前言:虽然Async hooks至此还是实验性API,但是他的确可以解决应用中的一些问题,比如日志和调用栈跟踪。本文从应用和原理方便介绍一下Node.js的Async hooks。
在Node.js的env对象中有一个AsyncHooks对象,负责Node.js进程中async_hooks的管理。我们看一下定义。
class AsyncHooks : public MemoryRetainer {
public:
enum Fields {
// 五种钩子
kInit,
kBefore,
kAfter,
kDestroy,
kPromiseResolve,
// 钩子总数
kTotals,
// async_hooks开启的个数
kCheck,
// 记录栈的top指针
kStackLength,
// 数组大小
kFieldsCount,
};
enum UidFields {
kExecutionAsyncId,
kTriggerAsyncId,
// 当前async id的值
kAsyncIdCounter,
kDefaultTriggerAsyncId,
kUidFieldsCount,
};
private:
inline AsyncHooks();
// 异步资源的类型
std::array<v8::Eternal<v8::String>, AsyncWrap::PROVIDERS_LENGTH> providers_;
// 栈
AliasedFloat64Array async_ids_stack_;
// 整形数组,每个元素值的意义和Fields对应
AliasedUint32Array fields_;
// 整形数组,每个元素值的意义和UidFields对应
AliasedFloat64Array async_id_fields_;
};
结构图如下 接下来看一下env的AsyncHooks对象提供了哪些API,这些API是上层的基础。
我们看一下env对象中获取AsyncHooks对象对应字段的API。
// 获取对应的字段
inline AliasedUint32Array& AsyncHooks::fields() {
return fields_;
}
inline AliasedFloat64Array& AsyncHooks::async_id_fields() {
return async_id_fields_;
}
inline AliasedFloat64Array& AsyncHooks::async_ids_stack() {
return async_ids_stack_;
}
// 获取资源类型
inline v8::Local<v8::String> AsyncHooks::provider_string(int idx) {
return providers_[idx].Get(env()->isolate());
}
// 新建资源的时候,获取新的async id
inline double Environment::new_async_id() {
async_hooks()->async_id_fields()[AsyncHooks::kAsyncIdCounter] += 1;
return async_hooks()->async_id_fields()[AsyncHooks::kAsyncIdCounter];
}
// 获取当前async id
inline double Environment::execution_async_id() {
return async_hooks()->async_id_fields()[AsyncHooks::kExecutionAsyncId];
}
// 获取当前trigger async id
inline double Environment::trigger_async_id() {
return async_hooks()->async_id_fields()[AsyncHooks::kTriggerAsyncId];
}
// 获取默认的trigger async id,如果没有设置,则获取当前的async id
inline double Environment::get_default_trigger_async_id() {
double default_trigger_async_id = async_hooks()->async_id_fields()[AsyncHooks::kDefaultTriggerAsyncId];
// If defaultTriggerAsyncId isn't set, use the executionAsyncId
if (default_trigger_async_id < 0)
default_trigger_async_id = execution_async_id();
return default_trigger_async_id;
}
inline void AsyncHooks::push_async_ids(double async_id,
double trigger_async_id) {
// 获取当前栈顶指针
uint32_t offset = fields_[kStackLength];
// 不够则扩容
if (offset * 2 >= async_ids_stack_.Length())
grow_async_ids_stack();
// 把旧的上下文压栈
async_ids_stack_[2 * offset] = async_id_fields_[kExecutionAsyncId];
async_ids_stack_[2 * offset + 1] = async_id_fields_[kTriggerAsyncId];
// 栈指针加一
fields_[kStackLength] += 1;
// 记录当前上下文
async_id_fields_[kExecutionAsyncId] = async_id;
async_id_fields_[kTriggerAsyncId] = trigger_async_id;
}
// 和上面的逻辑相反
inline bool AsyncHooks::pop_async_id(double async_id) {
if (fields_[kStackLength] == 0) return false;
uint32_t offset = fields_[kStackLength] - 1;
async_id_fields_[kExecutionAsyncId] = async_ids_stack_[2 * offset];
async_id_fields_[kTriggerAsyncId] = async_ids_stack_[2 * offset + 1];
fields_[kStackLength] = offset;
return fields_[kStackLength] > 0;
}
接着看一下异步资源的基类AsyncWrap。所有依赖于C、C++层实现的资源(比如TCP、UDP)都会继承AsyncWrap。看看该类的定义。
class AsyncWrap : public BaseObject {
private:
ProviderType provider_type_ = PROVIDER_NONE;
double async_id_ = kInvalidAsyncId;
double trigger_async_id_;
};
我们看到每个AsyncWrap对象都有asyncid、trigger_async_id_和provider_type_属性,这正是在init回调里拿到的数据。我们看看AsyncWrap的构造函数。接下来看一下新建一个资源(AsyncWrap)时的逻辑。
AsyncWrap::AsyncWrap(Environment* env,
Local<Object> object,
ProviderType provider,
double execution_async_id,
bool silent)
: AsyncWrap(env, object) {
// 资源类型
provider_type_ = provider;
AsyncReset(execution_async_id, silent);
}
void AsyncWrap::AsyncReset(Local<Object> resource, double execution_async_id,
bool silent) {
// 获取一个新的async id,execution_async_id默认是kInvalidAsyncId
async_id_ = execution_async_id == kInvalidAsyncId ? env()->new_async_id()
: execution_async_id;
// 获取trigger async id
trigger_async_id_ = env()->get_default_trigger_async_id();
// 执行init钩子
EmitAsyncInit(env(), resource,
env()->async_hooks()->provider_string(provider_type()),
async_id_, trigger_async_id_);
}
接着看EmitAsyncInit
void AsyncWrap::EmitAsyncInit(Environment* env,
Local<Object> object,
Local<String> type,
double async_id,
double trigger_async_id) {
AsyncHooks* async_hooks = env->async_hooks();
HandleScope scope(env->isolate());
Local<Function> init_fn = env->async_hooks_init_function();
Local<Value> argv[] = {
Number::New(env->isolate(), async_id),
type,
Number::New(env->isolate(), trigger_async_id),
object,
};
TryCatchScope try_catch(env, TryCatchScope::CatchMode::kFatal);
// 执行init回调
USE(init_fn->Call(env->context(), object, arraysize(argv), argv));
}
那么env->async_hooks_init_function()的值是什么呢?这是在Node.js初始化时设置的。
const { nativeHooks } = require('internal/async_hooks');
internalBinding('async_wrap').setupHooks(nativeHooks);
SetupHooks的实现如下
static void SetupHooks(const FunctionCallbackInfo<Value>& args) {
Environment* env = Environment::GetCurrent(args);
Local<Object> fn_obj = args[0].As<Object>();
#define SET_HOOK_FN(name) \
do { \
Local<Value> v = \
fn_obj->Get(env->context(), \
FIXED_ONE_BYTE_STRING(env->isolate(), #name)) \
.ToLocalChecked(); \
CHECK(v->IsFunction()); \
env->set_async_hooks_##name##_function(v.As<Function>()); \
} while (0)
// 保存到env中
SET_HOOK_FN(init);
SET_HOOK_FN(before);
SET_HOOK_FN(after);
SET_HOOK_FN(destroy);
SET_HOOK_FN(promise_resolve);
#undef SET_HOOK_FN
}
nativeHooks的实现如下
nativeHooks: {
init: emitInitNative,
before: emitBeforeNative,
after: emitAfterNative,
destroy: emitDestroyNative,
promise_resolve: emitPromiseResolveNative
}
这些Hooks会执行对应的回调,比如emitInitNative
function emitInitNative(asyncId, type, triggerAsyncId, resource) {
for (var i = 0; i < active_hooks.array.length; i++) {
if (typeof active_hooks.array[i][init_symbol] === 'function') {
active_hooks.array[i][init_symbol](
asyncId, type, triggerAsyncId,
resource
);
}
}
}
active_hooks.array的值就是我们在业务代码里设置的钩子,每次调研createHooks的时候就对应数组的一个元素。
当业务代码异步请求底层API,并且底层满足条件时,就会执行上层的回调,比如监听一个socket时,有连接到来。Node.js就会调用MakeCallback函数执行回调。
MaybeLocal<Value> AsyncWrap::MakeCallback(const Local<Function> cb,
int argc,
Local<Value>* argv) {
// 当前AsyncWrap对象对应的执行上下文
ProviderType provider = provider_type();
async_context context { get_async_id(), get_trigger_async_id() };
MaybeLocal<Value> ret = InternalMakeCallback(env(), object(), cb, argc, argv, context);
return ret;
}
MakeCallback中会调用InternalMakeCallback。
MaybeLocal<Value> InternalMakeCallback(Environment* env,
Local<Object> recv,
const Local<Function> callback,
int argc,
Local<Value> argv[],
async_context asyncContext) {
// 新建一个scope
InternalCallbackScope scope(env, recv, asyncContext);
// 执行回调
callback->Call(env->context(), recv, argc, argv);
// 关闭scope
scope.Close();
}
我们看看新建和关闭scope都做了什么事情。
InternalCallbackScope::InternalCallbackScope(Environment* env,
Local<Object> object,
const async_context& asyncContext,
int flags)
: env_(env),
async_context_(asyncContext),
object_(object),
skip_hooks_(flags & kSkipAsyncHooks),
skip_task_queues_(flags & kSkipTaskQueues) {
// v14版本中,是先触发before再push上下文,顺序是不对的,v16已经改过来。
// 当前执行上下文入栈
env->async_hooks()->push_async_ids(async_context_.async_id,
async_context_.trigger_async_id);
// 触发before钩子
if (asyncContext.async_id != 0 && !skip_hooks_) {
AsyncWrap::EmitBefore(env, asyncContext.async_id);
}
pushed_ids_ = true;
}
在scope里会把当前AsyncWrap对象的执行上下文作为当前执行上下文,并且触发before钩子,然后执行业务回调,所以我们在回调里获取当前执行上下文时就拿到了AsyncWrap对应的值( 调用executionAsyncId),接着看Close
void InternalCallbackScope::Close() {
// 执行
if (pushed_ids_)
env_->async_hooks()->pop_async_id(async_context_.async_id);
if (async_context_.async_id != 0 && !skip_hooks_) {
AsyncWrap::EmitAfter(env_, async_context_.async_id);
}
}
Close在执行回调后被调用,主要是恢复当前执行上下文并且触发after钩子。
并不是所有的异步资源都是底层实现的,比如定时器,tick也被定义为异步资源,因为他们都是和回调相关。这种异步资源是在JS层实现的,这里只分析Timeout。
我们看一下执行setTimeout时的核心逻辑。
function setTimeout(callback, after, arg1, arg2, arg3) {
const timeout = new Timeout(callback, after, args, false, true);
return timeout;
}
function Timeout(callback, after, args, isRepeat, isRefed) {
initAsyncResource(this, 'Timeout');
}
function initAsyncResource(resource, type) {
// 获取新的async id
const asyncId = resource[async_id_symbol] = newAsyncId();
const triggerAsyncId = resource[trigger_async_id_symbol] = getDefaultTriggerAsyncId();
// 是否设置了init钩子,是则触发回调
if (initHooksExist())
emitInit(asyncId, type, triggerAsyncId, resource);
}
执行setTimeout时,Node.js会创建一个Timeout对象,设置async_hooks相关的上下文并记录到Timeout对象中。然后触发init钩子。
function emitInitScript(asyncId, type, triggerAsyncId, resource) {
emitInitNative(asyncId, type, triggerAsyncId, resource);
}
以上代码会执行每个async_hooks对象的init回调(通常我们只有一个async_hooks对象)。
当定时器到期时,会执行回调,我们看看相关的逻辑。
// 触发before钩子
emitBefore(asyncId, timer[trigger_async_id_symbol]);
// 执行回调
timer._onTimeout();
// 触发after回调
emitAfter(asyncId);
我们看到执行超时回调的前后会触发对应的钩子。
function emitBeforeScript(asyncId, triggerAsyncId) {
// 和底层的push_async_ids逻辑一样
pushAsyncIds(asyncId, triggerAsyncId);
// 如果有回调则执行
if (async_hook_fields[kBefore] > 0)
emitBeforeNative(asyncId);
}
function emitAfterScript(asyncId) {
// 设置了after回调则emit
if (async_hook_fields[kAfter] > 0)
emitAfterNative(asyncId);
// 和底层的pop_async_ids逻辑一样
popAsyncIds(asyncId);
}
JS层的实现和底层是保持一致的。如果我们在setTimeout回调里新建一个资源,比如再次执行setTimeout,这时候trigger async id就是第一个setTimeout对应的async id,所以就连起来了,后面我们会看到具体的例子。
Node.js为了避免过多通过参数传递的方式传递async id,就设计了DefaultTriggerAsyncIdScope。DefaultTriggerAsyncIdScope的作用类似在多个函数外维护一个变量,多个函数都可以通过DefaultTriggerAsyncIdScope获得trigger async id,而不需要通过层层传递的方式,他的实现非常简单。
class DefaultTriggerAsyncIdScope {
private:
AsyncHooks* async_hooks_;
double old_default_trigger_async_id_;
};
inline AsyncHooks::DefaultTriggerAsyncIdScope ::DefaultTriggerAsyncIdScope(
Environment* env, double default_trigger_async_id)
: async_hooks_(env->async_hooks()) {
// 记录旧的id,设置新的id
old_default_trigger_async_id_ =
async_hooks_->async_id_fields()[AsyncHooks::kDefaultTriggerAsyncId];
async_hooks_->async_id_fields()[AsyncHooks::kDefaultTriggerAsyncId] =
default_trigger_async_id;
}
// 恢复
inline AsyncHooks::DefaultTriggerAsyncIdScope ::~DefaultTriggerAsyncIdScope() {
async_hooks_->async_id_fields()[AsyncHooks::kDefaultTriggerAsyncId] =
old_default_trigger_async_id_;
}
DefaultTriggerAsyncIdScope主要是记录旧的id,然后把新的id设置到env中,当其他函数调用get_default_trigger_async_id时就可以获取设置的async id。同样JS层也实现了类似的API。
function defaultTriggerAsyncIdScope(triggerAsyncId, block, ...args) {
const oldDefaultTriggerAsyncId = async_id_fields[kDefaultTriggerAsyncId];
async_id_fields[kDefaultTriggerAsyncId] = triggerAsyncId;
try {
return block(...args);
} finally {
async_id_fields[kDefaultTriggerAsyncId] = oldDefaultTriggerAsyncId;
}
}
在执行block函数时,可以获取到设置的值,而不需要传递,执行完block后恢复。我们看看如何使用。下面摘自net模块的代码。
// 获取handle里的async id
this[async_id_symbol] = getNewAsyncId(this._handle);
defaultTriggerAsyncIdScope(this[async_id_symbol],
process.nextTick,
emitListeningNT,
this);
我们看一下这里具体的情况。在defaultTriggerAsyncIdScope中会以emitListeningNT为入参执行process.nextTick。我们看看nextTick的实现。
function nextTick(callback) {
// 获取新的async id
const asyncId = newAsyncId();
// 获取默认的trigger async id,即刚才设置的
const triggerAsyncId = getDefaultTriggerAsyncId();
const tickObject = {
[async_id_symbol]: asyncId,
[trigger_async_id_symbol]: triggerAsyncId,
callback,
args
};
if (initHooksExist())
// 创建了新的资源,触发init钩子
emitInit(asyncId, 'TickObject', triggerAsyncId, tickObject);
queue.push(tickObject);
}
我们看到在nextTick中通过getDefaultTriggerAsyncId拿到了trigger async id。
function getDefaultTriggerAsyncId() {
const defaultTriggerAsyncId = async_id_fields[kDefaultTriggerAsyncId];
if (defaultTriggerAsyncId < 0)
return async_id_fields[kExecutionAsyncId];
return defaultTriggerAsyncId;
}
getDefaultTriggerAsyncId返回的就是刚才通过defaultTriggerAsyncIdScope设置的async id。所以在触发TickObject的init钩子时用户就可以拿到对应的id。不过更重要的时,在异步执行nextTick的任务时,还可以拿到原始的trigger async id。因为该id记录在tickObject中。我们看看执行tick任务时的逻辑。
function processTicksAndRejections() {
let tock;
do {
while (tock = queue.shift()) {
// 拿到对应的async 上下文
const asyncId = tock[async_id_symbol];
emitBefore(asyncId, tock[trigger_async_id_symbol]);
try {
const callback = tock.callback;
callback();
} finally {
if (destroyHooksExist())
emitDestroy(asyncId);
}
emitAfter(asyncId);
}
} while (!queue.isEmpty() || processPromiseRejections());
}
资源销毁的时候也会触发对应的钩子,不过不同的是这个钩子是异步触发的。无论是JS还是好C++层触发销毁钩子的时候,逻辑都是一致的。
void AsyncWrap::EmitDestroy(Environment* env, double async_id) {
// 之前为空则设置回调
if (env->destroy_async_id_list()->empty()) {
env->SetUnrefImmediate(&DestroyAsyncIdsCallback);
}
// async id入队
env->destroy_async_id_list()->push_back(async_id);
}
template <typename Fn>
void Environment::SetUnrefImmediate(Fn&& cb) {
CreateImmediate(std::move(cb), false);
}
template <typename Fn>
void Environment::CreateImmediate(Fn&& cb, bool ref) {
auto callback = std::make_unique<NativeImmediateCallbackImpl<Fn>>(
std::move(cb), ref);
// 加入任务队列
native_immediates_.Push(std::move(callback));
}
在事件循环的check阶段就会执行里面的任务,从而执行回调DestroyAsyncIdsCallback。
void AsyncWrap::DestroyAsyncIdsCallback(Environment* env) {
Local<Function> fn = env->async_hooks_destroy_function();
do {
std::vector<double> destroy_async_id_list;
destroy_async_id_list.swap(*env->destroy_async_id_list());
// 遍历销毁的async id
for (auto async_id : destroy_async_id_list) {
HandleScope scope(env->isolate());
Local<Value> async_id_value = Number::New(env->isolate(), async_id);
// 执行JS层回调
MaybeLocal<Value> ret = fn->Call(env->context(), Undefined(env->isolate()), 1, &async_id_value);
}
} while (!env->destroy_async_id_list()->empty());
}
我们通常以以下方式使用Async hooks
const async_hooks = require('async_hooks');
async_hooks.createHook({
init(asyncId, type, triggerAsyncId) {},
before(asyncId) {},
after(asyncId) {},
destroy(asyncId) {},
promiseResolve(asyncId),
})
.enable();
async_hooks是对资源生命周期的抽象,资源就是操作对象和回调的抽象。async_hooks定义了五个生命周期钩子,当资源的状态到达某个周期节点时,async_hooks就会触发对应的钩子。下面我们看一下具体的实现。我们首先看一下createHook。
function createHook(fns) {
return new AsyncHook(fns);
}
createHook是对AsyncHook的封装
class AsyncHook {
constructor({ init, before, after, destroy, promiseResolve }) {
// 记录回调
this[init_symbol] = init;
this[before_symbol] = before;
this[after_symbol] = after;
this[destroy_symbol] = destroy;
this[promise_resolve_symbol] = promiseResolve;
}
}
AsyncHook的初始化很简单,创建一个AsyncHook对象记录回调函数。创建了AsyncHook之后,我们需要调用AsyncHook的enable函数手动开启。
class AsyncHook {
enable() {
// 获取一个AsyncHook对象数组和一个整形数组
const [hooks_array, hook_fields] = getHookArrays();
// 执行过enable了则不需要再执行
if (hooks_array.includes(this))
return this;
// 做些统计
const prev_kTotals = hook_fields[kTotals];
hook_fields[kTotals] = hook_fields[kInit] += +!!this[init_symbol];
hook_fields[kTotals] += hook_fields[kBefore] += +!!this[before_symbol];
hook_fields[kTotals] += hook_fields[kAfter] += +!!this[after_symbol];
hook_fields[kTotals] += hook_fields[kDestroy] += +!!this[destroy_symbol];
hook_fields[kTotals] +=
hook_fields[kPromiseResolve] += +!!this[promise_resolve_symbol];
// 当前对象插入数组中
hooks_array.push(this);
// 如果之前的数量是0,本次操作后大于0则开启底层的逻辑
if (prev_kTotals === 0 && hook_fields[kTotals] > 0) {
enableHooks();
}
return this;
}
}
1 hooks_array:是一个AsyncHook对象数组,主要用于记录用户创建了哪些AsyncHook对象,然后哪些AsyncHook对象里都设置了哪些钩子,在回调的时候就会遍历这个对象数组,执行里面的回调。 2 hook_fields:对应底层的async_hook_fields。 3 enableHooks:
function enableHooks() {
// 记录async_hooks的开启个数
async_hook_fields[kCheck] += 1;
}
至此,async_hooks的初始化就完成了,我们发现逻辑非常简单。下面我们看一下他是如何串起来的。下面我们以TCP模块为例。
const { createHook, executionAsyncId } = require('async_hooks');
const fs = require('fs');
const net = require('net');
createHook({
init(asyncId, type, triggerAsyncId) {
fs.writeSync(
1,
`${type}(${asyncId}): trigger: ${triggerAsyncId} execution: ${executionAsyncId()}\n`);
}
}).enable();
net.createServer((conn) => {}).listen(8080);
以上代码输出
init: type: TCPSERVERWRAP asyncId: 2 trigger id: 1 executionAsyncId(): 1 triggerAsyncId(): 0
init: type: TickObject asyncId: 3 trigger id: 2 executionAsyncId(): 1 triggerAsyncId(): 0
before: asyncId: 3 executionAsyncId(): 3 triggerAsyncId(): 2
after: asyncId: 3 executionAsyncId(): 3 triggerAsyncId(): 2
下面我们来分析具体过程。我们知道创建资源的时候会执行init回调,具体逻辑在listen函数中,在listen函数中,通过层层调用会执行new TCP新建一个对象,表示服务器。TCP是C++层导出的类,刚才我们说过,TCP会继承AsyncWrap,新建AsyncWrap对象的时候会触发init钩子,结构图如下。 对应输出
init: type: TCPSERVERWRAP asyncId: 2 trigger id: 1 executionAsyncId(): 1 triggerAsyncId(): 0
那TickObject是怎么来的呢?我们接着看listen里的另一段逻辑。
this[async_id_symbol] = getNewAsyncId(this._handle);
defaultTriggerAsyncIdScope(this[async_id_symbol],
process.nextTick,
emitListeningNT,
this);
上面的代码我们刚才已经分析过,在执行process.nextTick的时候会创建一个TickObject对象封装执行上下文和回调。
const asyncId = newAsyncId();
const triggerAsyncId = getDefaultTriggerAsyncId();
const tickObject = {
[async_id_symbol]: asyncId,
[trigger_async_id_symbol]: triggerAsyncId,
callback,
args
};
emitInit(asyncId, 'TickObject', triggerAsyncId, tickObject);
这次再次触发了init钩子,结构如下(nextTick通过getDefaultTriggerAsyncId获取的id是defaultTriggerAsyncIdScope设置的id)。 对应输出
init: type: TickObject asyncId: 3 trigger id: 2 executionAsyncId(): 1 triggerAsyncId(): 0
接着执行tick任务。
const asyncId = tock[async_id_symbol];
emitBefore(asyncId, tock[trigger_async_id_symbol]);
try {
tock.callback();
} finally {
if (destroyHooksExist())
emitDestroy(asyncId);
}
emitAfter(asyncId);
emitBefore时,结构图如下。 对应输出
before: asyncId: 3 executionAsyncId(): 3 triggerAsyncId(): 2
after: asyncId: 3 executionAsyncId(): 3 triggerAsyncId(): 2
执行完我们的JS代码后,所有入栈的上下文都会被清空,结构图如下。 如果这时候有一个连接建立会输出什么呢?当有连接建立时,会执行C++层的OnConnection。 OnConnection会创建一个新的TCP对象表示和客户端通信的对象。
MaybeLocal<Object> TCPWrap::Instantiate(Environment* env,
AsyncWrap* parent,
TCPWrap::SocketType type) {
EscapableHandleScope handle_scope(env->isolate());
AsyncHooks::DefaultTriggerAsyncIdScope trigger_scope(parent);
return handle_scope.EscapeMaybe(
constructor->NewInstance(env->context(), 1, &type_value));
}
首先定义了一个AsyncHooks::DefaultTriggerAsyncIdScope。DefaultTriggerAsyncIdScope用于设置默认default_trigger_async_id为parent的async id(值是2),执行Instantiate时会执行析构函数恢复原来状态。接着NewInstance的时候就会新建一个TCPWrap对象,从而创建一个AsyncWrap对象。然后触发init钩子,结构图如下。 对应输出
init: type: TCPWRAP asyncId: 4 trigger id: 2 executionAsyncId(): 0 triggerAsyncId(): 0
创建完对象后,通过AsyncWrap::MakeCallback回调JS层,刚才我们已经分析过AsyncWrap::MakeCallback会触发before和after钩子,触发before钩子时,结构图如下。 对应输出
before: asyncId: 2 executionAsyncId(): 2 triggerAsyncId(): 1
同样,在回调函数里执行executionAsyncId和triggerAsyncId拿到的内容是一样的。触发after后再恢复上下文,所以输出也是一样的。
after: asyncId: 2 executionAsyncId(): 2 triggerAsyncId(): 1
异步资源并不是Node.js内置的,Node.js只是提供了一套机制,业务层也可以使用。Node.js也提供了一个类给业务使用,核心代码如下。
class AsyncResource {
constructor(type, opts = {}) {
let triggerAsyncId = opts;
let requireManualDestroy = false;
if (typeof opts !== 'number') {
triggerAsyncId = opts.triggerAsyncId === undefined ?
getDefaultTriggerAsyncId() : opts.triggerAsyncId;
requireManualDestroy = !!opts.requireManualDestroy;
}
const asyncId = newAsyncId();
this[async_id_symbol] = asyncId;
this[trigger_async_id_symbol] = triggerAsyncId;
if (initHooksExist()) {
emitInit(asyncId, type, triggerAsyncId, this);
}
}
runInAsyncScope(fn, thisArg, ...args) {
const asyncId = this[async_id_symbol];
emitBefore(asyncId, this[trigger_async_id_symbol]);
const ret = thisArg === undefined ?
fn(...args) :
ReflectApply(fn, thisArg, args);
emitAfter(asyncId);
return ret;
}
emitDestroy() {
if (this[destroyedSymbol] !== undefined) {
this[destroyedSymbol].destroyed = true;
}
emitDestroy(this[async_id_symbol]);
return this;
}
asyncId() {
return this[async_id_symbol];
}
triggerAsyncId() {
return this[trigger_async_id_symbol];
}
}
使用方式如下。
const { AsyncResource, executionAsyncId,triggerAsyncId } = require('async_hooks');
const asyncResource = new AsyncResource('Demo');
asyncResource.runInAsyncScope(() => {
console.log(executionAsyncId(), triggerAsyncId())
});
runInAsyncScope中会把asyncResource的执行上下文设置为当前执行上下文,async id是2,trigger async id是1,所以在回调里执行executionAsyncId输出的是2,triggerAsyncId输出的是1。
AsyncLocalStorage是基于AsyncResource实现的一个维护异步逻辑中公共上下文的类。我们可以把他理解为Redis。我们看一下怎么使用。
const { AsyncLocalStorage } = require('async_hooks');
const asyncLocalStorage = new AsyncLocalStorage();
function logWithId(msg) {
const id = asyncLocalStorage.getStore();
console.log(`${id !== undefined ? id : '-'}:`, msg);
}
asyncLocalStorage.run(1, () => {
logWithId('start');
setImmediate(() => {
logWithId('finish');
});
});
执行上面代码会输出
1: start
1: finish
run的时候初始化公共的上下文,然后在run里执行的异步代码也可以拿得到这个公共上下文,这个在记录日志traceId时就会很有用,否则我们就需要把traceId传遍代码每个需要的地方。下面我们看一下实现。
我们先看一下创建AsyncLocalStorage的逻辑
class AsyncLocalStorage {
constructor() {
this.kResourceStore = Symbol('kResourceStore');
this.enabled = false;
}
}
创建AsyncLocalStorage的时候很简单,主要是置状态为false,并且设置kResourceStore的值为Symbol('kResourceStore')。设置为Symbol('kResourceStore')而不是‘kResourceStore‘很重要,我们后面会看到。继续看一下执行AsyncLocalStorage.run的逻辑。
run(store, callback, ...args) {
// 新建一个AsyncResource
const resource = new AsyncResource('AsyncLocalStorage', defaultAlsResourceOpts);
// 通过runInAsyncScope把resource的执行上下文设置完当前的执行上下文
return resource.emitDestroy().runInAsyncScope(() => {
this.enterWith(store);
return ReflectApply(callback, null, args);
});
}
设置完上下文之后执行runInAsyncScope的回调,回调里首先执行里enterWith。
enterWith(store) {
// 修改AsyncLocalStorage状态
this._enable();
// 获得当前执行上下文对于多资源,也就是run里创建的resource
const resource = executionAsyncResource();
// 把公共上下文挂载到对象上
resource[this.kResourceStore] = store;
}
_enable() {
if (!this.enabled) {
this.enabled = true;
ArrayPrototypePush(storageList, this);
storageHook.enable();
}
}
挂载完公共上下文后,就执行业务回调。回调里可以通过asyncLocalStorage.getStore()获得设置的公共上下文。
getStore() {
if(this.enabled) {
const resource = executionAsyncResource();
return resource[this.kResourceStore];
}
}
getStore的原理很简单,就是首先拿到当前执行上下文对应的资源,然后根据AsyncLocalStorage的kResourceStore的值从resource中拿到公共上下文。如果是同步执行getStore,那么executionAsyncResource返回的就是我们在run的时候创建的AsyncResource,但是如果是异步getStore那么怎么办呢?因为这时候executionAsyncResource返回的不再是我们创建的AsyncResource,也就拿不到他挂载的公共上下文。为了解决这个问题,Node.js对公共上下文进行了传递。
const storageList = []; // AsyncLocalStorage对象数组
const storageHook = createHook({
init(asyncId, type, triggerAsyncId, resource) {
const currentResource = executionAsyncResource();
for (let i = 0; i < storageList.length; ++i) {
storageList[i]._propagate(resource, currentResource);
}
}
});
_propagate(resource, triggerResource) {
const store = triggerResource[this.kResourceStore];
if (this.enabled) {
resource[this.kResourceStore] = store;
}
}
我们看到Node.js内部创建了一个Hooks,在每次资源创建的时候,Node.js会把当前执行上下文对应的资源中的一个或多个key(根据storageList里对象的this.kResourceStore字段)对应的值挂载到新创建的资源中。所以在asyncLocalStorage.getStore()时即使不是我们在执行run时创建的资源对象,也可以获得具体asyncLocalStorage对象所设置的资源,我们再来看一个例子。
const { AsyncLocalStorage } = require('async_hooks');
const asyncLocalStorage = new AsyncLocalStorage();
const asyncLocalStorage2 = new AsyncLocalStorage();
function logWithId(msg) {
console.log(asyncLocalStorage2.getStore());
const id = asyncLocalStorage.getStore();
console.log(`${id !== undefined ? id : '-'}:`, msg);
}
asyncLocalStorage.run(0, () => {
asyncLocalStorage2.enterWith({hello: "world"});
logWithId('start');
setImmediate(() => {
logWithId('finish');
});
});
除了通过asyncLocalStorage.run设置上下文,我们通过asyncLocalStorage2.enterWith也给对象上下文的资源对象挂载一个新属性,key是Symbol('kResourceStore'),值是{hello: "world"},然后在logWithId中输出asyncLocalStorage2.getStore()。从输出中可以看到成功从资源中获得挂载的所有上下文。
{ hello: 'world' }
0: start
{ hello: 'world' }
0: finish
我们也可以修改源码验证
Immediate {
_idleNext: null,
_idlePrev: null,
_onImmediate: [Function (anonymous)],
_argv: undefined,
_destroyed: true,
[Symbol(refed)]: null,
[Symbol(asyncId)]: 6,
[Symbol(triggerId)]: 2,
[Symbol(kResourceStore)]: 0,
[Symbol(kResourceStore)]: { hello: 'world' }
}
可以看到资源对象挂载里两个key为Symbol(kResourceStore)的属性。
const async_hooks = require('async_hooks');
const eid = async_hooks.executionAsyncId();
const tid = async_hooks.triggerAsyncId();
console.log(eid, tid);
以上代码中,输出1和0。对应的API实现如下。
// 获取当前的async id
function executionAsyncId() {
return async_id_fields[kExecutionAsyncId];
}
// 获取当前的trigger async id,即触发当前代码的async id
function triggerAsyncId() {
return async_id_fields[kTriggerAsyncId];
}
那么async_id_fields的初始化是什么呢?从env.h定义中可以看到async_idfields(async_id_fields是上层使用的名称,对应底层的async_idfields)是AliasedFloat64Array类型。
AliasedFloat64Array async_id_fields_;
AliasedFloat64Array是个类型别名。
typedef AliasedBufferBase<double, v8::Float64Array> AliasedFloat64Array;
AliasedBufferBase的构造函数如下
AliasedBufferBase(v8::Isolate* isolate, const size_t count)
: isolate_(isolate), count_(count), byte_offset_(0) {
const v8::HandleScope handle_scope(isolate_);
const size_t size_in_bytes = MultiplyWithOverflowCheck(sizeof(NativeT), count);
v8::Local<v8::ArrayBuffer> ab = v8::ArrayBuffer::New(isolate_, size_in_bytes);
// ...
}
底层是一个ArrayBuffer。
Local<ArrayBuffer> v8::ArrayBuffer::New(Isolate* isolate, size_t byte_length) {
i::Isolate* i_isolate = reinterpret_cast<i::Isolate*>(isolate);
LOG_API(i_isolate, ArrayBuffer, New);
ENTER_V8_NO_SCRIPT_NO_EXCEPTION(i_isolate);
i::MaybeHandle<i::JSArrayBuffer> result =
i_isolate->factory()->NewJSArrayBufferAndBackingStore(
byte_length, i::InitializedFlag::kZeroInitialized);
// ...
}
ArrayBuffer::New在申请内存时传入了i::InitializedFlag::kZeroInitialized。从V8定义中可以看到会初始化内存的内容为0。
// Whether the backing store memory is initialied to zero or not.
enum class InitializedFlag : uint8_t { kUninitialized, kZeroInitialized };
回到例子中,为什么输出会是1和0而不是0和0呢?答案在Node.js启动时的这段代码。
{
InternalCallbackScope callback_scope(
env.get(),
Local<Object>(),
// async id和trigger async id
{ 1, 0 },
InternalCallbackScope::kAllowEmptyResource |
InternalCallbackScope::kSkipAsyncHooks);
// 执行我们的js
LoadEnvironment(env.get());
}
InternalCallbackScope刚才已经分析过,他会把1和0设置为当前的执行上下文。然后在LoadEnvironment里执行我的JS代码时获取到的值就是1和0。那么如果我们改成以下代码会输出什么呢?
const async_hooks = require('async_hooks');
Promise.resolve().then(() => {
const eid = async_hooks.executionAsyncId();
const tid = async_hooks.triggerAsyncId();
console.log(eid, tid);
})
以上代码会输出0和。因为执行完我们的JS代码后,InternalCallbackScope就被析构了,从而恢复为0和0。