Feature/send large message (#14)

* introduce send fragment

* can pass a fin frame

* can send messages which are a perfect multiple of the chunk size

* set fin only for last fragment

* cleanup

* last fragment should be of type CONTINUATION

* Add simple send and receive programs

* speedups receiving + better way to wait for thing

* receive speedup by using linked list of chunks instead of large array

* document bug

* use chunks to receive data

* trailing spaces
This commit is contained in:
Benjamin Sergeant
2019-02-20 18:59:07 -08:00
committed by GitHub
parent dd4e29542c
commit 932bb732e0
72 changed files with 9117 additions and 260 deletions

View File

@ -16,6 +16,7 @@
#include <memory>
#include <mutex>
#include <atomic>
#include <list>
#include "IXWebSocketSendInfo.h"
#include "IXWebSocketPerMessageDeflate.h"
@ -23,8 +24,9 @@
#include "IXWebSocketHttpHeaders.h"
#include "IXCancellationRequest.h"
#include "IXWebSocketHandshake.h"
#include "IXProgressCallback.h"
namespace ix
namespace ix
{
class Socket;
@ -66,7 +68,8 @@ namespace ix
int timeoutSecs);
void poll();
WebSocketSendInfo sendBinary(const std::string& message);
WebSocketSendInfo sendBinary(const std::string& message,
const OnProgressCallback& onProgressCallback);
WebSocketSendInfo sendPing(const std::string& message);
void close();
ReadyStateValues getReadyState() const;
@ -76,7 +79,6 @@ namespace ix
private:
std::string _url;
std::string _origin;
struct wsheader_type {
unsigned header_size;
@ -96,13 +98,31 @@ namespace ix
uint8_t masking_key[4];
};
// Buffer for reading from our socket. That buffer is never resized.
std::vector<uint8_t> _readbuf;
// Contains all messages that were fetched in the last socket read.
// This could be a mix of control messages (Close, Ping, etc...) and
// data messages. That buffer
std::vector<uint8_t> _rxbuf;
// Contains all messages that are waiting to be sent
std::vector<uint8_t> _txbuf;
mutable std::mutex _txbufMutex;
std::vector<uint8_t> _receivedData;
// Hold fragments for multi-fragments messages in a list. We support receiving very large
// messages (tested messages up to 700M) and we cannot put them in a single
// buffer that is resized, as this operation can be slow when a buffer has its
// size increased 2 fold, while appending to a list has a fixed cost.
std::list<std::vector<uint8_t>> _chunks;
// Fragments are 32K long
static constexpr size_t kChunkSize = 1 << 15;
// Underlying TCP socket
std::shared_ptr<Socket> _socket;
// Hold the state of the connection (OPEN, CLOSED, etc...)
std::atomic<ReadyStateValues> _readyState;
OnCloseCallback _onCloseCallback;
@ -111,13 +131,14 @@ namespace ix
size_t _closeWireSize;
mutable std::mutex _closeDataMutex;
// Data used for Per Message Deflate compression (with zlib)
WebSocketPerMessageDeflate _perMessageDeflate;
WebSocketPerMessageDeflateOptions _perMessageDeflateOptions;
std::atomic<bool> _enablePerMessageDeflate;
// Used to cancel dns lookup + socket connect + http upgrade
std::atomic<bool> _requestInitCancellation;
// Optional Heartbeat
int _heartBeatPeriod;
static const int kDefaultHeartBeatPeriod;
@ -129,11 +150,18 @@ namespace ix
bool heartBeatPeriodExceeded();
void sendOnSocket();
WebSocketSendInfo sendData(wsheader_type::opcode_type type,
WebSocketSendInfo sendData(wsheader_type::opcode_type type,
const std::string& message,
bool compress);
bool compress,
const OnProgressCallback& onProgressCallback = nullptr);
void emitMessage(MessageKind messageKind,
void sendFragment(wsheader_type::opcode_type type,
bool fin,
std::string::const_iterator begin,
std::string::const_iterator end,
bool compress);
void emitMessage(MessageKind messageKind,
const std::string& message,
const wsheader_type& ws,
const OnMessageCallback& onMessageCallback);
@ -148,5 +176,7 @@ namespace ix
unsigned getRandomUnsigned();
void unmaskReceiveBuffer(const wsheader_type& ws);
std::string getMergedChunks() const;
};
}