feedc0de/boost_beast
1
0
Fork 0
forked from boostorg/beast
Code Pull Requests Activity

Add flat_stream to experimental:

Browse Source 
This adds a new stream wrapper class template designed to address
a performance shortcoming of boost::asio::ssl::stream.
This commit is contained in:
Vinnie Falco
2018-05-02 08:32:07 -07:00
parent 46e658a91b
commit 6108cf3eb7
17 changed files with 801 additions and 850 deletions
Download Patch File Download Diff File Expand all files Collapse all files
include/boost/beast/experimental/core/detail/flat_stream.hpp
+64
View File
@@ -0,0 +1,64 @@
//
// Copyright (c) 2016-2017 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)
//
// Official repository: https://github.com/boostorg/beast
//
#ifndef BOOST_BEAST_CORE_DETAIL_FLAT_STREAM_HPP
#define BOOST_BEAST_CORE_DETAIL_FLAT_STREAM_HPP
#include <boost/asio/buffer.hpp>
#include <cstdlib>
#include <iterator>
namespace boost {
namespace beast {
namespace detail {
class flat_stream_base
{
public:
// Largest buffer size we will flatten.
// 16KB is the upper limit on reasonably sized HTTP messages.
static std::size_t constexpr coalesce_limit = 16 * 1024;
// calculates the coalesce settings for a buffer sequence
template<class BufferSequence>
static
std::pair<std::size_t, bool>
coalesce(BufferSequence const& buffers, std::size_t limit)
{
std::pair<std::size_t, bool> result{0, false};
auto first = boost::asio::buffer_sequence_begin(buffers);
auto last = boost::asio::buffer_sequence_end(buffers);
if(first != last)
{
result.first = boost::asio::buffer_size(*first);
if(result.first < limit)
{
auto it = first;
auto prev = first;
while(++it != last)
{
auto const n =
boost::asio::buffer_size(*it);
if(result.first + n > limit)
break;
result.first += n;
prev = it;
}
result.second = prev != first;
}
}
return result;
}
};
} // detail
} // beast
} // boost
#endif
include/boost/beast/experimental/core/flat_stream.hpp
+353
View File
@@ -0,0 +1,353 @@
//
// Copyright (c) 2016-2017 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)
//
// Official repository: https://github.com/boostorg/beast
//
#ifndef BOOST_BEAST_CORE_FLAT_STREAM_HPP
#define BOOST_BEAST_CORE_FLAT_STREAM_HPP
#include <boost/beast/core/error.hpp>
#include <boost/beast/core/type_traits.hpp>
#include <boost/beast/experimental/core/detail/flat_stream.hpp>
#include <boost/asio/async_result.hpp>
#include <cstdlib>
#include <utility>
namespace boost {
namespace beast {
/** Stream wrapper to improve ssl::stream write performance.
This wrapper flattens writes for buffer sequences having length
greater than 1 and total size below a predefined amount, using
a dynamic memory allocation. It is primarily designed to overcome
a performance limitation of the current version of `boost::asio::ssl::stream`,
which does not use OpenSSL's scatter/gather interface for its
low-level read some and write some operations.
@par Example
To use the @ref flat_stream template with SSL streams, declare
a variable of the correct type. Parameters passed to the constructor
will be forwarded to the next layer's constructor:
@code
flat_stream<ssl::stream<ip::tcp::socket>> fs{ioc, ctx};
@endcode
Alternatively you can write
@code
ssl::stream<ip::tcp::socket> ss{ioc, ctx};
flat_stream<ssl::stream<ip::tcp::socket>&> fs{ss};
@endcode
The resulting stream may be passed to any stream algorithms which
operate on synchronous or asynchronous read or write streams,
examples include:
@li `boost::asio::read`, `boost::asio::async_read`
@li `boost::asio::write`, `boost::asio::async_write`
@li `boost::asio::read_until`, `boost::asio::async_read_until`
The stream may also be used as a template parameter in other
stream wrappers, such as for websocket:
@code
websocket::stream<flat_stream<ssl::stream<ip::tcp::socket>>> ws{ioc, ctx};
@endcode
@tparam NextLayer The type representing the next layer, to which
data will be read and written during operations. For synchronous
operations, the type must support the @b SyncStream concept. For
asynchronous operations, the type must support the @b AsyncStream
concept. This type will usually be some variation of
`boost::asio::ssl::stream`.
@par Concepts
@b AsyncStream
@b SyncStream
@see
@li https://github.com/boostorg/beast/issues/1108
@li https://github.com/boostorg/asio/issues/100
@li https://stackoverflow.com/questions/38198638/openssl-ssl-write-from-multiple-buffers-ssl-writev
@li https://stackoverflow.com/questions/50026167/performance-drop-on-port-from-beast-1-0-0-b66-to-boost-1-67-0-beast
*/
template<class NextLayer>
class flat_stream
#ifndef BOOST_BEAST_DOXYGEN
: private detail::flat_stream_base
#endif
{
// Largest buffer size we will flatten.
// 16KB is the upper limit on reasonably sized HTTP messages.
static std::size_t constexpr max_size = 16 * 1024;
template<class, class> class write_op;
NextLayer stream_;
public:
/// The type of the next layer.
using next_layer_type =
typename std::remove_reference<NextLayer>::type;
/// The type of the lowest layer.
using lowest_layer_type = boost::beast::get_lowest_layer<next_layer_type>;
/// The type of the executor associated with the object.
using executor_type = typename next_layer_type::executor_type;
flat_stream(flat_stream&&) = default;
flat_stream(flat_stream const&) = default;
flat_stream& operator=(flat_stream&&) = default;
flat_stream& operator=(flat_stream const&) = default;
/** Destructor
The treatment of pending operations will be the same as that
of the next layer.
*/
~flat_stream() = default;
/** Constructor
Arguments, if any, are forwarded to the next layer's constructor.
*/
template<class... Args>
explicit
flat_stream(Args&&... args);
//--------------------------------------------------------------------------
/** Get the executor associated with the object.
This function may be used to obtain the executor object that the
stream uses to dispatch handlers for asynchronous operations.
@return A copy of the executor that stream will use to dispatch handlers.
*/
executor_type
get_executor() noexcept
{
return stream_.get_executor();
}
/** Get a reference to the next layer
This function returns a reference to the next layer
in a stack of stream layers.
@return A reference to the next layer in the stack of
stream layers.
*/
next_layer_type&
next_layer()
{
return stream_;
}
/** Get a reference to the next layer
This function returns a reference to the next layer in a
stack of stream layers.
@return A reference to the next layer in the stack of
stream layers.
*/
next_layer_type const&
next_layer() const
{
return stream_;
}
/** Get a reference to the lowest layer
This function returns a reference to the lowest layer
in a stack of stream layers.
@return A reference to the lowest layer in the stack of
stream layers.
*/
lowest_layer_type&
lowest_layer()
{
return stream_.lowest_layer();
}
/** Get a reference to the lowest layer
This function returns a reference to the lowest layer
in a stack of stream layers.
@return A reference to the lowest layer in the stack of
stream layers. Ownership is not transferred to the caller.
*/
lowest_layer_type const&
lowest_layer() const
{
return stream_.lowest_layer();
}
//--------------------------------------------------------------------------
/** Read some data from the stream.
This function is used to read data from the stream. The function call will
block until one or more bytes of data has been read successfully, or until
an error occurs.
@param buffers The buffers into which the data will be read.
@returns The number of bytes read.
@throws boost::system::system_error Thrown on failure.
@note The `read_some` operation may not read all of the requested number of
bytes. Consider using the function `boost::asio::read` if you need to ensure
that the requested amount of data is read before the blocking operation
completes.
*/
template<class MutableBufferSequence>
std::size_t
read_some(MutableBufferSequence const& buffers);
/** Read some data from the stream.
This function is used to read data from the stream. The function call will
block until one or more bytes of data has been read successfully, or until
an error occurs.
@param buffers The buffers into which the data will be read.
@param ec Set to indicate what error occurred, if any.
@returns The number of bytes read.
@note The `read_some` operation may not read all of the requested number of
bytes. Consider using the function `boost::asio::read` if you need to ensure
that the requested amount of data is read before the blocking operation
completes.
*/
template<class MutableBufferSequence>
std::size_t
read_some(
MutableBufferSequence const& buffers,
error_code& ec);
/** Start an asynchronous read.
This function is used to asynchronously read one or more bytes of data from
the stream. The function call always returns immediately.
@param buffers The buffers into which the data will be read. Although the
buffers object may be copied as necessary, ownership of the underlying
buffers is retained by the caller, which must guarantee that they remain
valid until the handler is called.
@param handler The handler to be called when the read operation completes.
Copies will be made of the handler as required. The equivalent function
signature of the handler must be:
@code void handler(
const boost::system::error_code& error, // Result of operation.
std::size_t bytes_transferred // Number of bytes read.
); @endcode
@note The `read_some` operation may not read all of the requested number of
bytes. Consider using the function `boost::asio::async_read` if you need
to ensure that the requested amount of data is read before the asynchronous
operation completes.
*/
template<
class MutableBufferSequence,
class ReadHandler>
BOOST_ASIO_INITFN_RESULT_TYPE(
ReadHandler, void(error_code, std::size_t))
async_read_some(
MutableBufferSequence const& buffers,
ReadHandler&& handler);
/** Write some data to the stream.
This function is used to write data on the stream. The function call will
block until one or more bytes of data has been written successfully, or
until an error occurs.
@param buffers The data to be written.
@returns The number of bytes written.
@throws boost::system::system_error Thrown on failure.
@note The `write_some` operation may not transmit all of the data to the
peer. Consider using the function `boost::asio::write` if you need to
ensure that all data is written before the blocking operation completes.
*/
template<class ConstBufferSequence>
std::size_t
write_some(ConstBufferSequence const& buffers);
/** Write some data to the stream.
This function is used to write data on the stream. The function call will
block until one or more bytes of data has been written successfully, or
until an error occurs.
@param buffers The data to be written.
@param ec Set to indicate what error occurred, if any.
@returns The number of bytes written.
@note The `write_some` operation may not transmit all of the data to the
peer. Consider using the function `boost::asio::write` if you need to
ensure that all data is written before the blocking operation completes.
*/
template<class ConstBufferSequence>
std::size_t
write_some(
ConstBufferSequence const& buffers,
error_code& ec);
/** Start an asynchronous write.
This function is used to asynchronously write one or more bytes of data to
the stream. The function call always returns immediately.
@param buffers The data to be written to the stream. Although the buffers
object may be copied as necessary, ownership of the underlying buffers is
retained by the caller, which must guarantee that they remain valid until
the handler is called.
@param handler The handler to be called when the write operation completes.
Copies will be made of the handler as required. The equivalent function
signature of the handler must be:
@code void handler(
const boost::system::error_code& error, // Result of operation.
std::size_t bytes_transferred // Number of bytes written.
); @endcode
@note The `async_write_some` operation may not transmit all of the data to
the peer. Consider using the function `boost::asio::async_write` if you need
to ensure that all data is written before the asynchronous operation completes.
*/
template<
class ConstBufferSequence,
class WriteHandler>
BOOST_ASIO_INITFN_RESULT_TYPE(
WriteHandler, void(error_code, std::size_t))
async_write_some(
ConstBufferSequence const& buffers,
WriteHandler&& handler);
};
} // beast
} // boost
#include <boost/beast/experimental/core/impl/flat_stream.ipp>
#endif
include/boost/beast/experimental/core/impl/flat_stream.ipp
+304
View File
@@ -0,0 +1,304 @@
//
// Copyright (c) 2016-2017 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)
//
// Official repository: https://github.com/boostorg/beast
//
#ifndef BOOST_BEAST_CORE_IMPL_FLAT_STREAM_IPP
#define BOOST_BEAST_CORE_IMPL_FLAT_STREAM_IPP
#include <boost/beast/core/buffers_prefix.hpp>
#include <boost/beast/websocket/teardown.hpp>
#include <boost/asio/associated_allocator.hpp>
#include <boost/asio/associated_executor.hpp>
#include <boost/asio/buffer.hpp>
#include <boost/asio/coroutine.hpp>
#include <boost/asio/handler_continuation_hook.hpp>
#include <boost/asio/handler_invoke_hook.hpp>
namespace boost {
namespace beast {
template<class NextLayer>
template<class ConstBufferSequence, class Handler>
class flat_stream<NextLayer>::write_op
: public boost::asio::coroutine
{
using alloc_type = typename
boost::asio::associated_allocator_t<Handler>::template
rebind<char>::other;
struct deleter
{
alloc_type alloc;
std::size_t size = 0;
explicit
deleter(alloc_type const& alloc_)
: alloc(alloc_)
{
}
void
operator()(char* p)
{
alloc.deallocate(p, size);
}
};
flat_stream<NextLayer>& s_;
ConstBufferSequence b_;
std::unique_ptr<char, deleter> p_;
Handler h_;
public:
template<class DeducedHandler>
write_op(
flat_stream<NextLayer>& s,
ConstBufferSequence const& b,
DeducedHandler&& h)
: s_(s)
, b_(b)
, p_(nullptr, deleter{
(boost::asio::get_associated_allocator)(h)})
, h_(std::forward<DeducedHandler>(h))
{
}
using allocator_type =
boost::asio::associated_allocator_t<Handler>;
allocator_type
get_allocator() const noexcept
{
return (boost::asio::get_associated_allocator)(h_);
}
using executor_type = boost::asio::associated_executor_t<
Handler, decltype(std::declval<NextLayer&>().get_executor())>;
executor_type
get_executor() const noexcept
{
return (boost::asio::get_associated_executor)(
h_, s_.get_executor());
}
void
operator()(
boost::system::error_code ec,
std::size_t bytes_transferred);
friend
bool asio_handler_is_continuation(write_op* op)
{
using boost::asio::asio_handler_is_continuation;
return asio_handler_is_continuation(
std::addressof(op->h_));
}
template<class Function>
friend
void asio_handler_invoke(Function&& f, write_op* op)
{
using boost::asio::asio_handler_invoke;
asio_handler_invoke(f, std::addressof(op->h_));
}
};
template<class NextLayer>
template<class ConstBufferSequence, class Handler>
void
flat_stream<NextLayer>::
write_op<ConstBufferSequence, Handler>::
operator()(
error_code ec,
std::size_t bytes_transferred)
{
BOOST_ASIO_CORO_REENTER(*this)
{
BOOST_ASIO_CORO_YIELD
{
auto const result = coalesce(b_, coalesce_limit);
if(result.second)
{
p_.get_deleter().size = result.first;
p_.reset(p_.get_deleter().alloc.allocate(
p_.get_deleter().size));
boost::asio::buffer_copy(
boost::asio::buffer(
p_.get(), p_.get_deleter().size),
b_, result.first);
s_.stream_.async_write_some(
boost::asio::buffer(
p_.get(), p_.get_deleter().size),
std::move(*this));
}
else
{
s_.stream_.async_write_some(
boost::beast::buffers_prefix(result.first, b_),
std::move(*this));
}
}
p_.reset();
h_(ec, bytes_transferred);
}
}
//------------------------------------------------------------------------------
template<class NextLayer>
template<class... Args>
flat_stream<NextLayer>::
flat_stream(Args&&... args)
: stream_(std::forward<Args>(args)...)
{
}
template<class NextLayer>
template<class MutableBufferSequence>
std::size_t
flat_stream<NextLayer>::
read_some(MutableBufferSequence const& buffers)
{
static_assert(boost::beast::is_sync_read_stream<next_layer_type>::value,
"SyncReadStream requirements not met");
static_assert(boost::asio::is_mutable_buffer_sequence<
MutableBufferSequence>::value,
"MutableBufferSequence requirements not met");
error_code ec;
auto n = read_some(buffers, ec);
if(ec)
BOOST_THROW_EXCEPTION(boost::system::system_error{ec});
return n;
}
template<class NextLayer>
template<class MutableBufferSequence>
std::size_t
flat_stream<NextLayer>::
read_some(MutableBufferSequence const& buffers, error_code& ec)
{
return stream_.read_some(buffers, ec);
}
template<class NextLayer>
template<
class MutableBufferSequence,
class ReadHandler>
BOOST_ASIO_INITFN_RESULT_TYPE(
ReadHandler, void(error_code, std::size_t))
flat_stream<NextLayer>::
async_read_some(
MutableBufferSequence const& buffers,
ReadHandler&& handler)
{
static_assert(boost::beast::is_async_read_stream<next_layer_type>::value,
"AsyncReadStream requirements not met");
static_assert(boost::asio::is_mutable_buffer_sequence<
MutableBufferSequence >::value,
"MutableBufferSequence requirements not met");
return stream_.async_read_some(
buffers, std::forward<ReadHandler>(handler));
}
template<class NextLayer>
template<class ConstBufferSequence>
std::size_t
flat_stream<NextLayer>::
write_some(ConstBufferSequence const& buffers)
{
static_assert(boost::beast::is_sync_write_stream<next_layer_type>::value,
"SyncWriteStream requirements not met");
static_assert(boost::asio::is_const_buffer_sequence<
ConstBufferSequence>::value,
"ConstBufferSequence requirements not met");
auto const result = coalesce(buffers, coalesce_limit);
if(result.second)
{
std::unique_ptr<char[]> p{new char[result.first]};
auto const b = boost::asio::buffer(p.get(), result.first);
boost::asio::buffer_copy(b, buffers);
return stream_.write_some(b);
}
return stream_.write_some(
boost::beast::buffers_prefix(result.first, buffers));
}
template<class NextLayer>
template<class ConstBufferSequence>
std::size_t
flat_stream<NextLayer>::
write_some(ConstBufferSequence const& buffers, error_code& ec)
{
static_assert(boost::beast::is_sync_write_stream<next_layer_type>::value,
"SyncWriteStream requirements not met");
static_assert(boost::asio::is_const_buffer_sequence<
ConstBufferSequence>::value,
"ConstBufferSequence requirements not met");
auto const result = coalesce(buffers, coalesce_limit);
if(result.second)
{
std::unique_ptr<char[]> p{new char[result.first]};
auto const b = boost::asio::buffer(p.get(), result.first);
boost::asio::buffer_copy(b, buffers);
return stream_.write_some(b, ec);
}
return stream_.write_some(
boost::beast::buffers_prefix(result.first, buffers), ec);
}
template<class NextLayer>
template<
class ConstBufferSequence,
class WriteHandler>
BOOST_ASIO_INITFN_RESULT_TYPE(
WriteHandler, void(error_code, std::size_t))
flat_stream<NextLayer>::
async_write_some(
ConstBufferSequence const& buffers,
WriteHandler&& handler)
{
static_assert(boost::beast::is_async_write_stream<next_layer_type>::value,
"AsyncWriteStream requirements not met");
static_assert(boost::asio::is_const_buffer_sequence<
ConstBufferSequence>::value,
"ConstBufferSequence requirements not met");
BOOST_BEAST_HANDLER_INIT(
WriteHandler, void(error_code, std::size_t));
write_op<ConstBufferSequence, BOOST_ASIO_HANDLER_TYPE(
WriteHandler, void(error_code, std::size_t))>{
*this, buffers, std::move(init.completion_handler)}({}, 0);
return init.result.get();
}
template<class NextLayer>
void
teardown(
boost::beast::websocket::role_type role,
flat_stream<NextLayer>& s,
error_code& ec)
{
using boost::beast::websocket::teardown;
teardown(role, s.next_layer(), ec);
}
template<class NextLayer, class TeardownHandler>
void
async_teardown(
boost::beast::websocket::role_type role,
flat_stream<NextLayer>& s,
TeardownHandler&& handler)
{
using boost::beast::websocket::async_teardown;
async_teardown(role, s.next_layer(), std::move(handler));
}
} // beast
} // boost
#endif