP connection drops during the opening handshake.
  • Made it easier to customize authentication with check_credentials().
  • Provided additional information in ConnectionClosed exceptions.
  • Clarified several exceptions or log messages.
  • Restructured documentation.
  • Improved API documentation.
  • Extended FAQ.

Bug fixes

•
Avoided a crash when receiving a ping while the connection is closing.

May 27, 2021

Security fix

Version 8.0 was vulnerable to timing attacks on HTTP Basic Auth passwords (CVE-2021-33880).

May 15, 2021

Bug fixes

  • Restored compatibility of python -m websockets with Python < 3.9.
  • Restored compatibility with mypy.

May 2, 2021

Bug fixes

•
Fixed issues with the packaging of the 9.0 release.

May 1, 2021

Backwards-incompatible changes

Aliases provide compatibility for all previously public APIs according to the backwards-compatibility policy

  • Headers and MultipleValuesError are moved from websockets.http to websockets.datastructures. If you're using them, you should adjust the import path.
  • The client, server, protocol, and auth modules were moved from the websockets package to a websockets.legacy sub-package. Despite the name, they're still fully supported.
  • The framing, handshake, headers, http, and uri modules in the websockets package are deprecated. These modules provided low-level APIs for reuse by other projects, but they didn't reach that goal. Keeping these APIs public makes it more difficult to improve websockets.

These changes pave the path for a refactoring that should be a transparent upgrade for most uses and facilitate integration by other projects.

While Python supports this, tools relying on static code analysis don't. This breaks auto-completion in an IDE or type checking with mypy.

If you depend on such tools, use the real import paths, which can be found in the API documentation, for example:

from websockets.client import connect
from websockets.server import serve

New features

•
Added compatibility with Python 3.9.

  • Added support for IRIs in addition to URIs.
  • Added close codes 1012, 1013, and 1014.
  • Raised an error when passing a dict to send().
  • Improved error reporting.

Bug fixes

  • Fixed sending fragmented, compressed messages.
  • Fixed Host header sent when connecting to an IPv6 address.
  • Fixed creating a client or a server with an existing Unix socket.
  • Aligned maximum cookie size with popular web browsers.
  • Ensured cancellation always propagates, even on Python versions where CancelledError inherits Exception.

November 1, 2019

New features

•
Added compatibility with Python 3.8.

July 31, 2019

Bug fixes

  • Restored the ability to pass a socket with the sock parameter of serve().
  • Removed an incorrect assertion when a connection drops.

July 21, 2019

Bug fixes

•
Restored the ability to import WebSocketProtocolError from websockets.

July 7, 2019

Backwards-incompatible changes

websockets 7.0 is the last version supporting Python 3.4 and 3.5.

If you're passing a process_request argument to serve() or WebSocketServerProtocol, or if you're overriding process_request() in a subclass, define it with async def instead of def. Previously, both were supported.

For backwards compatibility, functions are still accepted, but mixing functions and coroutines won't work in some inheritance scenarios.

If you were setting max_queue=0 to make the queue of incoming messages unbounded, change it to max_queue=None.

of WebSocketCommonProtocol are deprecated."

Use local_address in servers and remote_address in clients instead of host and port.

to ProtocolError."

An alias provides backwards compatibility.

If you're using this low-level API, you must change your code.

New features

  • Added basic_auth_protocol_factory() to enforce HTTP Basic Auth on the server side.
  • connect() handles redirects from the server during the handshake.
  • connect() supports overriding host and port.
  • Added unix_connect() for connecting to Unix sockets.
  • Added support for asynchronous generators in send() to generate fragmented messages incrementally.
  • Enabled readline in the interactive client.
  • Added type hints (PEP 484).
  • Added a FAQ to the documentation.
  • Added documentation for extensions.
  • Documented how to optimize memory usage.

  • send(), ping(), and pong() support bytes-like types bytearray and memoryview in addition to bytes.
  • Added ConnectionClosedOK and ConnectionClosedError subclasses of ConnectionClosed to tell apart normal connection termination from errors.
  • Changed WebSocketServer.close() to perform a proper closing handshake instead of failing the connection.
  • Improved error messages when HTTP parsing fails.
  • Improved API documentation.

Bug fixes

  • Prevented spurious log messages about ConnectionClosed exceptions in keepalive ping task. If you were using ping_timeout=None as a workaround, you can remove it.
  • Avoided a crash when a extra_headers callable returns None.

November 1, 2018

Backwards-incompatible changes

websockets now sends Ping frames at regular intervals and closes the connection if it doesn't receive a matching Pong frame. See WebSocketCommonProtocol for details.

Previously, connections handlers were canceled. Now, connections are closed with close code 1001 (going away).

From the perspective of the connection handler, this is the same as if the remote endpoint was disconnecting. This removes the need to prepare for CancelledError in connection handlers.

You can restore the previous behavior by adding the following line at the beginning of connection handlers:

def handler(websocket, path):
    closed = asyncio.ensure_future(websocket.wait_closed())
    closed.add_done_callback(lambda task: task.cancel())
concurrently raises a RuntimeError."

Concurrent calls lead to non-deterministic behavior because there are no guarantees about which coroutine will receive which message.

and connect() is renamed to close_timeout ."

This prevents confusion with ping_timeout.

For backwards compatibility, timeout is still supported.

Include None in the list rather than '' to allow requests that don't contain an Origin header.

A ping — as in ping = await websocket.ping() — for which no pong was received yet used to be canceled when the connection is closed, so that await ping raised CancelledError.

Now await ping raises ConnectionClosed like other public APIs.

New features

  • Added process_request and select_subprotocol arguments to serve() and WebSocketServerProtocol to facilitate customization of process_request() and select_subprotocol().
  • Added support for sending fragmented messages.
  • Added the wait_closed() method to protocols.
  • Added an interactive client: python -m websockets <uri>.

  • Improved handling of multiple HTTP headers with the same name.
  • Improved error messages when a required HTTP header is missing.

Bug fixes

•
Fixed a data loss bug in recv(): canceling it at the wrong time could result in messages being dropped.

July 16, 2018

Backwards-incompatible changes

several APIs are updated to use it."
  • The request_headers argument of process_request() is now a Headers instead of an http.client.HTTPMessage.
  • The request_headers and response_headers attributes of WebSocketCommonProtocol are now Headers instead of http.client.HTTPMessage.
  • The raw_request_headers and raw_response_headers attributes of WebSocketCommonProtocol are removed. Use raw_items() instead.
  • Functions defined in the handshake module now receive Headers in argument instead of get_header or set_header functions. This affects libraries that rely on low-level APIs.
  • Functions defined in the http module now return HTTP headers as Headers instead of lists of (name, value) pairs.

Since Headers and http.client.HTTPMessage provide similar APIs, much of the code dealing with HTTP headers won't require changes.

New features

•
Added compatibility with Python 3.7.

May 24, 2018

Bug fixes

•
Fixed a regression in 5.0 that broke some invocations of serve() and connect().

May 22, 2018

Security fix

Version 4.0 was vulnerable to denial of service by memory exhaustion because it didn't enforce max_size when decompressing compressed messages (CVE-2018-1000518).

Backwards-incompatible changes

parse_uri and WebSocketURI."

If you're unpacking WebSocketURI into four variables, adjust your code to account for that fifth field.

New features

  • connect() performs HTTP Basic Auth when the URI contains credentials.
  • unix_serve() can be used as an asynchronous context manager on Python ≥ 3.5.1.
  • Added the closed property to protocols.
  • Added new examples in the documentation.

  • Iterating on incoming messages no longer raises an exception when the connection terminates with close code 1001 (going away).
  • A plain HTTP request now receives a 426 Upgrade Required response and doesn't log a stack trace.
  • If a ping() doesn't receive a pong, it's canceled when the connection is closed.
  • Reported the cause of ConnectionClosed exceptions.
  • Stopped logging stack traces when the TCP connection dies prematurely.
  • Prevented writing to a closing TCP connection during unclean shutdowns.
  • Made connection termination more robust to network congestion.
  • Prevented processing of incoming frames after failing the connection.
  • Updated documentation with new features from Python 3.6.
  • Improved several sections of the documentation.

Bug fixes

  • Prevented TypeError due to missing close code on connection close.
  • Fixed a race condition in the closing handshake that raised InvalidState.

November 2, 2017

Bug fixes

•
Fixed issues with the packaging of the 4.0 release.

November 2, 2017

Backwards-incompatible changes

websockets 3.4 is the last version supporting Python 3.3.

Compression is enabled by default.

In August 2017, Firefox and Chrome support the permessage-deflate extension, but not Safari and IE.

Compression should improve performance but it increases RAM and CPU use.

If you want to disable compression, add compression=None when calling serve() or connect().

Use protocol.state.name instead of protocol.state_name.

New features

  • WebSocketCommonProtocol instances can be used as asynchronous iterators on Python ≥ 3.6. They yield incoming messages.
  • Added unix_serve() for listening on Unix sockets.
  • Added the sockets attribute to the return value of serve().
  • Allowed extra_headers to override Server and User-Agent headers.

  • Reorganized and extended documentation.
  • Rewrote connection termination to increase robustness in edge cases.
  • Reduced verbosity of "Failing the WebSocket connection" logs.

Bug fixes

  • Aborted connections if they don't close within the configured timeout.
  • Stopped leaking pending tasks when cancel() is called on a connection while it's being closed.

August 20, 2017

Backwards-incompatible changes

by InvalidStatusCode."

This exception is raised when connect() receives an invalid response status code from the server.

New features

  • serve() can be used as an asynchronous context manager on Python ≥ 3.5.1.
  • Added support for customizing handling of incoming connections with process_request().
  • Made read and write buffer sizes configurable.

  • Renamed serve() and connect()'s klass argument to create_protocol to reflect that it can also be a callable. For backwards compatibility, klass is still supported.
  • Rewrote HTTP handling for simplicity and performance.
  • Added an optional C extension to speed up low-level operations.

Bug fixes

•
Providing a sock argument to connect() no longer crashes.

March 29, 2017

New features

•
Ensured compatibility with Python 3.6.

•
Reduced noise in logs caused by connection resets.

Bug fixes

•
Avoided crashing on concurrent writes on slow connections.

August 17, 2016

New features

•
Added timeout, max_size, and max_queue arguments to connect() and serve().

•
Made server shutdown more robust.

April 21, 2016

New features

•
Added flow control for incoming data.

Bug fixes

•
Avoided a warning when closing a connection before the opening handshake.

December 25, 2015

Backwards-incompatible changes

raises an exception when the connection is closed."

recv() used to return None when the connection was closed. This required checking the return value of every call:

message = await websocket.recv()
if message is None:
    return

Now it raises a ConnectionClosed exception instead. This is more Pythonic. The previous code can be simplified to:

message = await websocket.recv()

When implementing a server, there's no strong reason to handle such exceptions. Let them bubble up, terminate the handler coroutine, and the server will simply ignore them.

In order to avoid stranding projects built upon an earlier version, the previous behavior can be restored by passing legacy_recv=True to serve(), connect(), WebSocketServerProtocol, or WebSocketClientProtocol.

New features

  • connect() can be used as an asynchronous context manager on Python ≥ 3.5.1.
  • ping() and pong() support data passed as str in addition to bytes.
  • Made state_name attribute on protocols a public API.

  • Updated documentation with await and async syntax from Python 3.5.
  • Worked around an asyncio bug affecting connection termination under load.
  • Improved documentation.

November 18, 2015

New features

•
Added compatibility with Python 3.5.

•
Refreshed documentation.

August 18, 2015

New features

  • Added local_address and remote_address attributes on protocols.
  • Closed open connections with code 1001 when a server shuts down.

Bug fixes

•
Avoided TCP fragmentation of small frames.

July 28, 2015

New features

  • Provided access to handshake request and response HTTP headers.
  • Allowed customizing handshake request and response HTTP headers.
  • Added support for running on a non-default event loop.

  • Improved documentation.
  • Sent a 403 status code instead of 400 when request Origin isn't allowed.
  • Clarified that the closing handshake can be initiated by the client.
  • Set the close code and reason more consistently.
  • Strengthened connection termination.

Bug fixes

•
Canceling recv() no longer drops the next message.

January 31, 2015

New features

  • Added support for subprotocols.
  • Added loop argument to connect() and serve().

November 3, 2014

•
Improved compliance of close codes.

July 28, 2014

New features

•
Added support for limiting message size.

April 26, 2014

New features

  • Added host, port and secure attributes on protocols.
  • Added support for providing and checking Origin.

February 16, 2014

Backwards-incompatible changes

ping(), and pong() are now coroutines."

They used to be functions.

Instead of:

websocket.send(message)

you must write:

await websocket.send(message)

New features

•
Added flow control for outgoing data.

November 14, 2013

New features

•
Initial public release.

Thanks for taking the time to contribute to websockets!

This project and everyone participating in it is governed by the Code of Conduct. By participating, you are expected to uphold this code. Please report inappropriate behavior to aymeric DOT augustin AT fractalideas DOT com.

(If I'm the person with the inappropriate behavior, please accept my apologies. I know I can mess up. I can't expect you to tell me, but if you choose to do so, I'll do my best to handle criticism constructively. -- Aymeric)

Bug reports, patches and suggestions are welcome!

Please open an issue or send a pull request.

Feedback about the documentation is especially valuable, as the primary author feels more confident about writing code than writing docs :-)

If you're wondering why things are done in a certain way, the design document provides lots of details about the internals of websockets.

GitHub issues aren't a good medium for handling questions. There are better places to ask questions, for example Stack Overflow.

If you want to ask a question anyway, please make sure that:

  • it's a question about websockets and not about asyncio;
  • it isn't answered in the documentation;
  • it wasn't asked already.

A good question can be written as a suggestion to improve the documentation.

websockets appears to be quite popular for interfacing with Bitcoin or other cryptocurrency trackers. I'm strongly opposed to Bitcoin's carbon footprint.

I'm aware of efforts to build proof-of-stake models. I'll care once the total energy consumption of all cryptocurrencies drops to a non-bullshit level.

You already negated all of humanity's efforts to develop renewable energy. Please stop heating the planet where my children will have to live.

Since websockets is released under an open-source license, you can use it for any purpose you like. However, I won't spend any of my time to help you.

I will summarily close issues related to Bitcoin or cryptocurrency in any way.

Copyright (c) Aymeric Augustin and contributors

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

  • Redistributions of source code must retain the above copyright notice, this list of conditions and the following disclaimer.
  • Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.
  • Neither the name of the copyright holder nor the names of its contributors may be used to endorse or promote products derived from this software without specific prior written permission.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

[image]

Tidelift is working with the maintainers of websockets and thousands of other open source projects to deliver commercial support and maintenance for the open source dependencies you use to build your applications. Save time, reduce risk, and improve code health, while paying the maintainers of the exact dependencies you use.

The Tidelift Subscription is a managed open source subscription for application dependencies covering millions of open source projects across JavaScript, Python, Java, PHP, Ruby, .NET, and more.

Your subscription includes:

•
Security updates
•
Tidelift’s security response team coordinates patches for new breaking security vulnerabilities and alerts immediately through a private channel, so your software supply chain is always secure.
•
Licensing verification and indemnification
•
Tidelift verifies license information to enable easy policy enforcement and adds intellectual property indemnification to cover creators and users in case something goes wrong. You always have a 100% up-to-date bill of materials for your dependencies to share with your legal team, customers, or partners.
•
Maintenance and code improvement
•
Tidelift ensures the software you rely on keeps working as long as you need it to work. Your managed dependencies are actively maintained and we recruit additional maintainers where required.
•
Package selection and version guidance
•
We help you choose the best open source packages from the start—and then guide you through updates to stay on the best releases as new issues arise.
•
Roadmap input
•
Take a seat at the table with the creators behind the software you use. Tidelift’s participating maintainers earn more income as their software is used by more subscribers, so they’re interested in knowing what you need.
•
Tooling and cloud integration
•
Tidelift works with GitHub, GitLab, BitBucket, and more. We support every cloud platform (and other deployment targets, too).

The end result? All of the capabilities you expect from commercial-grade software, for the full breadth of open source you use. That means less time grappling with esoteric open source trivia, and more time building your own applications—and your business.

Aymeric Augustin

2013-2023, Aymeric Augustin and contributors

December 16, 2023 12.0