流媒体学习之路(mediasoup)——信令传输（3）
文章目录
一、Node.js部分
mediasoup——c++部分的信令交互主要通过Socket管道来进行的。Node.js部分与c++部分通过本机网络进行通话，实现交互。
前面两篇提到，Node.js部分worker建立后需要建立channel来进行通信，我们再来回顾一下channel.js部分通信的代码。下面是Worker的构造函数：
constructor({ logLevel, logTags, rtcMinPort, rtcMaxPort, dtlsCertificateFile, dtlsPrivateKeyFile, appData }) {
super();
// Closed flag.
this._closed = false;
// Routers set.
this._routers = new Set();
// Observer instance.
this._observer = new EnhancedEventEmitter_1.EnhancedEventEmitter();
logger.debug('constructor()');
let spawnBin = workerBin;
let spawnArgs = [];
if (process.env.MEDIASOUP_USE_VALGRIND === 'true') {
spawnBin = process.env.MEDIASOUP_VALGRIND_BIN || 'valgrind';
if (process.env.MEDIASOUP_VALGRIND_OPTIONS)
spawnArgs = spawnArgs.concat(process.env.MEDIASOUP_VALGRIND_OPTIONS.split(/\s+/));
spawnArgs.push(workerBin);
}
if (typeof logLevel === 'string' && logLevel)
spawnArgs.push(`--logLevel=${logLevel}`);
for (const logTag of (Array.isArray(logTags) ? logTags : [])) {
if (typeof logTag === 'string' && logTag)
spawnArgs.push(`--logTag=${logTag}`);
}
if (typeof rtcMinPort === 'number' && !Number.isNaN(rtcMinPort))
spawnArgs.push(`--rtcMinPort=${rtcMinPort}`);
if (typeof rtcMaxPort === 'number' && !Number.isNaN(rtcMaxPort))
spawnArgs.push(`--rtcMaxPort=${rtcMaxPort}`);
if (typeof dtlsCertificateFile === 'string' && dtlsCertificateFile)
spawnArgs.push(`--dtlsCertificateFile=${dtlsCertificateFile}`);
if (typeof dtlsPrivateKeyFile === 'string' && dtlsPrivateKeyFile)
spawnArgs.push(`--dtlsPrivateKeyFile=${dtlsPrivateKeyFile}`);
logger.debug('spawning worker process: %s %s', spawnBin, spawnArgs.join(' '));
this._child = child_process_1.spawn(
// command
spawnBin,
// args
spawnArgs,
// options
{
env: {
MEDIASOUP_VERSION: '3.6.13'
},
detached: false,
// fd 0 (stdin) : Just ignore it.
// fd 1 (stdout) : Pipe it for 3rd libraries that log their own stuff.
// fd 2 (stderr) : Same as stdout.
// fd 3 (channel) : Producer Channel fd.
// fd 4 (channel) : Consumer Channel fd.
// fd 5 (channel) : Producer PayloadChannel fd.
// fd 6 (channel) : Consumer PayloadChannel fd.
stdio: ['ignore', 'pipe', 'pipe', 'pipe', 'pipe', 'pipe', 'pipe'],
windowsHide: true
});
this._pid = this._child.pid;
this._channel = new Channel_1.Channel({
producerSocket: this._child.stdio[3],
consumerSocket: this._child.stdio[4],
pid: this._pid
});
this._payloadChannel = new PayloadChannel_1.PayloadChannel({
// NOTE: TypeScript does not like more than 5 fds.
// @ts-ignore
producerSocket: this._child.stdio[5],
// @ts-ignore
consumerSocket: this._child.stdio[6]
});
this._appData = appData;
let spawnDone = false;
// Listen for 'running' notification.
this._channel.once(String(this._pid), (event) => {
if (!spawnDone && event === 'running') {
spawnDone = true;
logger.debug('worker process running [pid:%s]', this._pid);
this.emit('@success');
}
});
this._child.on('exit', (code, signal) => {
this._child = undefined;
this.close();
if (!spawnDone) {
spawnDone = true;
if (code === 42) {
logger.error('worker process failed due to wrong settings [pid:%s]', this._pid);
this.emit('@failure', new TypeError('wrong settings'));
}
else {
logger.error('worker process failed unexpectedly [pid:%s, code:%s, signal:%s]', this._pid, code, signal);
this.emit('@failure', new Error(`[pid:${this._pid}, code:${code}, signal:${signal}]`));
}
}
else {
logger.error('worker process died unexpectedly [pid:%s, code:%s, signal:%s]', this._pid, code, signal);
this.safeEmit('died', new Error(`[pid:${this._pid}, code:${code}, signal:${signal}]`));
}
});
this._child.on('error', (error) => {
this._child = undefined;
this.close();
if (!spawnDone) {
spawnDone = true;
logger.error('worker process failed [pid:%s]: %s', this._pid, error.message);
this.emit('@failure', error);
}
else {
logger.error('worker process error [pid:%s]: %s', this._pid, error.message);
this.safeEmit('died', error);
}
});
// Be ready for 3rd party worker libraries logging to stdout.
this._child.stdout.on('data', (buffer) => {
for (const line of buffer.toString('utf8').split('\n')) {
if (line)
workerLogger.debug(`(stdout) ${line}`);
}
});
// In case of a worker bug, mediasoup will log to stderr.
this._child.stderr.on('data', (buffer) => {
for (const line of buffer.toString('utf8').split('\n')) {
if (line)
workerLogger.error(`(stderr) ${line}`);
}
});
}
可见创建子进程（worker）后，创建了两个频道channel、payloadchannel。这两个对应cpp代码中的两个频道。创建channel时传入的三个参数——producerSocket, consumerSocket, pid 分别代表：子进程的生产者Socket、子进程消费者Socket以及子进程（worker-cpp）的进程id。
constructor({ producerSocket, consumerSocket, pid }) {
super();
// Closed flag.
this._closed = false;
// Next id for messages sent to the worker process.
this._nextId = 0;
// Map of pending sent requests.
this._sents = new Map();
logger.debug('constructor()');
this._producerSocket = producerSocket;
this._consumerSocket = consumerSocket;
// Read Channel responses/notifications from the worker.
this._consumerSocket.on('data', (buffer) => {
if (!this._recvBuffer) {
this._recvBuffer = buffer;
}
else {
this._recvBuffer = Buffer.concat([this._recvBuffer, buffer], this._recvBuffer.length + buffer.length);
}
if (this._recvBuffer.length > NS_PAYLOAD_MAX_LEN) {
logger.error('receiving buffer is full, discarding all data into it');
// Reset the buffer and exit.
this._recvBuffer = undefined;
return;
}
while (true) // eslint-disable-line no-constant-condition
{
let nsPayload;
try {
nsPayload = netstring.nsPayload(this._recvBuffer);
}
catch (error) {
logger.error('invalid netstring data received from the worker process: %s', String(error));
// Reset the buffer and exit.
this._recvBuffer = undefined;
return;
}
// Incomplete netstring message.
if (nsPayload === -1)
return;
try {
// We can receive JSON messages (Channel messages) or log strings.
switch (nsPayload[0]) {
// 123 = '{' (a Channel JSON messsage).
case 123:
this._processMessage(JSON.parse(nsPayload.toString('utf8')));
break;
// 68 = 'D' (a debug log).
case 68:
logger.debug(`[pid:${pid}] ${nsPayload.toString('utf8', 1)}`);
break;
// 87 = 'W' (a warn log).
case 87:
logger.warn(`[pid:${pid}] ${nsPayload.toString('utf8', 1)}`);
break;
// 69 = 'E' (an error log).
case 69:
logger.error(`[pid:${pid} ${nsPayload.toString('utf8', 1)}`);
break;
// 88 = 'X' (a dump log).
case 88:
// eslint-disable-next-line no-console
console.log(nsPayload.toString('utf8', 1));
break;
default:
// eslint-disable-next-line no-console
console.warn(`worker[pid:${pid}] unexpected data: %s`, nsPayload.toString('utf8', 1));
}
}
catch (error) {
logger.error('received invalid message from the worker process: %s', String(error));
}
// Remove the read payload from the buffer.
this._recvBuffer =
this._recvBuffer.slice(netstring.nsLength(this._recvBuffer));
if (!this._recvBuffer.length) {
this._recvBuffer = undefined;
return;
}
}
});
this._consumerSocket.on('end', () => (logger.debug('Consumer Channel ended by the worker process')));
this._consumerSocket.on('error', (error) => (logger.error('Consumer Channel error: %s', String(error))));
this._producerSocket.on('end', () => (logger.debug('Producer Channel ended by the worker process')));
this._producerSocket.on('error', (error) => (logger.error('Producer Channel error: %s', String(error))));
}
在channel中执行了一个死循环，等待接收buffer中传入数据后想worker进行通信。
二、C++部分
为了方便理解，我将cpp代码信令部分按语义进行了分层。分层图大致如下：
2.1 UnixStreamSocket创建
这里有一个很容易让人头晕的类名，就是UnixStreamSocket。这个类在命名空间Channel中以及全局空间中有同名的两个类。
a. 全局空间的类是生产者/消费者类的真正基类，它将会保存Socket的fd、最大buf大小、角色类型（CONSUMER 或者 PRODUCER）。
b. Channel内部的类则为主函数中调用的类，该类负责初始化生产者、消费者，同时该类也继承了生产者消费者类内的监听类。这样就可以在外层实现内部类函数的调用。
下面列一下伪代码：
class UnixStreamSocket
{
public:
UnixStreamSocket(int fd, size_t bufferSize, UnixStreamSocket::Role role);
};
namespace Channel {
class ConsumerSocket : public ::UnixStreamSocket{
public:
class Listener{
};
ConsumerSocket(int fd, size_t bufferSize, Listener* listener);
};
class UnixStreamSocket : public ConsumerSocket::Listener {
public:
explicit UnixStreamSocket(int consumerFd, int producerFd);
};
};
2.2 UnixStreamSocket
在C++代码的main函数中，最初就构造了两个channel。其中，Channel::UnixStreamSocket的构造中首先创建了一个名为uvHandle的uv_pipe_t私有对象。该对象将会开启libUV，这里做了一下libUV的基础简介：https://blog.csdn.net/qw225967/article/details/119319970
libUV中，初始化了uv_pipe，随后调用uv_pipe_open然后开始uv_read_start。这样libUV就进入了loop。
打开之后进入uv_read_start，这个函数是用来启动读的操作，它同样也有三个参数：
a. uvHandle，里面存放的是这个对象本身
b. onAlloc，当buffer不足时回调这个函数，便能重新创建一个buffer
c. onRead，接收pipe的另一端发送的数据
此处引用熠熠微光作者的文章：https://blog.csdn.net/Frederick_Fung/article/details/107063392
UnixStreamSocket::UnixStreamSocket(int fd, size_t bufferSize, UnixStreamSocket::Role role)
: bufferSize(bufferSize), role(role)
{
MS_TRACE_STD();
int err;
this->uvHandle = new uv_pipe_t;
this->uvHandle->data = static_cast<void*>(this);
err = uv_pipe_init(DepLibUV::GetLoop(), this->uvHandle, 0);
if (err != 0)
{
delete this->uvHandle;
this->uvHandle = nullptr;
MS_THROW_ERROR_STD("uv_pipe_init() failed: %s", uv_strerror(err));
}
err = uv_pipe_open(this->uvHandle, fd);
if (err != 0)
{
uv_close(reinterpret_cast<uv_handle_t*>(this->uvHandle), static_cast<uv_close_cb>(onClose));
MS_THROW_ERROR_STD("uv_pipe_open() failed: %s", uv_strerror(err));
}
if (this->role == UnixStreamSocket::Role::CONSUMER)
{
// Start reading.
err = uv_read_start(
reinterpret_cast<uv_stream_t*>(this->uvHandle),
static_cast<uv_alloc_cb>(onAlloc),
static_cast<uv_read_cb>(onRead));
if (err != 0)
{
uv_close(reinterpret_cast<uv_handle_t*>(this->uvHandle), static_cast<uv_close_cb>(onClose));
MS_THROW_ERROR_STD("uv_read_start() failed: %s", uv_strerror(err));
}
}
// NOTE: Don't allocate the buffer here. Instead wait for the first uv_alloc_cb().
}
2.3 OnRead函数：
inline static void onRead(uv_stream_t* handle, ssize_t nread, const uv_buf_t* buf)
{
auto* socket = static_cast<UnixStreamSocket*>(handle->data);
if (socket)
socket->OnUvRead(nread, buf);
}
libUV回调时传入了fd、读取的数据信息以及buf位置。
这里强转handle->data后取出socket，然后使用socket调用OnUvRead函数。为什么使用强转后可以调用呢？是因为uv_stream_t这个结构体第一个成员为void*，可以将其转为任何类型的指针。
2.4 OnUvRead函数：
在函数里只是判断了读到数据的大小，判断是否需要调用子类的读函数。
inline void UnixStreamSocket::OnUvRead(ssize_t nread, const uv_buf_t* /*buf*/)
{
MS_TRACE_STD();
if (nread == 0)
return;
// Data received.
if (nread > 0)
{
// Update the buffer data length.
this->bufferDataLen += static_cast<size_t>(nread);
// Notify the subclass.
UserOnUnixStreamRead();
}
// Peer disconnected.
else if (nread == UV_EOF || nread == UV_ECONNRESET)
{
this->isClosedByPeer = true;
// Close local side of the pipe.
Close();
// Notify the subclass.
UserOnUnixStreamSocketClosed();
}
// Some error.
else
{
MS_ERROR_STD("read error, closing the pipe: %s", uv_strerror(nread));
this->hasError = true;
// Close the socket.
Close();
// Notify the subclass.
UserOnUnixStreamSocketClosed();
}
}
2.5 UserOnUnixStreamRead：
void ConsumerSocket::UserOnUnixStreamRead()
{
MS_TRACE();
// Be ready to parse more than a single message in a single chunk.
while (true)
{
if (IsClosed())
return;
size_t readLen = this->bufferDataLen - this->msgStart;
char* msgStart = nullptr;
size_t msgLen;
int nsRet = netstring_read(
reinterpret_cast<char*>(this->buffer + this->msgStart), readLen, &msgStart, &msgLen);
if (nsRet != 0)
{
switch (nsRet)
{
case NETSTRING_ERROR_TOO_SHORT:
{
// Check if the buffer is full.
if (this->bufferDataLen == this->bufferSize)
{
// First case: the incomplete message does not begin at position 0 of
// the buffer, so move the incomplete message to the position 0.
if (this->msgStart != 0)
{
std::memmove(this->buffer, this->buffer + this->msgStart, readLen);
this->msgStart = 0;
this->bufferDataLen = readLen;
}
// Second case: the incomplete message begins at position 0 of the buffer.
// The message is too big, so discard it.
else
{
MS_ERROR(
"no more space in the buffer for the unfinished message being parsed, "
"discarding it");
this->msgStart = 0;
this->bufferDataLen = 0;
}
}
// Otherwise the buffer is not full, just wait.
return;
}
case NETSTRING_ERROR_TOO_LONG:
{
MS_ERROR("NETSTRING_ERROR_TOO_LONG");
break;
}
case NETSTRING_ERROR_NO_COLON:
{
MS_ERROR("NETSTRING_ERROR_NO_COLON");
break;
}
case NETSTRING_ERROR_NO_COMMA:
{
MS_ERROR("NETSTRING_ERROR_NO_COMMA");
break;
}
case NETSTRING_ERROR_LEADING_ZERO:
{
MS_ERROR("NETSTRING_ERROR_LEADING_ZERO");
break;
}
case NETSTRING_ERROR_NO_LENGTH:
{
MS_ERROR("NETSTRING_ERROR_NO_LENGTH");
break;
}
}
// Error, so reset and exit the parsing loop.
this->msgStart = 0;
this->bufferDataLen = 0;
return;
}
// If here it means that msgStart points to the beginning of a message
// with msgLen bytes length, so recalculate readLen.
readLen =
reinterpret_cast<const uint8_t*>(msgStart) - (this->buffer + this->msgStart) + msgLen + 1;
this->listener->OnConsumerSocketMessage(this, msgStart, msgLen);
// If there is no more space available in the buffer and that is because
// the latest parsed message filled it, then empty the full buffer.
if ((this->msgStart + readLen) == this->bufferSize)
{
this->msgStart = 0;
this->bufferDataLen = 0;
}
// If there is still space in the buffer, set the beginning of the next
// parsing to the next position after the parsed message.
else
{
this->msgStart += readLen;
}
// If there is more data in the buffer after the parsed message
// then parse again. Otherwise break here and wait for more data.
if (this->bufferDataLen > this->msgStart)
{
continue;
}
break;
}
}
子类的读函数中会解析信息，如果信息不止有一个块大小则会一直循环解析到信息完整。进行错误判断后就会调用监听者来进行上层回调到“业务处理层”。
2.6 OnConsumerSocketMessage
void UnixStreamSocket::OnConsumerSocketMessage(
ConsumerSocket* /*consumerSocket*/, char* msg, size_t msgLen)
{
MS_TRACE_STD();
try
{
json jsonMessage = json::parse(msg, msg + msgLen);
auto* request = new Channel::Request(this, jsonMessage);
// Notify the listener.
try
{
this->listener->OnChannelRequest(this, request);
}
catch (const MediaSoupTypeError& error)
{
request->TypeError(error.what());
}
catch (const MediaSoupError& error)
{
request->Error(error.what());
}
// Delete the Request.
delete request;
}
catch (const json::parse_error& error)
{
MS_ERROR_STD("JSON parsing error: %s", error.what());
}
catch (const MediaSoupError& error)
{
MS_ERROR_STD("discarding wrong Channel request");
}
}
该函数将接收到的信息转换成json格式向上传递。
而json的类型则是 Request类 的形式：
class Request
{
public:
enum class MethodId
{
WORKER_DUMP = 1,
WORKER_GET_RESOURCE_USAGE,
WORKER_UPDATE_SETTINGS,
WORKER_CREATE_ROUTER,
ROUTER_CLOSE,
ROUTER_DUMP,
ROUTER_CREATE_WEBRTC_TRANSPORT,
ROUTER_CREATE_PLAIN_TRANSPORT,
ROUTER_CREATE_PIPE_TRANSPORT,
ROUTER_CREATE_DIRECT_TRANSPORT,
ROUTER_CREATE_AUDIO_LEVEL_OBSERVER,
TRANSPORT_CLOSE,
TRANSPORT_DUMP,
TRANSPORT_GET_STATS,
TRANSPORT_CONNECT,
TRANSPORT_SET_MAX_INCOMING_BITRATE,
TRANSPORT_RESTART_ICE,
TRANSPORT_PRODUCE,
TRANSPORT_CONSUME,
TRANSPORT_PRODUCE_DATA,
TRANSPORT_CONSUME_DATA,
TRANSPORT_ENABLE_TRACE_EVENT,
PRODUCER_CLOSE,
PRODUCER_DUMP,
PRODUCER_GET_STATS,
PRODUCER_PAUSE,
PRODUCER_RESUME,
PRODUCER_ENABLE_TRACE_EVENT,
CONSUMER_CLOSE,
CONSUMER_DUMP,
CONSUMER_GET_STATS,
CONSUMER_PAUSE,
CONSUMER_RESUME,
CONSUMER_SET_PREFERRED_LAYERS,
CONSUMER_SET_PRIORITY,
CONSUMER_REQUEST_KEY_FRAME,
CONSUMER_ENABLE_TRACE_EVENT,
DATA_PRODUCER_CLOSE,
DATA_PRODUCER_DUMP,
DATA_PRODUCER_GET_STATS,
DATA_CONSUMER_CLOSE,
DATA_CONSUMER_DUMP,
DATA_CONSUMER_GET_STATS,
RTP_OBSERVER_CLOSE,
RTP_OBSERVER_PAUSE,
RTP_OBSERVER_RESUME,
RTP_OBSERVER_ADD_PRODUCER,
RTP_OBSERVER_REMOVE_PRODUCER
};
private:
static std::unordered_map<std::string, MethodId> string2MethodId;
public:
Request(Channel::UnixStreamSocket* channel, json& jsonRequest);
virtual ~Request();
void Accept();
void Accept(json& data);
void Error(const char* reason = nullptr);
void TypeError(const char* reason = nullptr);
public:
// Passed by argument.
Channel::UnixStreamSocket* channel{ nullptr };
uint32_t id{ 0u };
std::string method;
MethodId methodId;
json internal;
json data;
// Others.
bool replied{ false };
};
2.7 Request
Request的构造函数中，将json信息初始化。
Request::Request(Channel::UnixStreamSocket* channel, json& jsonRequest) : channel(channel)
{
MS_TRACE();
auto jsonIdIt = jsonRequest.find("id");
if (jsonIdIt == jsonRequest.end() || !Utils::Json::IsPositiveInteger(*jsonIdIt))
MS_THROW_ERROR("missing id");
this->id = jsonIdIt->get<uint32_t>();
auto jsonMethodIt = jsonRequest.find("method");
if (jsonMethodIt == jsonRequest.end() || !jsonMethodIt->is_string())
MS_THROW_ERROR("missing method");
this->method = jsonMethodIt->get<std::string>();
auto methodIdIt = Request::string2MethodId.find(this->method);
if (methodIdIt == Request::string2MethodId.end())
{
Error("unknown method");
MS_THROW_ERROR("unknown method '%s'", this->method.c_str());
}
this->methodId = methodIdIt->second;
auto jsonInternalIt = jsonRequest.find("internal");
if (jsonInternalIt != jsonRequest.end() && jsonInternalIt->is_object())
this->internal = *jsonInternalIt;
else
this->internal = json::object();
auto jsonDataIt = jsonRequest.find("data");
if (jsonDataIt != jsonRequest.end() && jsonDataIt->is_object())
this->data = *jsonDataIt;
else
this->data = json::object();
}
将json内容转换完成后把Request对象继续向上传递。
2.8 OnChannelRequest
当Worker接收到Request的数据后，便能做出相应的处理了，接下来进入Worker.cpp的OnChannelRequest()
switch里就会根据methodId做出相应的处理，当它为：
a. WORKER_DUMP，表示将Worker中的Router信息都打印出来
b. WORKER_GET_RESOURCE_USAGE，表示将RU的参数信息打印出来
c. WORKER_UPDATE_SETTINGS，表示更新设置
d. WORKER_CREATE_ROUTER，表示创建Router
e. ROUTER_CLOSE，表示关闭
inline void Worker::OnChannelRequest(Channel::UnixStreamSocket* /*channel*/, Channel::Request* request)
{
MS_TRACE();
MS_DEBUG_DEV(
"Channel request received [method:%s, id:%" PRIu32 "]", request->method.c_str(), request->id);
switch (request->methodId)
{
case Channel::Request::MethodId::WORKER_DUMP:
{
json data = json::object();
FillJson(data);
request->Accept(data);
break;
}
case Channel::Request::MethodId::WORKER_GET_RESOURCE_USAGE:
{
json data = json::object();
FillJsonResourceUsage(data);
request->Accept(data);
break;
}
case Channel::Request::MethodId::WORKER_UPDATE_SETTINGS:
{
Settings::HandleRequest(request);
break;
}
case Channel::Request::MethodId::WORKER_CREATE_ROUTER:
{
std::string routerId;
// This may throw.
SetNewRouterIdFromInternal(request->internal, routerId);
auto* router = new RTC::Router(routerId);
this->mapRouters[routerId] = router;
MS_DEBUG_DEV("Router created [routerId:%s]", routerId.c_str());
request->Accept();
break;
}
case Channel::Request::MethodId::ROUTER_CLOSE:
{
// This may throw.
RTC::Router* router = GetRouterFromInternal(request->internal);
// Remove it from the map and delete it.
this->mapRouters.erase(router->id);
delete router;
MS_DEBUG_DEV("Router closed [id:%s]", router->id.c_str());
request->Accept();
break;
}
// Any other request must be delivered to the corresponding Router.
default:
{
// This may throw.
RTC::Router* router = GetRouterFromInternal(request->internal);
router->HandleRequest(request);
break;
}
}
}
三、总结
本章主要捋了一下Channel这部分的信令传输流程，可以看到主要回调形式是通过观察者向上传递。分层处理了信令数据，后续我们会对各个信令的意义进行更详细的分析。
1174
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
