调试、诊断子线程最直接的方式就是像调试、诊断主线程一样,但是无论是动态开启还是静态开启,子线程都不可避免地需要内置一些相关的非业务代码,本文介绍另外一种对子线程代码无侵入的调试方式,另外也介绍一下通过子线程调试主线程的方式。
1.初始化子线程的Inspector
在Node.js启动子线程的时候,会初始化Inspector。
- env_->InitializeInspector(std::move(inspector_parent_handle_));
在分析InitializeInspector之前,我们先看一下inspector_parent_handle_。
- std::unique_ptr<inspector::ParentInspectorHandle> inspector_parent_handle_;
inspector_parent_handle_是一个ParentInspectorHandle对象,这个对象是子线程和主线程通信的桥梁。我们看一下他的初始化逻辑(在主线程里执行)。
- inspector_parent_handle_ = env->inspector_agent()->GetParentHandle(thread_id_, url);
调用agent的GetParentHandle获取一个ParentInspectorHandle对象。
- std::unique_ptr<ParentInspectorHandle> Agent::GetParentHandle(int thread_id, const std::string& url) {
- return client_->getWorkerManager()->NewParentHandle(thread_id, url);
- }
内部其实是通过client_->getWorkerManager()对象的NewParentHandle方法获取ParentInspectorHandle对象,接下来我们看一下WorkerManager的NewParentHandle。
- std::unique_ptr<ParentInspectorHandle> WorkerManager::NewParentHandle(int thread_id, const std::string& url) {
- bool wait = !delegates_waiting_on_start_.empty();
- return std::make_unique<ParentInspectorHandle>(thread_id, url, thread_, wait);
- }
- ParentInspectorHandle::ParentInspectorHandle(
- int id, const std::string& url,
- std::shared_ptr<MainThreadHandle> parent_thread,
- bool wait_for_connect
- )
- : id_(id),
- url_(url),
- parent_thread_(parent_thread),
- wait_(wait_for_connect) {}
最终的架构图如下入所示。
分析完ParentInspectorHandle后继续看一下env_->InitializeInspector(std::move(inspector_parent_handle_))的逻辑(在子线程里执行)。
- int Environment::InitializeInspector(
- std::unique_ptr<inspector::ParentInspectorHandle> parent_handle) {
- std::string inspector_path;
- inspector_path = parent_handle->url();
- inspector_agent_->SetParentHandle(std::move(parent_handle));
- inspector_agent_->Start(inspector_path,
- options_->debug_options(),
- inspector_host_port(),
- is_main_thread());
- }
首先把ParentInspectorHandle对象保存到agent中,然后调用agent的Start方法。
- bool Agent::Start(...) {
- // 新建client对象
- client_ = std::make_shared<NodeInspectorClient>(parent_env_, is_main);
- // 调用agent中保存的ParentInspectorHandle对象的WorkerStarted
- parent_handle_->WorkerStarted(client_->getThreadHandle(), ...);
- }
Agent::Start创建了一个client对象,然后调用ParentInspectorHandle对象的WorkerStarted方法(刚才SetParentHandle的时候保存的),我们看一下这时候的架构图。
接着看parent_handle_->WorkerStarted。
- void ParentInspectorHandle::WorkerStarted(
- std::shared_ptr<MainThreadHandle> worker_thread, bool waiting) {
- std::unique_ptr<Request> request(
- new WorkerStartedRequest(id_, url_, worker_thread, waiting));
- parent_thread_->Post(std::move(request));
- }
WorkerStarted创建了一个WorkerStartedRequest请求,然后通过parent_thread_->Post提交,parent_thread_是MainThreadInterface对象。
- void MainThreadInterface::Post(std::unique_ptr<Request> request) {
- Mutex::ScopedLock scoped_lock(requests_lock_);
- // 之前是空则需要唤醒消费者
- bool needs_notify = requests_.empty();
- // 消息入队
- requests_.push_back(std::move(request));
- if (needs_notify) {
- // 获取当前对象的一个弱引用
- std::weak_ptr<MainThreadInterface>* interface_ptr = new std::weak_ptr<MainThreadInterface>(shared_from_this());
- // 请求V8执行RequestInterrupt入参对应的回调
- isolate_->RequestInterrupt([](v8::Isolate* isolate, void* opaque) {
- // 把执行时传入的参数转成MainThreadInterface
- std::unique_ptr<std::weak_ptr<MainThreadInterface>> interface_ptr {
- static_cast<std::weak_ptr<MainThreadInterface>*>(opaque)
- };
- // 判断对象是否还有效,是则调用DispatchMessages
- if (auto iface = interface_ptr->lock()) iface->DispatchMessages();
- }, static_cast<void*>(interface_ptr));
- }
- // 唤醒消费者
- incoming_message_cond_.Broadcast(scoped_lock);
- }
我们看看这时候的架构图。
接着看回调里执行MainThreadInterface对象DispatchMessages方法的逻辑。
- void MainThreadInterface::DispatchMessages() {
- // 遍历请求队列
- requests_.swap(dispatching_message_queue_);
- while (!dispatching_message_queue_.empty()) {
- MessageQueue::value_type task;
- std::swap(dispatching_message_queue_.front(), task);
- dispatching_message_queue_.pop_front();
- // 执行任务函数
- task->Call(this);
- }
- }
task是WorkerStartedRequest对象,看一下Call方法的代码。
- void Call(MainThreadInterface* thread) override {
- auto manager = thread->inspector_agent()->GetWorkerManager();
- manager->WorkerStarted(id_, info_, waiting_);
- }
接着调用agent的WorkerManager的WorkerStarted。
- void WorkerManager::WorkerStarted(int session_id,
- const WorkerInfo& info,
- bool waiting) {
- children_.emplace(session_id, info);
- for (const auto& delegate : delegates_) {
- Report(delegate.second, info, waiting);
- }
- }
WorkerStarted记录了一个id和上下文,因为delegates_初始化的时候是空的,所以不会执行。至此,子线程Inspector初始化的逻辑就分析完了,结构图如下。
我们发现,和主线程不一样,主线程会启动一个WebSocket服务器接收客户端的连接请求,而子线程只是初始化了一些数据结构。下面我们看一下基于这些数据结构,主线程是如何动态开启调试子线程的。
2.主线程开启调试子线程的能力
我们可以以以下方式开启对子线程的调试。
- const { Worker, workerData } = require('worker_threads');
- const { Session } = require('inspector');
- // 新建一个新的通信通道
- const session = new Session();
- session.connect();
- // 创建子线程
- const worker = new Worker('./httpServer.js', {workerData: {port: 80}});
- // 子线程启动成功后开启调试子线程的能力
- worker.on('online', () => {
- session.post("NodeWorker.enable",
- {waitForDebuggerOnStart: false},
- (err) => {
- err && console.log("NodeWorker.enable", err);
- });
- });
- // 防止主线程退出
- setInterval(() => {}, 100000);
我们先来分析一下connect函数的逻辑。
- connect() {
- this[connectionSymbol] = new Connection((message) => this[onMessageSymbol](message));
- }
新建了一个Connection对象并传入一个回调函数,该回调函数在收到消息时被回调。Connection是C++层导出的对象,由模版类JSBindingsConnection实现。
- template <typename ConnectionType>
- class JSBindingsConnection {}
我们看看导出的路逻辑。
- JSBindingsConnection<Connection>::Bind(env, target);
接着看Bind。
- static void Bind(Environment* env, Local<Object> target) {
- // class_name是Connection
- Local<String> class_name = ConnectionType::GetClassName(env);
- Local<FunctionTemplate> tmpl = env->NewFunctionTemplate(JSBindingsConnection::New);
- tmpl->InstanceTemplate()->SetInternalFieldCount(1);
- tmpl->SetClassName(class_name);
- tmpl->Inherit(AsyncWrap::GetConstructorTemplate(env));
- env->SetProtoMethod(tmpl, "dispatch", JSBindingsConnection::Dispatch);
- env->SetProtoMethod(tmpl, "disconnect", JSBindingsConnection::Disconnect);
- target->Set(env->context(),
- class_name,
- tmpl->GetFunction(env->context()).ToLocalChecked())
- .ToChecked();
- }
当我们在JS层执行new Connection的时候,就会执行JSBindingsConnection::New。
- static void New(const FunctionCallbackInfo<Value>& info) {
- Environment* env = Environment::GetCurrent(info);
- Local<Function> callback = info[0].As<Function>();
- new JSBindingsConnection(env, info.This(), callback);
- }
我们看看新建一个JSBindingsConnection对象时的逻辑。
- JSBindingsConnection(Environment* env,
- Local<Object> wrap,
- Local<Function> callback)
- : AsyncWrap(env, wrap, PROVIDER_INSPECTORJSBINDING),
- callback_(env->isolate(), callback) {
- Agent* inspector = env->inspector_agent();
- session_ = LocalConnection::Connect(
- inspector, std::make_unique<JSBindingsSessionDelegate>(env, this)
- );}static std::unique_ptr<InspectorSession> Connect(
- Agent* inspector,
- std::unique_ptr<InspectorSessionDelegate> delegate
- ) {
- return inspector->Connect(std::move(delegate), false);
- }
最终是传入了一个JSBindingsSessionDelegate对象调用Agent的Connect方法。
- std::unique_ptr<InspectorSession> Agent::Connect(
- std::unique_ptr<InspectorSessionDelegate> delegate,
- bool prevent_shutdown) {
- int session_id = client_->connectFrontend(std::move(delegate),
- prevent_shutdown);
- // JSBindingsConnection对象的session_字段指向的对象
- return std::unique_ptr<InspectorSession>(
- new SameThreadInspectorSession(session_id, client_)
- );
- }
Agent的Connect方法继续调用client_->connectFrontend。
- int connectFrontend(std::unique_ptr<InspectorSessionDelegate> delegate,
- bool prevent_shutdown) {
- int session_id = next_session_id_++;
- channels_[session_id] = std::make_unique<ChannelImpl>(env_,
- client_,
- getWorkerManager(),
- std::move(delegate),
- getThreadHandle(),
- prevent_shutdown);
- return session_id;
- }
connectFrontend新建了一个ChannelImpl对象,在新建ChannelImpl时,会初始化子线程处理的逻辑。
- explicit ChannelImpl(Environment* env,
- const std::unique_ptr<V8Inspector>& inspector,
- std::shared_ptr<WorkerManager> worker_manager,
- std::unique_ptr<InspectorSessionDelegate> delegate,
- std::shared_ptr<MainThreadHandle> main_thread_,
- bool prevent_shutdown)
- : delegate_(std::move(delegate)), prevent_shutdown_(prevent_shutdown),
- retaining_context_(false) {
- session_ = inspector->connect(CONTEXT_GROUP_ID, this, StringView());
- // Node.js拓展命令的处理分发器
- node_dispatcher_ = std::make_unique<protocol::UberDispatcher>(this);
- // trace相关
- tracing_agent_ = std::make_unique<protocol::TracingAgent>(env, main_thread_);
- tracing_agent_->Wire(node_dispatcher_.get());
- // 处理子线程相关
- if (worker_manager) {
- worker_agent_ = std::make_unique<protocol::WorkerAgent>(worker_manager);
- worker_agent_->Wire(node_dispatcher_.get());
- }
- // 处理runtime
- runtime_agent_ = std::make_unique<protocol::RuntimeAgent>();
- runtime_agent_->Wire(node_dispatcher_.get());
- }
我们这里只关注处理子线程相关的逻辑。看一下 worker_agent_->Wire。
- void WorkerAgent::Wire(UberDispatcher* dispatcher) {
- frontend_.reset(new NodeWorker::Frontend(dispatcher->channel()));
- NodeWorker::Dispatcher::wire(dispatcher, this);
- auto manager = manager_.lock();
- workers_ = std::make_shared<NodeWorkers>(frontend_, manager->MainThread());
- }
这时候的架构图如下
接着看一下NodeWorker::Dispatcher::wire(dispatcher, this)的逻辑。
- void Dispatcher::wire(UberDispatcher* uber, Backend* backend){
- std::unique_ptr<DispatcherImpl> dispatcher(new DispatcherImpl(uber->channel(), backend));
- uber->setupRedirects(dispatcher->redirects());
- uber->registerBackend("NodeWorker", std::move(dispatcher));
- }
首先新建了一个DispatcherImpl对象。
- DispatcherImpl(FrontendChannel* frontendChannel, Backend* backend)
- : DispatcherBase(frontendChannel)
- , m_backend(backend) {
- m_dispatchMap["NodeWorker.sendMessageToWorker"] = &DispatcherImpl::sendMessageToWorker;
- m_dispatchMap["NodeWorker.enable"] = &DispatcherImpl::enable;
- m_dispatchMap["NodeWorker.disable"] = &DispatcherImpl::disable;
- m_dispatchMap["NodeWorker.detach"] = &DispatcherImpl::detach;
- }
除了初始化一些字段,另外了一个kv数据结构,这个是一个路由配置,后面我们会看到它的作用。新建完DispatcherImpl后又调用了uber->registerBackend("NodeWorker", std::move(dispatcher))注册该对象。
- void UberDispatcher::registerBackend(const String& name, std::unique_ptr<protocol::DispatcherBase> dispatcher){
- m_dispatchers[name] = std::move(dispatcher);
- }
这时候的架构图如下。
我们看到这里其实是建立了一个路由体系,后面收到命令时就会根据这些路由配置进行转发,类似Node.js Express框架路由机制。这时候可以通过session的post给主线程发送NodeWorker.enable命令来开启子线程的调试。我们分析这个过程。
- post(method, params, callback) {
- // 忽略参数处理
- // 保存请求对应的回调
- if (callback) {
- this[messageCallbacksSymbol].set(id, callback);
- }
- // 调用C++的dispatch
- this[connectionSymbol].dispatch(JSONStringify(message));
- }
this[connectionSymbol]对应的是JSBindingsConnection对象。
- static void Dispatch(const FunctionCallbackInfo<Value>& info) {
- Environment* env = Environment::GetCurrent(info);
- JSBindingsConnection* session;
- ASSIGN_OR_RETURN_UNWRAP(&session, info.Holder());
- if (session->session_) {
- session->session_->Dispatch(
- ToProtocolString(env->isolate(), info[0])->string());
- }
- }
session_是一个SameThreadInspectorSession对象。
- void SameThreadInspectorSession::Dispatch(
- const v8_inspector::StringView& message) {
- auto client = client_.lock();
- client->dispatchMessageFromFrontend(session_id_, message);}void dispatchMessageFromFrontend(int session_id, const StringView& message) {
- channels_[session_id]->dispatchProtocolMessage(message);
- }
最终调用了ChannelImpl的dispatchProtocolMessage。
- void dispatchProtocolMessage(const StringView& message) {
- std::string raw_message = protocol::StringUtil::StringViewToUtf8(message);
- std::unique_ptr<protocol::DictionaryValue> value =
- protocol::DictionaryValue::cast(protocol::StringUtil::parseMessage(
- raw_message, false));
- int call_id;
- std::string method;
- // 解析命令
- node_dispatcher_->parseCommand(value.get(), &call_id, &method);
- // 判断命令是V8内置命令还是Node.js拓展的命令
- if (v8_inspector::V8InspectorSession::canDispatchMethod(
- Utf8ToStringView(method)->string())) {
- session_->dispatchProtocolMessage(message);
- } else {
- node_dispatcher_->dispatch(call_id, method, std::move(value),
- raw_message);
- }
- }
因为NodeWorker.enable是Node.js拓展的命令,所以会走到else里面的逻辑。根据路由配置找到该命令对应的处理逻辑(NodeWorker.enable以.切分,对应两级路由)。
- void UberDispatcher::dispatch(int callId, const String& in_method, std::unique_ptr<Value> parsedMessage, const ProtocolMessage& rawMessage){
- // 找到一级路由配置
- protocol::DispatcherBase* dispatcher = findDispatcher(method);
- std::unique_ptr<protocol::DictionaryValue> messageObject = DictionaryValue::cast(std::move(parsedMessage));
- // 交给一级路由处理器处理
- dispatcher->dispatch(callId, method, rawMessage, std::move(messageObject));
- }
NodeWorker.enable对应的路由处理器代码如下
- void DispatcherImpl::dispatch(int callId, const String& method, const ProtocolMessage& message, std::unique_ptr<protocol::DictionaryValue> messageObject){
- // 查找二级路由
- std::unordered_map<String, CallHandler>::iterator it = m_dispatchMap.find(method);
- protocol::ErrorSupport errors;
- // 找到处理函数
- (this->*(it->second))(callId, method, message, std::move(messageObject), &errors);
- }
dispatch继续寻找命令对应的处理函数,最终找到NodeWorker.enable命令的处理函数为DispatcherImpl::enable。
- void DispatcherImpl::enable(...){
- std::unique_ptr<DispatcherBase::WeakPtr> weak = weakPtr();
- DispatchResponse response = m_backend->enable(...);
- // 返回响应给命令(类似请求/响应模式)
- weak->get()->sendResponse(callId, response);
- }
根据架构图可以知道m_backend是WorkerAgent对象。
- DispatchResponse WorkerAgent::enable(bool waitForDebuggerOnStart) {
- auto manager = manager_.lock();
- std::unique_ptr<AgentWorkerInspectorDelegate> delegate(new AgentWorkerInspectorDelegate(workers_));
- event_handle_ = manager->SetAutoAttach(std::move(delegate));
- return DispatchResponse::OK();
- }
继续调用WorkerManager的SetAutoAttach方法。
- std::unique_ptr<WorkerManagerEventHandle> WorkerManager::SetAutoAttach(
- std::unique_ptr<WorkerDelegate> attach_delegate) {
- int id = ++next_delegate_id_;
- // 保存delegate
- delegates_[id] = std::move(attach_delegate);
- const auto& delegate = delegates_[id];
- // 通知子线程
- for (const auto& worker : children_) {
- Report(delegate, worker.second, false);
- }
- ...
- }
SetAutoAttach遍历子线程。
- void Report(const std::unique_ptr<WorkerDelegate>& delegate,
- const WorkerInfo& info, bool waiting) {
- if (info.worker_thread)
- delegate->WorkerCreated(info.title, info.url, waiting, info.worker_thread);
- }
info是一个WorkerInfo对象,该对象是子线程初始化和主线程建立关系的数据结构。delegate是AgentWorkerInspectorDelegate对象。
- void WorkerCreated(const std::string& title,
- const std::string& url,
- bool waiting,
- std::shared_ptr<MainThreadHandle> target) override {
- workers_->WorkerCreated(title, url, waiting, target);
- }
workers_是一个NodeWorkers对象。
- void NodeWorkers::WorkerCreated(const std::string& title,
- const std::string& url,
- bool waiting,
- std::shared_ptr<MainThreadHandle> target) {
- auto frontend = frontend_.lock();
- std::string id = std::to_string(++next_target_id_);
- // 处理数据通信的delegate
- auto delegate = thread_->MakeDelegateThreadSafe(
- std::unique_ptr<InspectorSessionDelegate>(
- new ParentInspectorSessionDelegate(id, shared_from_this())
- )
- );
- // 建立和子线程V8 Inspector的通信通道
- sessions_[id] = target->Connect(std::move(delegate), true);
- frontend->attachedToWorker(id, WorkerInfo(id, title, url), waiting);
- }
WorkerCreated建立了一条和子线程通信的通道,然后通知命令的发送方通道建立成功。这时候架构图如下。
接着看attachedToWorker。
- void Frontend::attachedToWorker(const String& sessionId, std::unique_ptr<protocol::NodeWorker::WorkerInfo> workerInfo, bool waitingForDebugger){
- std::unique_ptr<AttachedToWorkerNotification> messageData = AttachedToWorkerNotification::create()
- .setSessionId(sessionId)
- .setWorkerInfo(std::move(workerInfo))
- .setWaitingForDebugger(waitingForDebugger)
- .build();
- // 触发NodeWorker.attachedToWorker
- m_frontendChannel->sendProtocolNotification(InternalResponse::createNotification("NodeWorker.attachedToWorker", std::move(messageData)));
- }
继续看sendProtocolNotification
- void sendProtocolNotification(
- std::unique_ptr<Serializable> message) override {
- sendMessageToFrontend(message->serializeToJSON());
- }
- void sendMessageToFrontend(const StringView& message) {
- delegate_->SendMessageToFrontend(message);
- }
这里的delegate_是一个JSBindingsSessionDelegate对象。
- void SendMessageToFrontend(const v8_inspector::StringView& message)
- override {
- Isolate* isolate = env_->isolate();
- HandleScope handle_scope(isolate);
- Context::Scope context_scope(env_->context());
- MaybeLocal<String> v8string = String::NewFromTwoByte(isolate,
- message.characters16(),
- NewStringType::kNormal, message.length()
- );
- Local<Value> argument = v8string.ToLocalChecked().As<Value>();
- // 收到消息执行回调
- connection_->OnMessage(argument);
- }
- // 执行JS层回调
- void OnMessage(Local<Value> value) {
- MakeCallback(callback_.Get(env()->isolate()), 1, &value);
- }
JS层回调逻辑如下。
- [onMessageSymbol](message) {
- const parsed = JSONParse(message);
- // 收到的消息如果是某个请求的响应,则有个id字段记录了请求对应的id,否则则触发事件
- if (parsed.id) {
- const callback = this[messageCallbacksSymbol].get(parsed.id);
- this[messageCallbacksSymbol].delete(parsed.id);
- if (callback) {
- callback(null, parsed.result);
- }
- } else {
- this.emit(parsed.method, parsed);
- this.emit('inspectorNotification', parsed);
- }
- }
主线程拿到Worker Session对一个的id,后续就可以通过命令NodeWorker.sendMessageToWorker加上该id和子线程通信。大致原理如下,主线程通过自己的channel和子线程的channel进行通信,从而达到控制子线程的目的。
我们分析一下NodeWorker.sendMessageToWorker命令的逻辑,对应处理函数为DispatcherImpl::sendMessageToWorker。
- void DispatcherImpl::sendMessageToWorker(...){
- std::unique_ptr<DispatcherBase::WeakPtr> weak = weakPtr();
- DispatchResponse response = m_backend->sendMessageToWorker(in_message, in_sessionId);
- // 响应
- weak->get()->sendResponse(callId, response);
- return;
- }
继续分析m_backend->sendMessageToWorker。
- DispatchResponse WorkerAgent::sendMessageToWorker(const String& message,
- const String& sessionId) {
- workers_->Receive(sessionId, message);
- return DispatchResponse::OK();
- }
- void NodeWorkers::Receive(const std::string& id, const std::string& message) {
- auto it = sessions_.find(id);
- it->second->Dispatch(Utf8ToStringView(message)->string());
- }
sessions_对应的是和子线程的通信的数据结构CrossThreadInspectorSession。看一下该对象的Dispatch方法。
- void Dispatch(const StringView& message) override {
- state_.Call(&MainThreadSessionState::Dispatch,
- StringBuffer::create(message));
- }
再次调了MainThreadSessionState::Dispatch
- void Dispatch(std::unique_ptr<StringBuffer> message) {
- session_->Dispatch(message->string());
- }
session_是SameThreadInspectorSession对象。继续看它的Dispatch方法。
- void SameThreadInspectorSession::Dispatch(
- const v8_inspector::StringView& message) {
- auto client = client_.lock();
- client->dispatchMessageFromFrontend(session_id_, message);}void dispatchMessageFromFrontend(int session_id, const StringView& message) {
- channels_[session_id]->dispatchProtocolMessage(message);
- }
通过层层调用,最终拿到了一个合子线程通信的channel,dispatchProtocolMessage方法刚才已经分析过,该方法会根据命令做不同的处理,因为我们这里发送的是V8内置的命令,所以会交给V8 Inspector处理。当V8 Inspector处理完后,会通过ChannelImpl的sendResponse返回结果。
- void sendResponse(
- int callId,
- std::unique_ptr<v8_inspector::StringBuffer> message) override {
- sendMessageToFrontend(message->string());
- }
- void sendMessageToFrontend(const StringView& message) {
- delegate_->SendMessageToFrontend(message);
- }
这里的delegate_是ParentInspectorSessionDelegate对象。
- void SendMessageToFrontend(const v8_inspector::StringView& msg) override {
- std::string message = protocol::StringUtil::StringViewToUtf8(msg);
- workers_->Send(id_, message);
- }
- void NodeWorkers::Send(const std::string& id, const std::string& message) {
- auto frontend = frontend_.lock();
- if (frontend)
- frontend->receivedMessageFromWorker(id, message);
- }
- void Frontend::receivedMessageFromWorker(const String& sessionId, const String& message){
- std::unique_ptr<ReceivedMessageFromWorkerNotification> messageData = ReceivedMessageFromWorkerNotification::create()
- .setSessionId(sessionId)
- .setMessage(message)
- .build();
- // 触发NodeWorker.receivedMessageFromWorker
- m_frontendChannel->sendProtocolNotification(InternalResponse::createNotification("NodeWorker.receivedMessageFromWorker", std::move(messageData)));
- }
m_frontendChannel是主线程的ChannelImpl对象。
- void sendProtocolNotification(
- std::unique_ptr<Serializable> message) override {
- sendMessageToFrontend(message->serializeToJSON());
- }
- void sendMessageToFrontend(const StringView& message) {
- delegate_->SendMessageToFrontend(message);
- }
delegate_是C++层传入的JSBindingsSessionDelegate对象。最终通过JSBindingsSessionDelegate对象回调JS层,之前已经分析过就不再赘述。至此,主线程就具备了控制子线程的能力,但是控制方式有很多种。
2.1 使用通用的V8命令
通过下面代码收集子线程的CPU Profile信息。
- const { Worker, workerData } = require('worker_threads');
- const { Session } = require('inspector');
- const session = new Session();
- session.connect();
- let id = 1;
- function post(sessionId, method, params, callback) {
- session.post('NodeWorker.sendMessageToWorker', {
- sessionId,
- message: JSON.stringify({ id: id++, method, params })
- }, callback);
- }
- session.on('NodeWorker.attachedToWorker', (data) => {
- post(data.params.sessionId, 'Profiler.enable');
- post(data.params.sessionId, 'Profiler.start');
- // 收集一段时间后提交停止收集命令
- setTimeout(() => {
- post(data.params.sessionId, 'Profiler.stop');
- }, 10000)
- });
- session.on('NodeWorker.receivedMessageFromWorker', ({ params: { message }}) => {
- const data = JSON.parse(message);
- console.log(data);
- });
- const worker = new Worker('./httpServer.js', {workerData: {port: 80}});
- worker.on('online', () => {
- session.post("NodeWorker.enable",{waitForDebuggerOnStart: false}, (err) => { console.log(err, "NodeWorker.enable");});
- });
- setInterval(() => {}, 100000);
通过这种方式可以通过命令控制子线程的调试和数据收集。
2.2 在子线程中动态执行脚本
可以通过执行脚本开启子线程的WebSocket服务,像调试主线程一样。
- const { Worker, workerData } = require('worker_threads');
- const { Session } = require('inspector');
- const session = new Session();
- session.connect();
- let workerSessionId;
- let id = 1;
- function post(method, params) {
- session.post('NodeWorker.sendMessageToWorker', {
- sessionId: workerSessionId,
- message: JSON.stringify({ id: id++, method, params })
- });
- }
- session.on('NodeWorker.receivedMessageFromWorker', ({ params: { message }}) => {
- const data = JSON.parse(message);
- console.log(data);
- });
- session.on('NodeWorker.attachedToWorker', (data) => {
- workerSessionId = data.params.sessionId;
- post("Runtime.evaluate", {
- includeCommandLineAPI: true,
- expression: `const inspector = process.binding('inspector');
- inspector.open();
- inspector.url();
- `
- }
- );
- });
- const worker = new Worker('./httpServer.js', {workerData: {port: 80}});
- worker.on('online', () => {
- session.post("NodeWorker.enable",{waitForDebuggerOnStart: false}, (err) => { err && console.log("NodeWorker.enable", err);});
- });
- setInterval(() => {}, 100000);
执行上面的代码就拿到以下输出
- {
- id: 1,
- result: {
- result: {
- type: 'string',
- value: 'ws://127.0.0.1:9229/c0ca16c8-55aa-4651-9776-fca1b27fc718'
- }
- }
- }
通过该地址,客户端就可以对子线程进行调试了。上面代码里使用process.binding而不是require加载inspector,因为刚才通过NodeWorker.enable命令为子线程创建了一个到子线程Inspector的channel,而JS模块里判断如果channel非空则报错Inspector已经打开。所以这里需要绕过这个限制,直接加载C++模块开启WebSocket服务器。
3.子线程调试主线程
不仅可以通过主线程调试子线程,还可以通过子线程调试主线程。Node.js在子线程暴露了connectToMainThread方法连接到主线程的Inspector(只能在work_threads中使用),实现的原理和之前分析的类似,主要是子线程连接到主线程的V8 Inspector,通过和该Inspector完成对主线程的控制。看下面一个例子。主线程代码
- const { Worker, workerData } = require('worker_threads');const http = require('http');const worker = new Worker('./worker.js', {workerData: {port: 80}});
- http.createServer((_, res) => {
- res.end('main');
- }).listen(8000);
worker.js代码如下:
- const fs = require('fs');
- const { workerData: { port } } = require('worker_threads');
- const { Session } = require('inspector');
- const session = new Session();
- session.connectToMainThread();
- session.post('Profiler.enable');
- session.post('Profiler.start');
- setTimeout(() => {
- session.post('Profiler.stop', (err, data) => {
- if (data.profile) {
- fs.writeFileSync('./profile.cpuprofile', JSON.stringify(data.profile));
- }
- });
- }, 5000)