Rust tokio 如何以异步非阻塞方式运行大量任务

tokio 官方给了一个完整的例子：手动构建 runtime ，利用 block_on 来运行多个任务。
tokio 的任务是由 tokio::spawn 之类的函数产生的 JoinHandle 类型，而且是个 Future 。

而下面利用 #[tokio::main] 和 await 编写了等价的版本（为了直观对比任务完成的实际顺序和总耗时，我对 sleep 的时间做了一些简化）：

use std::time::Instant;
use tokio::time::{sleep, Duration};

#[tokio::main]
async fn main() -> std::io::Result<()> {
    let now = Instant::now();

    let mut handles = Vec::with_capacity(10);
    for i in 0..10 {
        handles.push(tokio::spawn(my_bg_task(i)));
    }

    // Do something time-consuming while the background tasks execute.
    std::thread::sleep(Duration::from_millis(120));
    println!("Finished time-consuming task.");

    // Wait for all of them to complete.
    for handle in handles {
        handle.await?;
    }

    println!("总耗时：{} ms", now.elapsed().as_millis());
    Ok(())
}

async fn my_bg_task(i: u64) {
    let millis = 100;
    println!("Task {} sleeping for {} ms.", i, millis);
    sleep(Duration::from_millis(millis)).await;
    println!("Task {} stopping.", i);
}

输出结果：

Task 0 sleeping for 100 ms.
Task 1 sleeping for 100 ms.
Task 2 sleeping for 100 ms.
Task 3 sleeping for 100 ms.
Task 4 sleeping for 100 ms.
Task 5 sleeping for 100 ms.
Task 6 sleeping for 100 ms.
Task 7 sleeping for 100 ms.
Task 8 sleeping for 100 ms.
Task 9 sleeping for 100 ms.
Task 9 stopping.
Task 0 stopping.
Task 1 stopping.
Task 2 stopping.
Task 3 stopping.
Task 4 stopping.
Task 5 stopping.
Task 6 stopping.
Task 7 stopping.
Task 8 stopping.
Finished time-consuming task.
总耗时：120 ms

如果把主线程的的 sleep 时间改成 100 ms：std::thread::sleep(Duration::from_millis(100));
则产生下面的结果：

Task 0 sleeping for 100 ms.
Task 1 sleeping for 100 ms.
Task 2 sleeping for 100 ms.
Task 3 sleeping for 100 ms.
Task 4 sleeping for 100 ms.
Task 5 sleeping for 100 ms.
Task 6 sleeping for 100 ms.
Task 7 sleeping for 100 ms.
Task 8 sleeping for 100 ms.
Task 9 sleeping for 100 ms.
Finished time-consuming task.
Task 3 stopping.
Task 0 stopping.
Task 1 stopping.
Task 2 stopping.
Task 9 stopping.
Task 4 stopping.
Task 5 stopping.
Task 6 stopping.
Task 7 stopping.
Task 8 stopping.
总耗时：103 ms

可以看到，my_bg_task 实际是异步非阻塞执行的 👍 ：

• 异步：因为每个任务不必等待其结果就可以开始下一个任务，即；

// 异步
Task 0 sleeping for 100 ms.
Task 1 sleeping for 100 ms.
...

// 同步
Task 0 sleeping for 100 ms.
Task 0 stopping.
Task 1 sleeping for 100 ms.
Task 1 stopping.
...

• 非阻塞：每个任务之间可以快速切换，不必等待其他任务完成才切换，这个例子表现在：
  • 任务 0-9 以乱序方式 stop
  • Finished time-consuming task. 与 Task x stopping. 的打印顺序只与任务各自的运行 (sleep) 时间有关，与源代码的声明执行顺序无关。只有任务之间快速切换才能做到这一点。回顾官网的例子：10 个任务的 sleep 时间线性递减 （let millis = 1000 - 50 * i;），从 6 个任务开始小于主线程 sleep 任务的时间（750 ms），而等待 10 个任务执行的语句 for handle in handles { ... } 显然位于 std::thread::sleep 之后，所以任务之间非阻塞执行的话，打印结果为 sleep 时间越短的任务先完成，时间越长的任务后完成，总耗时为任务中的最长耗时：

Task 0 sleeping for 1000 ms.
Task 1 sleeping for 950 ms.
Task 2 sleeping for 900 ms.
Task 3 sleeping for 850 ms.
Task 4 sleeping for 800 ms.
Task 5 sleeping for 750 ms.
Task 6 sleeping for 700 ms.
Task 7 sleeping for 650 ms.
Task 8 sleeping for 600 ms.
Task 9 sleeping for 550 ms.
Task 9 stopping.
Task 8 stopping.
Task 7 stopping.
Task 6 stopping.
Finished time-consuming task.
Task 5 stopping.
Task 4 stopping.
Task 3 stopping.
Task 2 stopping.
Task 1 stopping.
Task 0 stopping.
总耗时：1001 ms // 非常完美

一般情况下，对于 async block/fn 你至少有以下一些做法：

1. 对 async block/fn 调用 .await 来等待结果；
2. 对可列举的少数 Future 调用 join! 或者 select! 来同时等待多个结果 或者 等待多个分支的第一个结果；
3. 对大量 Future 调用 join 或者 select 一类支持传入 Vec / iter 参数类型的函数，比如这个例子中的 for handle in handles { ... } 部分就可以改写成 futures::future::join_all(handles).await; ；
4. 把 async block/fn 变成任务，然后调用 Runtime::block_on （等价地，对任务 await）来执行许多任务。

容易犯的错误是，希望异步非阻塞时，对所有 async block/fn 进行了 await，而没有进行任务化处理（即 把 Future 通过 spwan 函数转化成任务）：

use std::time::Instant;
use tokio::time::{sleep, Duration};

#[tokio::main]
async fn main() {
    let now = Instant::now();

    let mut handles = Vec::with_capacity(10);
    for i in 0..10 {
        handles.push(my_bg_task(i)); // 没有把 Future 变成任务
    }

    std::thread::sleep(Duration::from_millis(120));
    println!("Finished time-consuming task.");

    for handle in handles {
        handle.await; // 而且每个 handle 必须执行完才能执行下一个 handle
    }
    println!("总耗时：{} ms", now.elapsed().as_millis());
}

async fn my_bg_task(i: u64) {
    let millis = 100;
    println!("Task {} sleeping for {} ms.", i, millis);
    sleep(Duration::from_millis(millis)).await;
    println!("Task {} stopping.", i);
}

运行结果：同步阻塞

Finished time-consuming task.
Task 0 sleeping for 100 ms.
Task 0 stopping.
Task 1 sleeping for 100 ms.
Task 1 stopping.
Task 2 sleeping for 100 ms.
Task 2 stopping.
Task 3 sleeping for 100 ms.
Task 3 stopping.
Task 4 sleeping for 100 ms.
Task 4 stopping.
Task 5 sleeping for 100 ms.
Task 5 stopping.
Task 6 sleeping for 100 ms.
Task 6 stopping.
Task 7 sleeping for 100 ms.
Task 7 stopping.
Task 8 sleeping for 100 ms.
Task 8 stopping.
Task 9 sleeping for 100 ms.
Task 9 stopping.
总耗时：1130 ms

---

或者像这样：

use std::time::Instant;
use tokio::time::{sleep, Duration};

#[tokio::main]
async fn main() {
    let now = Instant::now();

    let mut handles = Vec::with_capacity(10);
    for i in 0..10 {
        handles.push(my_bg_task(i)); // 没有把 Future 变成任务
    }

    std::thread::sleep(Duration::from_millis(120));
    println!("Finished time-consuming task.");

    futures::future::join_all(handles).await; // 但是 join_all 会等待所有 Future 并发执行完
    println!("总耗时：{} ms", now.elapsed().as_millis());
}

async fn my_bg_task(i: u64) {
    let millis = 100;
    println!("Task {} sleeping for {} ms.", i, millis);
    sleep(Duration::from_millis(millis)).await;
    println!("Task {} stopping.", i);
}

运行结果：异步阻塞

Finished time-consuming task.
Task 0 sleeping for 100 ms.
Task 1 sleeping for 100 ms.
Task 2 sleeping for 100 ms.
Task 3 sleeping for 100 ms.
Task 4 sleeping for 100 ms.
Task 5 sleeping for 100 ms.
Task 6 sleeping for 100 ms.
Task 7 sleeping for 100 ms.
Task 8 sleeping for 100 ms.
Task 9 sleeping for 100 ms.
Task 0 stopping.
Task 1 stopping.
Task 2 stopping.
Task 3 stopping.
Task 4 stopping.
Task 5 stopping.
Task 6 stopping.
Task 7 stopping.
Task 8 stopping.
Task 9 stopping.
总耗时：221 ms

​

P.S. 关于代码中 std::thread::sleep 和 tokio::time::sleep 的区别，参考这篇文章 Async: What is blocking? (by Alice Ryhl) 。