rippled/doc/websocket.qbk

[/
    Copyright (c) 2013-2016 Vinnie Falco (vinnie dot falco at gmail dot com)

    Distributed under the Boost Software License, Version 1.0. (See accompanying
    file LICENSE_1_0.txt or copy at http://www.boost.org/LICENSE_1_0.txt)
]

[section:websocket WebSocket]

The WebSocket Protocol enables two-way communication between a client
running untrusted code in a controlled environment to a remote host that has
opted-in to communications from that code. The protocol consists of an opening
handshake followed by basic message framing, layered over TCP.  The goal of
this technology is to provide a mechanism for browser-based applications that
need two-way communication with servers that does not rely on opening multiple
HTTP connections.

Beast.WebSocket provides developers with a robust WebSocket implementation
built on Boost.Asio with a consistent asynchronous model using a modern
C++ approach.

The WebSocket protocol is described fully in
[@https://tools.ietf.org/html/rfc6455 rfc6455]



[section:motivation Motivation]

Today's web applications increasingly rely on alternatives to standard HTTP
to achieve performance and/or responsiveness. While WebSocket implementations
are widely available in common web development languages such as Javascript,
good implementations in C++ are scarce. A survey of existing C++ WebSocket
solutions reveals interfaces which lack symmetry, impose performance penalties,
and needlessly restrict implementation strategies.

Beast.WebSocket is built on Boost.Asio, a robust cross platform networking
framework that is part of Boost and also offered as a standalone library.
A proposal to add networking functionality to the C++ standard library,
based on Boost.Asio, is under consideration by the standards committee.
Since the final approved networking interface for the C++ standard library
will likely closely resemble the current interface of Boost.Asio, it is
logical for Beast.WebSocket to use Boost.Asio as its network transport.

Beast.WebSocket takes advantage of Boost.Asio's extensible asynchronous
model, handler allocation, and handler invocation hooks. Calls to
Beast.WebSocket asynchronous initiation functions allow callers the choice
of using a completion handler, stackful or stackless coroutines, futures,
or user defined customizations (for example, Boost.Fiber). The
implementation uses handler invocation hooks
([@http://www.boost.org/doc/libs/1_61_0/doc/html/boost_asio/reference/asio_handler_invoke.html `asio_handler_invoke`]),
providing execution guarantees on composed operations in a manner
identical to Boost.Asio. The implementation also uses handler allocation hooks
([@http://www.boost.org/doc/libs/1_61_0/doc/html/boost_asio/reference/asio_handler_allocate.html `asio_handler_allocate`])
when allocating memory internally for composed operations.

There is no need for inheritance or virtual members in a
[link beast.ref.websocket__stream `beast::websocket::stream`].
All operations are templated and transparent to the compiler, allowing for
maximum inlining and optimization.

[note The documentation which follows assumes familiarity with
both Boost.Asio and the WebSocket protocol specification described in
[@https://tools.ietf.org/html/rfc6455 rfc6455] ]

[endsect]



[section:creating Creating the socket]

The interface to Beast's WebSocket implementation is a single template
class [link beast.ref.websocket__stream `beast::websocket::stream`] which
wraps a "next layer" object. The next layer object must meet the requirements
of [link beast.types.streams.SyncStream [*`SyncReadStream`]] if synchronous
operations are performed, or
[link beast.types.streams.AsyncStream [*`AsyncStream`]] if asynchronous
operations are performed, or both. Arguments supplied during construction are
passed to next layer's constructor. Here we declare two websockets which have
ownership of the next layer:
```
boost::asio::io_service ios;
beast::websocket::stream<boost::asio::ip::tcp::socket> ws(ios);

boost::asio::ssl::context ctx(boost::asio::ssl::context::sslv23);
beast::websocket::stream<
    boost::asio::ssl::stream<boost::asio::ip::tcp::socket>> wss(ios, ctx);
```

For servers that can handshake in multiple protocols, it may be desired
to wrap an object that already exists. This socket can be moved in:
```
    boost::asio::ip::tcp::socket&& sock;
    ...
    beast::websocket::stream<
        boost::asio::ip::tcp::socket> ws(std::move(sock));
```

Or, the wrapper can be constructed with a non-owning reference. In
this case, the caller is responsible for managing the lifetime of the
underlying socket being wrapped:
```
    boost::asio::ip::tcp::socket sock;
    ...
    beast::websocket::stream<boost::asio::ip::tcp::socket&> ws(sock);
```

The layer being wrapped can be accessed through the websocket's "next layer",
permitting callers to interact directly with its interface.
```
    boost::asio::ssl::context ctx(boost::asio::ssl::context::sslv23);
    beast::websocket::stream<
        boost::asio::ssl::stream<boost::asio::ip::tcp::socket>> ws(ios, ctx);
    ...
    ws.next_layer().shutdown(); // ssl::stream shutdown
```

[important Initiating read and write operations on the next layer while
websocket operations are being performed can break invariants, and
result in undefined behavior. ]

[endsect]



[section:connecting Making connections]

Connections are established by using the interfaces which already exist
for the next layer. For example, making an outgoing connection:
```
    std::string const host = "mywebapp.com";
    boost::asio::io_service ios;
    boost::asio::ip::tcp::resolver r(ios);
    beast::websocket::stream<boost::asio::ip::tcp::socket> ws(ios);
    boost::asio::connect(ws.next_layer(),
        r.resolve(boost::asio::ip::tcp::resolver::query{host, "ws"}));
```

Accepting an incoming connection:
```
void do_accept(boost::asio::ip::tcp::acceptor& acceptor)
{
    beast::websocket::stream<boost::asio::ip::tcp::socket> ws(acceptor.get_io_service());
    acceptor.accept(ws.next_layer());
}
```

[note Examples use synchronous interfaces for clarity of exposition. ]

[endsect]



[section:handshaking Handshaking]

A WebSocket session begins when one side sends the HTTP Upgrade request
for websocket, and the other side sends an appropriate HTTP response
indicating that the request was accepted and that the connection has
been upgraded. The HTTP Upgrade request must include the Host HTTP field,
and the URI of the resource to request. `handshake` is used to send the
request with the required host and resource strings.
```
    beast::websocket::stream<boost::asio::ip::tcp::socket> ws(ios);
    ...
    ws.set_option(beast::websocket::keep_alive(true));
    ws.handshake("ws.example.com:80", "/cgi-bin/bitcoin-prices");
```

The [link beast.ref.websocket__stream `beast::websocket::stream`] automatically
handles receiving and processing the HTTP response to the handshake request.
The call to handshake is successful if a HTTP response is received with the
101 "Switching Protocols" status code. On failure, an error is returned or an
exception is thrown. Depending on the keep alive setting, the socket may remain
open for a subsequent handshake attempt

Performing a handshake for an incoming websocket upgrade request operates
similarly. If the handshake fails, an error is returned or exception thrown:
```
    beast::websocket::stream<boost::asio::ip::tcp::socket> ws(ios);
    ...
    ws.accept();
```

Servers that can handshake in multiple protocols may have already read data
on the connection, or might have already received an entire HTTP request
containing the upgrade request. Overloads of `accept` allow callers to
pass in this additional buffered handshake data.
```
void do_accept(boost::asio::ip::tcp::socket& sock)
{
    boost::asio::streambuf sb;
    boost::asio::read_until(sock, sb, "\r\n\r\n");
    ...
    beast::websocket::stream<boost::asio::ip::tcp::socket&> ws(sock);
    ws.accept(sb.data());
    ...
}
```

Alternatively, the caller can pass an entire HTTP request if it was
obtained elsewhere:
```
void do_accept(boost::asio::ip::tcp::socket& sock)
{
    boost::asio::streambuf sb;
    beast::http::request<http::empty_body> request;
    beast::http::read(sock, request);
    if(beast::http::is_upgrade(request))
    {
        websocket::stream<ip::tcp::socket&> ws(sock);
        ws.accept(request);
        ...
    }
}
```

[endsect]



[section:messages Messages]

After the WebSocket handshake is accomplished, callers may send and receive
messages using the message oriented interface. This interface requires that
all of the buffers representing the message are known ahead of time:
```
void echo(beast::websocket::stream<boost::asio::ip::tcp::socket>& ws)
{
    beast::streambuf sb;
    beast::websocket::opcode::value op;
    ws.read(sb);

    ws.set_option(beast::websocket::message_type(op));
    ws.write(sb.data());
    sb.consume(sb.size());
}
```

[important Calls to [link beast.ref.websocket__stream.set_option `set_option`]
must be made from the same implicit or explicit strand as that used to perform
other operations. ]

[endsect]



[section:frames Frames]

Some use-cases make it impractical or impossible to buffer the entire
message ahead of time:

* Streaming multimedia to an endpoint.
* Sending a message that does not fit in memory at once.
* Providing incremental results as they become available.

For these cases, the frame oriented interface may be used. This
example reads and echoes a complete message using this interface:
```
void echo(beast::websocket::stream<boost::asio::ip::tcp::socket>& ws)
{
    beast::streambuf sb;
    beast::websocket::frame_info fi;
    for(;;)
    {
        ws.read_frame(fi, sb);
        if(fi.fin)
            break;
    }
    ws.set_option(beast::websocket::message_type(fi.op));
    beast::consuming_buffers<
        beast::streambuf::const_buffers_type> cb(sb.data());
    for(;;)
    {
        using boost::asio::buffer_size;
        std::size_t size = std::min(buffer_size(cb));
        if(size > 512)
        {
            ws.write_frame(false, beast::prepare_buffers(512, cb));
            cb.consume(512);
        }
        else
        {
            ws.write_frame(true, cb);
            break;
        }
    }
}
```

[endsect]



[section:controlframes Control frames]

During read operations, the implementation automatically reads and processes
WebSocket control frames such as ping, pong, and close. Pings are replied
to as soon as possible, pongs are delivered to the pong callback. The receipt
of a close frame initiates the WebSocket close procedure, eventually resulting
in the error code [link beast.ref.websocket__error `error::closed`] being
delivered to the caller in a subsequent read operation, assuming no other error
takes place.

To ensure timely delivery of control frames, large messages are broken up
into smaller sized frames. The implementation chooses the size and number
of the frames making up the message. The automatic fragment size option
gives callers control over the size of these frames:
```
    ...
    ws.set_option(beast::websocket::auto_fragment_size(8192));
```

The WebSocket protocol defines a procedure and control message for initiating
a close of the session. Handling of close initiated by the remote end of the
connection is performed automatically. To manually initiate a close, use
[link beast.ref.websocket__stream.close `close`]:
```
    ws.close();
```

[note To receive the [link beast.ref.websocket__error `error::closed`]
error, a read operation is required. ]

[endsect]



[section:pongs Pong messages]

To receive pong control frames, callers may register a "pong callback" using
[link beast.ref.websocket__stream.set_option `set_option`]:

the following signature:
```
    void on_pong(ping_data const& payload);
    ...
    ws.set_option(pong_callback{&on_pong});
```

When a pong callback is registered, any pongs received through either
synchronous read functions or asynchronous read functions will invoke the
pong callback, passing the payload in the pong message as the argument.

Unlike regular completion handlers used in calls to asynchronous initiation
functions, the pong callback only needs to be set once. The callback is not
reset when a pong is received. The same callback is used for both synchronous
and asynchronous reads. The pong callback is passive; in order to receive
pongs, a synchronous or asynchronous stream read function must be active.

[note When an asynchronous read function receives a pong, the the pong callback
is invoked in the same manner as that used to invoke the final completion
handler of the corresponding read function.]

[endsect]



[section:buffers Buffers]

Because calls to read data may return a variable amount of bytes, the
interface to calls that read data require an object that meets the requirements
of [link beast.types.DynamicBuffer [*`DynamicBuffer`]]. This concept is modeled on
[@http://www.boost.org/doc/libs/1_61_0/doc/html/boost_asio/reference/basic_streambuf.html `boost::asio::basic_streambuf`].

The implementation does not perform queueing or buffering of messages. If
desired, these features should be provided by callers. The impact of this
design is that library users are in full control of the allocation strategy
used to store data and the back-pressure applied on the read and write side
of the underlying TCP/IP connection.

[endsect]



[section:async Asynchronous interface]

Asynchronous versions are available for all functions:
```
websocket::opcode op;
ws.async_read(op, sb,
    [](boost::system::error_code const& ec)
    {
        ...
    });
```

Calls to asynchronous initiation functions support the extensible asynchronous
model developed by the Boost.Asio author, allowing for traditional completion
handlers, stackful or stackless coroutines, and even futures:
```
void echo(websocket::stream<ip::tcp::socket>& ws,
    boost::asio::yield_context yield)
{
    ws.async_read(sb, yield);
    std::future<websocket::error_code> fut =
        ws.async_write, sb.data(), boost::use_future);
    ...
}
```

[endsect]



[section:io_service The io_service]

The creation and operation of the
[@http://www.boost.org/doc/libs/1_61_0/doc/html/boost_asio/reference/io_service.html `boost::asio::io_service`]
associated with the underlying stream is left to the callers, permitting any
implementation strategy including one that does not require threads for
environments where threads are unavailable. Beast.WebSocket itself does not
use or require threads.

[endsect]



[section:safety Thread Safety]

Like a regular asio socket, a [link beast.ref.websocket__stream `stream`] is
not thread safe. Callers are responsible for synchronizing operations on the
socket using an implicit or explicit strand, as per the Asio documentation.
The asynchronous interface supports one active read and one active write
simultaneously. Undefined behavior results if two or more reads or two or
more writes are attempted concurrently. Caller initiated WebSocket ping, pong,
and close operations each count as an active write.

The implementation uses composed asynchronous operations internally; a high
level read can cause both reads and writes to take place on the underlying
stream. This behavior is transparent to callers.

[endsect]



[endsect]

[include quickref.xml]