异步上下文跟踪#

稳定性:2 - 稳定

源代码: lib/async_hooks.js

介绍#

这些类用于关联状态并在回调函数和 Promise 链中传播。它们允许在 Web 请求或任何其他异步持续时间的生命周期内存储数据。这与其他语言中的线程局部存储(thread-local storage)类似。

AsyncLocalStorageAsyncResource 类是 node:async_hooks 模块的一部分。

import { AsyncLocalStorage, AsyncResource } from 'node:async_hooks';const { AsyncLocalStorage, AsyncResource } = require('node:async_hooks');

类:AsyncLocalStorage#

这个类创建的存储在异步操作中保持一致。

虽然你可以在 node:async_hooks 模块之上构建自己的实现,但应优先选择 AsyncLocalStorage,因为它是一个高性能且内存安全的实现,其中包含了许多不易实现的显著优化。

以下示例使用 AsyncLocalStorage 构建一个简单的日志记录器,它为传入的 HTTP 请求分配 ID,并将这些 ID 包含在每个请求中记录的消息里。

import http from 'node:http';
import { AsyncLocalStorage } from 'node:async_hooks';

const asyncLocalStorage = new AsyncLocalStorage();

function logWithId(msg) {
  const id = asyncLocalStorage.getStore();
  console.log(`${id !== undefined ? id : '-'}:`, msg);
}

let idSeq = 0;
http.createServer((req, res) => {
  asyncLocalStorage.run(idSeq++, () => {
    logWithId('start');
    // Imagine any chain of async operations here
    setImmediate(() => {
      logWithId('finish');
      res.end();
    });
  });
}).listen(8080);

http.get('https://:8080');
http.get('https://:8080');
// Prints:
//   0: start
//   0: finish
//   1: start
//   1: finishconst http = require('node:http');
const { AsyncLocalStorage } = require('node:async_hooks');

const asyncLocalStorage = new AsyncLocalStorage();

function logWithId(msg) {
  const id = asyncLocalStorage.getStore();
  console.log(`${id !== undefined ? id : '-'}:`, msg);
}

let idSeq = 0;
http.createServer((req, res) => {
  asyncLocalStorage.run(idSeq++, () => {
    logWithId('start');
    // Imagine any chain of async operations here
    setImmediate(() => {
      logWithId('finish');
      res.end();
    });
  });
}).listen(8080);

http.get('https://:8080');
http.get('https://:8080');
// Prints:
//   0: start
//   0: finish
//   1: start
//   1: finish

每个 AsyncLocalStorage 实例都维护一个独立的存储上下文。多个实例可以安全地同时存在,而不会有互相干扰数据的风险。

new AsyncLocalStorage([options])#

  • options <Object>
    • defaultValue <any> 当没有提供存储时使用的默认值。
    • name <string> AsyncLocalStorage 值的名称。

创建一个新的 AsyncLocalStorage 实例。存储仅在 run() 调用内部或 enterWith() 调用之后提供。

静态方法:AsyncLocalStorage.bind(fn)#

  • fn <Function> 要绑定到当前执行上下文的函数。
  • 返回:<Function> 一个新函数,它在捕获的执行上下文中调用 fn

将给定的函数绑定到当前的执行上下文。

静态方法:AsyncLocalStorage.snapshot()#

  • 返回:<Function> 一个签名为 (fn: (...args) : R, ...args) : R 的新函数。

捕获当前的执行上下文并返回一个接受函数作为参数的函数。每当调用返回的函数时,它将在捕获的上下文中调用传递给它的函数。

const asyncLocalStorage = new AsyncLocalStorage();
const runInAsyncScope = asyncLocalStorage.run(123, () => AsyncLocalStorage.snapshot());
const result = asyncLocalStorage.run(321, () => runInAsyncScope(() => asyncLocalStorage.getStore()));
console.log(result);  // returns 123 

AsyncLocalStorage.snapshot() 可以替代 AsyncResource 用于简单的异步上下文跟踪,例如:

class Foo {
  #runInAsyncScope = AsyncLocalStorage.snapshot();

  get() { return this.#runInAsyncScope(() => asyncLocalStorage.getStore()); }
}

const foo = asyncLocalStorage.run(123, () => new Foo());
console.log(asyncLocalStorage.run(321, () => foo.get())); // returns 123 

asyncLocalStorage.disable()#

稳定性:1 - 实验性

禁用 AsyncLocalStorage 实例。所有后续对 asyncLocalStorage.getStore() 的调用都将返回 undefined,直到再次调用 asyncLocalStorage.run()asyncLocalStorage.enterWith()

调用 asyncLocalStorage.disable() 时,所有与该实例关联的当前上下文都将退出。

asyncLocalStorage 可以被垃圾回收之前,必须调用 asyncLocalStorage.disable()。这不适用于 asyncLocalStorage 提供的存储,因为这些对象会与相应的异步资源一起被垃圾回收。

asyncLocalStorage 在当前进程中不再使用时,请使用此方法。

asyncLocalStorage.getStore()#

返回当前存储。如果在通过调用 asyncLocalStorage.run()asyncLocalStorage.enterWith() 初始化的异步上下文之外调用,则返回 undefined

asyncLocalStorage.enterWith(store)#

稳定性:1 - 实验性

在当前同步执行的剩余部分进入该上下文,然后在任何后续的异步调用中持久化该存储。

示例

const store = { id: 1 };
// Replaces previous store with the given store object
asyncLocalStorage.enterWith(store);
asyncLocalStorage.getStore(); // Returns the store object
someAsyncOperation(() => {
  asyncLocalStorage.getStore(); // Returns the same object
}); 

这种转换将持续整个同步执行过程。这意味着,例如,如果在一个事件处理程序中进入了上下文,那么后续的事件处理程序也将在该上下文中运行,除非使用 AsyncResource 将它们明确绑定到另一个上下文。这就是为什么应该优先使用 run() 而不是 enterWith(),除非有充分的理由使用后一种方法。

const store = { id: 1 };

emitter.on('my-event', () => {
  asyncLocalStorage.enterWith(store);
});
emitter.on('my-event', () => {
  asyncLocalStorage.getStore(); // Returns the same object
});

asyncLocalStorage.getStore(); // Returns undefined
emitter.emit('my-event');
asyncLocalStorage.getStore(); // Returns the same object 

asyncLocalStorage.name#

如果提供的话,为 AsyncLocalStorage 实例的名称。

asyncLocalStorage.run(store, callback[, ...args])#

在一个上下文中同步运行一个函数并返回其返回值。存储在回调函数之外是不可访问的。存储对于在回调函数内创建的任何异步操作都是可访问的。

可选的 args 会被传递给回调函数。

如果回调函数抛出错误,run() 也会抛出该错误。调用堆栈不会受此调用的影响,并且上下文会被退出。

示例

const store = { id: 2 };
try {
  asyncLocalStorage.run(store, () => {
    asyncLocalStorage.getStore(); // Returns the store object
    setTimeout(() => {
      asyncLocalStorage.getStore(); // Returns the store object
    }, 200);
    throw new Error();
  });
} catch (e) {
  asyncLocalStorage.getStore(); // Returns undefined
  // The error will be caught here
} 

asyncLocalStorage.exit(callback[, ...args])#

稳定性:1 - 实验性

在一个上下文之外同步运行一个函数并返回其返回值。存储在回调函数或回调函数内创建的异步操作中是不可访问的。在回调函数内进行的任何 getStore() 调用都将始终返回 undefined

可选的 args 会被传递给回调函数。

如果回调函数抛出错误,exit() 也会抛出该错误。调用堆栈不会受此调用的影响,并且上下文会被重新进入。

示例

// Within a call to run
try {
  asyncLocalStorage.getStore(); // Returns the store object or value
  asyncLocalStorage.exit(() => {
    asyncLocalStorage.getStore(); // Returns undefined
    throw new Error();
  });
} catch (e) {
  asyncLocalStorage.getStore(); // Returns the same object or value
  // The error will be caught here
} 

async/await 一起使用#

如果在一个异步函数中,只有一个 await 调用需要在某个上下文中运行,应使用以下模式:

async function fn() {
  await asyncLocalStorage.run(new Map(), () => {
    asyncLocalStorage.getStore().set('key', value);
    return foo(); // The return value of foo will be awaited
  });
} 

在这个例子中,存储仅在回调函数和被 foo 调用的函数中可用。在 run 之外调用 getStore 将返回 undefined

问题排查:上下文丢失#

在大多数情况下,AsyncLocalStorage 都能正常工作。在极少数情况下,当前存储会在某个异步操作中丢失。

如果你的代码是基于回调的,使用 util.promisify() 将其 Promise 化就足够了,这样它就能与原生 Promise 一起工作。

如果你需要使用基于回调的 API,或者你的代码假定了一种自定义的 thenable 实现,请使用 AsyncResource 类将异步操作与正确的执行上下文关联起来。通过在你怀疑导致上下文丢失的调用之后记录 asyncLocalStorage.getStore() 的内容,来找到导致上下文丢失的函数调用。当代码记录 undefined 时,最后被调用的那个回调函数很可能就是导致上下文丢失的原因。

类:AsyncResource#

AsyncResource 类被设计为由嵌入者的异步资源来扩展。通过这种方式,用户可以轻松地触发他们自己资源的生命周期事件。

当一个 AsyncResource 被实例化时,init 钩子将被触发。

以下是 AsyncResource API 的概述。

import { AsyncResource, executionAsyncId } from 'node:async_hooks';

// AsyncResource() is meant to be extended. Instantiating a
// new AsyncResource() also triggers init. If triggerAsyncId is omitted then
// async_hook.executionAsyncId() is used.
const asyncResource = new AsyncResource(
  type, { triggerAsyncId: executionAsyncId(), requireManualDestroy: false },
);

// Run a function in the execution context of the resource. This will
// * establish the context of the resource
// * trigger the AsyncHooks before callbacks
// * call the provided function `fn` with the supplied arguments
// * trigger the AsyncHooks after callbacks
// * restore the original execution context
asyncResource.runInAsyncScope(fn, thisArg, ...args);

// Call AsyncHooks destroy callbacks.
asyncResource.emitDestroy();

// Return the unique ID assigned to the AsyncResource instance.
asyncResource.asyncId();

// Return the trigger ID for the AsyncResource instance.
asyncResource.triggerAsyncId();const { AsyncResource, executionAsyncId } = require('node:async_hooks');

// AsyncResource() is meant to be extended. Instantiating a
// new AsyncResource() also triggers init. If triggerAsyncId is omitted then
// async_hook.executionAsyncId() is used.
const asyncResource = new AsyncResource(
  type, { triggerAsyncId: executionAsyncId(), requireManualDestroy: false },
);

// Run a function in the execution context of the resource. This will
// * establish the context of the resource
// * trigger the AsyncHooks before callbacks
// * call the provided function `fn` with the supplied arguments
// * trigger the AsyncHooks after callbacks
// * restore the original execution context
asyncResource.runInAsyncScope(fn, thisArg, ...args);

// Call AsyncHooks destroy callbacks.
asyncResource.emitDestroy();

// Return the unique ID assigned to the AsyncResource instance.
asyncResource.asyncId();

// Return the trigger ID for the AsyncResource instance.
asyncResource.triggerAsyncId();

new AsyncResource(type[, options])#

  • type <string> 异步事件的类型。
  • options <Object>
    • triggerAsyncId <number> 创建此异步事件的执行上下文的 ID。默认值:executionAsyncId()
    • requireManualDestroy <boolean> 如果设置为 true,则在对象被垃圾回收时禁用 emitDestroy。通常不需要设置此项(即使是手动调用 emitDestroy),除非检索了资源的 asyncId 并用它调用了敏感 API 的 emitDestroy。当设置为 false 时,只有在至少有一个活动的 destroy 钩子时,垃圾回收时的 emitDestroy 调用才会发生。默认值: false

用法示例:

class DBQuery extends AsyncResource {
  constructor(db) {
    super('DBQuery');
    this.db = db;
  }

  getInfo(query, callback) {
    this.db.get(query, (err, data) => {
      this.runInAsyncScope(callback, null, err, data);
    });
  }

  close() {
    this.db = null;
    this.emitDestroy();
  }
} 

静态方法:AsyncResource.bind(fn[, type[, thisArg]])#

  • fn <Function> 要绑定到当前执行上下文的函数。
  • type <string> 一个可选的名称,用于关联底层的 AsyncResource
  • thisArg <any>

将给定的函数绑定到当前的执行上下文。

asyncResource.bind(fn[, thisArg])#

  • fn <Function> 要绑定到当前 AsyncResource 的函数。
  • thisArg <any>

将给定的函数绑定到此 AsyncResource 的作用域中执行。

asyncResource.runInAsyncScope(fn[, thisArg, ...args])#

  • fn <Function> 在此异步资源的执行上下文中调用的函数。
  • thisArg <any> 用于函数调用的接收者。
  • ...args <any> 传递给函数的可选参数。

在异步资源的执行上下文中,使用提供的参数调用提供的函数。这将建立上下文,触发 AsyncHooks 的 before 回调,调用函数,触发 AsyncHooks 的 after 回调,然后恢复原始的执行上下文。

asyncResource.emitDestroy()#

调用所有 destroy 钩子。这应该只被调用一次。如果调用超过一次,将会抛出错误。这**必须**被手动调用。如果资源被留给垃圾回收器(GC)收集,那么 destroy 钩子将永远不会被调用。

asyncResource.asyncId()#

  • 返回:<number> 分配给该资源的唯一 asyncId

asyncResource.triggerAsyncId()#

  • 返回:<number> 与传递给 AsyncResource 构造函数的 triggerAsyncId 相同。

Worker 线程池使用 AsyncResource#

以下示例展示了如何使用 AsyncResource 类为 Worker 池正确地提供异步跟踪。其他资源池,如数据库连接池,可以遵循类似的模型。

假设任务是计算两个数字的和,使用一个名为 task_processor.js 的文件,内容如下:

import { parentPort } from 'node:worker_threads';
parentPort.on('message', (task) => {
  parentPort.postMessage(task.a + task.b);
});const { parentPort } = require('node:worker_threads');
parentPort.on('message', (task) => {
  parentPort.postMessage(task.a + task.b);
});

围绕它构建的 Worker 池可以使用以下结构:

import { AsyncResource } from 'node:async_hooks';
import { EventEmitter } from 'node:events';
import { Worker } from 'node:worker_threads';

const kTaskInfo = Symbol('kTaskInfo');
const kWorkerFreedEvent = Symbol('kWorkerFreedEvent');

class WorkerPoolTaskInfo extends AsyncResource {
  constructor(callback) {
    super('WorkerPoolTaskInfo');
    this.callback = callback;
  }

  done(err, result) {
    this.runInAsyncScope(this.callback, null, err, result);
    this.emitDestroy();  // `TaskInfo`s are used only once.
  }
}

export default class WorkerPool extends EventEmitter {
  constructor(numThreads) {
    super();
    this.numThreads = numThreads;
    this.workers = [];
    this.freeWorkers = [];
    this.tasks = [];

    for (let i = 0; i < numThreads; i++)
      this.addNewWorker();

    // Any time the kWorkerFreedEvent is emitted, dispatch
    // the next task pending in the queue, if any.
    this.on(kWorkerFreedEvent, () => {
      if (this.tasks.length > 0) {
        const { task, callback } = this.tasks.shift();
        this.runTask(task, callback);
      }
    });
  }

  addNewWorker() {
    const worker = new Worker(new URL('task_processor.js', import.meta.url));
    worker.on('message', (result) => {
      // In case of success: Call the callback that was passed to `runTask`,
      // remove the `TaskInfo` associated with the Worker, and mark it as free
      // again.
      worker[kTaskInfo].done(null, result);
      worker[kTaskInfo] = null;
      this.freeWorkers.push(worker);
      this.emit(kWorkerFreedEvent);
    });
    worker.on('error', (err) => {
      // In case of an uncaught exception: Call the callback that was passed to
      // `runTask` with the error.
      if (worker[kTaskInfo])
        worker[kTaskInfo].done(err, null);
      else
        this.emit('error', err);
      // Remove the worker from the list and start a new Worker to replace the
      // current one.
      this.workers.splice(this.workers.indexOf(worker), 1);
      this.addNewWorker();
    });
    this.workers.push(worker);
    this.freeWorkers.push(worker);
    this.emit(kWorkerFreedEvent);
  }

  runTask(task, callback) {
    if (this.freeWorkers.length === 0) {
      // No free threads, wait until a worker thread becomes free.
      this.tasks.push({ task, callback });
      return;
    }

    const worker = this.freeWorkers.pop();
    worker[kTaskInfo] = new WorkerPoolTaskInfo(callback);
    worker.postMessage(task);
  }

  close() {
    for (const worker of this.workers) worker.terminate();
  }
}const { AsyncResource } = require('node:async_hooks');
const { EventEmitter } = require('node:events');
const path = require('node:path');
const { Worker } = require('node:worker_threads');

const kTaskInfo = Symbol('kTaskInfo');
const kWorkerFreedEvent = Symbol('kWorkerFreedEvent');

class WorkerPoolTaskInfo extends AsyncResource {
  constructor(callback) {
    super('WorkerPoolTaskInfo');
    this.callback = callback;
  }

  done(err, result) {
    this.runInAsyncScope(this.callback, null, err, result);
    this.emitDestroy();  // `TaskInfo`s are used only once.
  }
}

class WorkerPool extends EventEmitter {
  constructor(numThreads) {
    super();
    this.numThreads = numThreads;
    this.workers = [];
    this.freeWorkers = [];
    this.tasks = [];

    for (let i = 0; i < numThreads; i++)
      this.addNewWorker();

    // Any time the kWorkerFreedEvent is emitted, dispatch
    // the next task pending in the queue, if any.
    this.on(kWorkerFreedEvent, () => {
      if (this.tasks.length > 0) {
        const { task, callback } = this.tasks.shift();
        this.runTask(task, callback);
      }
    });
  }

  addNewWorker() {
    const worker = new Worker(path.resolve(__dirname, 'task_processor.js'));
    worker.on('message', (result) => {
      // In case of success: Call the callback that was passed to `runTask`,
      // remove the `TaskInfo` associated with the Worker, and mark it as free
      // again.
      worker[kTaskInfo].done(null, result);
      worker[kTaskInfo] = null;
      this.freeWorkers.push(worker);
      this.emit(kWorkerFreedEvent);
    });
    worker.on('error', (err) => {
      // In case of an uncaught exception: Call the callback that was passed to
      // `runTask` with the error.
      if (worker[kTaskInfo])
        worker[kTaskInfo].done(err, null);
      else
        this.emit('error', err);
      // Remove the worker from the list and start a new Worker to replace the
      // current one.
      this.workers.splice(this.workers.indexOf(worker), 1);
      this.addNewWorker();
    });
    this.workers.push(worker);
    this.freeWorkers.push(worker);
    this.emit(kWorkerFreedEvent);
  }

  runTask(task, callback) {
    if (this.freeWorkers.length === 0) {
      // No free threads, wait until a worker thread becomes free.
      this.tasks.push({ task, callback });
      return;
    }

    const worker = this.freeWorkers.pop();
    worker[kTaskInfo] = new WorkerPoolTaskInfo(callback);
    worker.postMessage(task);
  }

  close() {
    for (const worker of this.workers) worker.terminate();
  }
}

module.exports = WorkerPool;

如果没有 WorkerPoolTaskInfo 对象添加的显式跟踪,回调似乎会与单个 Worker 对象相关联。然而,Worker 的创建与任务的创建无关,并且不提供任务何时被调度的信息。

这个池可以按如下方式使用:

import WorkerPool from './worker_pool.js';
import os from 'node:os';

const pool = new WorkerPool(os.availableParallelism());

let finished = 0;
for (let i = 0; i < 10; i++) {
  pool.runTask({ a: 42, b: 100 }, (err, result) => {
    console.log(i, err, result);
    if (++finished === 10)
      pool.close();
  });
}const WorkerPool = require('./worker_pool.js');
const os = require('node:os');

const pool = new WorkerPool(os.availableParallelism());

let finished = 0;
for (let i = 0; i < 10; i++) {
  pool.runTask({ a: 42, b: 100 }, (err, result) => {
    console.log(i, err, result);
    if (++finished === 10)
      pool.close();
  });
}

AsyncResourceEventEmitter 集成#

EventEmitter 触发的事件监听器可能在与调用 eventEmitter.on() 时活跃的执行上下文不同的上下文中运行。

以下示例展示了如何使用 AsyncResource 类将事件监听器与正确的执行上下文正确关联。同样的方法可以应用于 Stream 或类似的事件驱动类。

import { createServer } from 'node:http';
import { AsyncResource, executionAsyncId } from 'node:async_hooks';

const server = createServer((req, res) => {
  req.on('close', AsyncResource.bind(() => {
    // Execution context is bound to the current outer scope.
  }));
  req.on('close', () => {
    // Execution context is bound to the scope that caused 'close' to emit.
  });
  res.end();
}).listen(3000);const { createServer } = require('node:http');
const { AsyncResource, executionAsyncId } = require('node:async_hooks');

const server = createServer((req, res) => {
  req.on('close', AsyncResource.bind(() => {
    // Execution context is bound to the current outer scope.
  }));
  req.on('close', () => {
    // Execution context is bound to the scope that caused 'close' to emit.
  });
  res.end();
}).listen(3000);