Thrift线程和状态机分析.pdf
目录
1. 工作线程和IO线程
启动Thrift时,可启动两类线程,一是TNonblockingIOThread,另一是Worker:
TNonblockingIOThread负责接受连接,和收发数据;而Worker负责回调服务端的用户函数。
TNonblockingIOThread::registerEvents主要做了两件事:
1) 注册TNonblockingIOThread::listenHandler(),这个是用来接受连接请求的;
2) 注册TNonblockingIOThread::notifyHandler(),这个是用来监听管道的。
TNonblockingIOThread和Worker两类线程间通过队列进行通讯,队列类型为std::queue
class ThreadManager::Task: public Runnable { public: void run() { // runnable_实际为TNonblockingServer::TConnection::Task runnable_->run(); }
private: // 这里的Runnable实际为TNonblockingServer::TConnection::Task // 在TNonblockingServer::TConnection::transition()中被push进来 boost::shared_ptr<Runnable> runnable_; }; |
2. TNonblockingServer::TConnection::transition()
transition()为状态切换函数,状态有两种:一是socket的状态,另一是rpc会话的状态。APP开头的是rpc会话的状态,SOCKET开头的是socket的状态。
在APP_READ_REQUEST状态发生在IO线程中,addTask()会将任务转交给或工作线程,然后由工作线程回调服务端的函数。
TNonblockingServer::TConnection::Task { public: void run() { // 回调 processor_->process(input_, output_, connectionContext_);
// 回调完后通知, // 从工作线程重回到IO线程 connection_->notifyIOThread(); // ioThread_->notify(this); // 这个将触发TNonblockingIOThread::notifyHandler() } };
TNonblockingIOThread::notifyHandler() { // 从管道中取出connection的指针地址 TNonblockingServer::TConnection* connection = NULL; int nBytes = recv(fd, cast_sockopt(&connection), kSize, 0); connection->transition(); // 进入状态转换函数 } |
3. RPC函数被调用过程
IO线程收到完整的RPC请求包时,以任务方式转给工作线程,然后由工作线程回调用户写的RPC函数。
完成的调用过程如下图所示:
任务从IO线程进入工作线程:
4. 管道和任务队列
IO线程以Task方式将连接交给工作线程,而工作线程在回调完后,以管道方式还回给IO线程。连接从IO线程进入到或工作线程后,会从libevent中删除,返回后再进入libevent。
5. 对象间关系
class TNonblockingServer: public TServer { public: void serve() // 用户可以直接调用server(),但直接调用run()是更好的做法 { // 创建socket监听 // 创建TNonblockingIOThread // 通过Thread启动TNonblockingIOThread } };
class TServer: public concurrency::Runnable { public: virtual void serve() = 0; virtual void run() // 用户也可以直接调用run() { serve(); } }; |
6. 相关代码摘要
// 线程 // thrift支持原生posix线程和boost线程
void PthreadThread::start() { // PthreadThread是一个Posix线程类 pthread_create(&pthread_, &thread_attr, threadMain, (void*)selfRef); }
static void* PthreadThread::threadMain(void* arg) { thread->runnable()->run(); }
// 以下为IO线程
/// Three states for sockets: recv frame size, recv data, and send mode enum TSocketState { SOCKET_RECV_FRAMING, SOCKET_RECV, SOCKET_SEND };
/** * Five states for the nonblocking server: * 1) initialize * 2) read 4 byte frame size * 3) read frame of data * 4) send back data (if any) * 5) force immediate connection close */ enum TAppState { APP_INIT, // 初始化 APP_READ_FRAME_SIZE, // 接收包大小 APP_READ_REQUEST, // 接收包数据 APP_WAIT_TASK, APP_SEND_RESULT, // 发送数据 APP_CLOSE_CONNECTION // 关闭连接 };
// 启动监听和IO线程 void TNonblockingServer::serve() { createAndListenOnSocket();
for (uint32_t id = 0; id < numIOThreads_; ++id) { // TNonblockingIOThread是一个Runnable // 以委托方式被运行在PthreadThread中 thread = new TNonblockingIOThread(this, id, listenFd, useHighPriorityIOThreads_); ioThreads_.push_back(thread); } for (uint32_t i = 1; i < ioThreads_.size(); ++i) { // PthreadThread thread; thread->start(); }
ioThreads_[0]->run(); // 这将阻塞调用线程 for (uint32_t i = 0; i < ioThreads_.size(); ++i) { ioThreads_[i]->join(); } }
void TNonblockingIOThread::run() { eventBase_ = event_base_new();
// IO线程在启动时会调用registerEvents() // 在registerEvents()中完成两个回调函数的注册:listenHandler和notifyHandler // listenHandler回调负责接受请求,并创建连接对象 registerEvents(); event_base_loop(eventBase_, 0); // libevent }
void TNonblockingIOThread::registerEvents() { // listenHandler和socket关联 event_set(&serverEvent_, listenSocket_, EV_READ | EV_PERSIST, TNonblockingIOThread::listenHandler, server_);
// notifyHandler和pipe关联 event_set(?ificationEvent_, getNotificationRecvFD(), EV_READ | EV_PERSIST, TNonblockingIOThread::notifyHandler, this); }
static void listenHandler(evutil_socket_t fd, short which, void* v) { ((TNonblockingServer*)v)->handleEvent(fd, which); }
void TNonblockingServer::handleEvent(int fd, short which) { accept(); createConnection(); }
TNonblockingServer::TConnection* TNonblockingServer::createConnection() { // 会将自己绑定到一个线程 // 采用轮询的方式选择线程 // int selectedThreadIdx = nextIOThread_; // nextIOThread_ = (nextIOThread_ + 1) % ioThreads_.size();
// std::stack // 使用了内存池connectionStack_
// App状态:APP_INIT // Socket状态:SOCKET_RECV_FRAMING }
static void eventHandler(evutil_socket_t fd, short /* which */, void* v) { assert(fd == ((TConnection*)v)->getTSocket()->getSocketFD()); ((TConnection*)v)->workSocket(); }
void TNonblockingServer::TConnection::setFlags(short eventFlags) { event_set(&event_, tSocket_->getSocketFD(), eventFlags_, TConnection::eventHandler, this); }
void TNonblockingServer::TConnection::setRead() { setFlags(EV_READ | EV_PERSIST); }
void TNonblockingServer::TConnection::setWrite() { setFlags(EV_WRITE | EV_PERSIST); }
void TNonblockingServer::TConnection::setIdle() { setFlags(0); }
void TNonblockingServer::TConnection::workSocket() { case SOCKET_RECV_FRAMING: TSocket::read(); // 接收包大小 transition(); case SOCKET_RECV: TSocket::read(); // 接收包数据 transition(); case SOCKET_SEND: TSocket::write_partial(); // 发送数据(非阻塞的) transition(); }
void TNonblockingIOThread::notifyHandler(evutil_socket_t fd, short which, void* v) { recv(); connection->transition(); }
// transition()为状态迁移函数 void TNonblockingServer::TConnection::transition() { case APP_INIT: setRead(); case APP_WAIT_TASK: setWrite(); case APP_READ_REQUEST: setIdle(); }
TNonblockingServer::TConnection::Task { public: void run() { // 回调 processor_->process(input_, output_, connectionContext_); // 回调完后通知, // 从工作线程重回到IO线程 // connection_的指针地址将通过管道传给工作线程 connection_->notifyIOThread(); // ioThread_->notify(this); } };
TNonblockingIOThread::notifyHandler() { // 从管道中取出connection的指针地址 TNonblockingServer::TConnection* connection = NULL; int nBytes = recv(fd, cast_sockopt(&connection), kSize, 0); connection->transition(); // 进入状态转换函数 }
// 以下为工作线程 class ThreadManager::Impl : public ThreadManager; class SimpleThreadManager : public ThreadManager::Impl; class ThreadManager::Worker: public Runnable; class ThreadManager::Task : public Runnable;
void SimpleThreadManager::start() { // workerCount_为工作线程数 addWorker(workerCount_); }
void ThreadManager::Impl::addWorker(size_t value) { for (size_t ix = 0; ix < value; ix++) { worker = new ThreadManager::Worker(this);
// thread为PthreadThread // 调用了worker->run(); thread->start(); } }
void ThreadManager::Worker::run() { ThreadManager::Task task; task->run(); }
class ThreadManager::Task: public Runnable { public: void run() { // runnable_实际为TNonblockingServer::TConnection::Task runnable_->run(); }
private: // 这里的Runnable实际为TNonblockingServer::TConnection::Task // 在TNonblockingServer::TConnection::transition()中被push进来
boost::shared_ptr };
void ThreadManager::Impl::add(shared_ptr {
// std::queue task = new ThreadManager::Task(value, expiration); tasks_.push(task); }
// 两者关系 class TNonblockingServer: public TServer { public:
TNonblockingServer(const boost::shared_ptr
private: // TNonblockingServer关联了ThreadManager
boost::shared_ptr };
// 工作线程将回调TNonblockingServer::TConnection::Task class TNonblockingServer::TConnection::Task: public Runnable { };
// task为TNonblockingServer::TConnection::Task
void TNonblockingServer::addTask(boost::shared_ptr { // 将任务交给工作线程 // threadManager_为SimpleThreadManager threadManager_->add(task, 0LL, taskExpireTime_); }
void TNonblockingServer::TConnection::transition() { case APP_READ_REQUEST: if (server_->isThreadPoolProcessing()) {
boost::shared_ptr new TNonblockingServer::TConnection::Task( processor_, inputProtocol_, outputProtocol_, this);
// server_为TNonblockingServer // 回调交给工作线程,IO线程不做这个工作 server_->addTask(task); // server_为TNonblockingServer } else { // 调用TNonblockingServer的构造函数时, // 如果没有指定参数ThreadManager,则会走这条分支 // 这种情况下,isThreadPoolProcessing()返回false processor_->process(inputProtocol_, outputProtocol_, connectionContext_); } }
void TNonblockingServer::TConnection::Task::run() { // 回调 processor_->process(input_, output_, connectionContext_); }
内嵌关系: 1) TNonblockingServer内嵌了类TConnection,而TConnection又内嵌了类Task 2) ThreadManager内嵌了类Impl、类Worker和类Task(注意区分于TConnection内嵌的Task),而Impl又是ThreadManager的子类,而Task是对Runnable的实现
class TNonblockingServer: public TServer { public: void serve() // 用户可以直接调用server(),但直接调用run()是更好的做法 { // 创建socket监听 // 创建TNonblockingIOThread // 通过Thread启动TNonblockingIOThread } };
class TServer: public concurrency::Runnable { public: virtual void serve() = 0; virtual void run() // 用户也可以直接调用run() { serve(); } }; |