How to synchronize a sequence of promises?

问题:

I have an array of promise objects that must be resolved in the same sequence in which they are listed in the array, i.e. we cannot attempt resolving an element till the previous one has been resolved (as method Promise.all([...]) does).

And if one element is rejected, I need the chain to reject at once, without attempting to resolve the following element.

How can I implement this, or is there an existing implementation for such sequence pattern?

function sequence(arr) {
    return new Promise(function (resolve, reject) {
        // try resolving all elements in \'arr\',
        // but strictly one after another;
    });
}

EDIT

The initial answers suggest we can only sequence results of such array elements, not their execution, because it is predefined in such example.

But then how to generate an array of promises in such a way as to avoid early execution?

Here\'s a modified example:

function sequence(nextPromise) {
    // while nextPromise() creates and returns another promise,
    // continue resolving it;
}

I wouldn\'t want to make it into a separate question, because I believe it is part of the same problem.

SOLUTION

Some answers below and discussions that followed went a bit astray, but the eventual solution that did exactly what I was looking for was implemented within spex library, as method sequence. The method can iterate through a sequence of dynamic length, and create promises as required by the business logic of your application.

Later on I turned it into a shared library for everyone to use.

回答1:

Here are some simple examples for how you sequence through an array executing each async operation serially (one after the other).

Let\'s suppose you have an array of items:

var arr = [...];

And, you want to carry out a specific async operation on each item in the array, one at a time serially such that the next operation does not start until the previous one has finished.

And, let\'s suppose you have a promise returning function for processing one of the items in the array:

Manual Iteration

function processItem(item) {
    // do async operation and process the result
    // return a promise
}

Then, you can do something like this:

function processArray(array, fn) {
    var index = 0;

    function next() {
        if (index < array.length) {
            fn(array[index++]).then(next);
        }
    }
    next();
}

processArray(arr, processItem);

Manual Iteration Returning Promise

If you wanted a promise returned from processArray() so you\'d know when it was done, you could add this to it:

function processArray(array, fn) {
    var index = 0;

    return new Promise(function(resolve, reject) {

        function next() {
            if (index < array.length) {
                fn(array[index++]).then(next, reject);
            } else {
                resolve();
            }
        }
        next();
    }
}

processArray(arr, processItem).then(function() {
    // all done here
}, function(reason) {
    // rejection happened
});

Note: this will stop the chain on the first rejection and pass that reason back to the processArray returned promise.

Iteration with .reduce()

If you wanted to do more of the work with promises, you could chain all the promises:

function processArray(array, fn) {
   return array.reduce(function(p, item) {
       return p.then(function() {
          return fn(item);
       });
   }, Promise.resolve());
}

processArray(arr, processItem).then(function(result) {
    // all done here
}, function(reason) {
    // rejection happened
});

Note: this will stop the chain on the first rejection and pass that reason back to the promise returned from processArray().

For a success scenario, the promise returned from processArray() will be resolved with the last resolved value of your fn callback. If you wanted to accumulate a list of results and resolve with that, you could collect the results in a closure array from fn and continue to return that array each time so the final resolve would be an array of results.

Iteration with .reduce() that Resolves With Array

And, since it now seems apparent that you want the final promise result to be an array of data (in order), here\'s a revision of the previous solution that produces that:

function processArray(array, fn) {
   var results = [];
   return array.reduce(function(p, item) {
       return p.then(function() {
           return fn(item).then(function(data) {
               results.push(data);
               return results;
           });
       });
   }, Promise.resolve());
}

processArray(arr, processItem).then(function(result) {
    // all done here
    // array of data here in result
}, function(reason) {
    // rejection happened
});

Working demo: http://jsfiddle.net/jfriend00/h3zaw8u8/

And a working demo that shows a rejection: http://jsfiddle.net/jfriend00/p0ffbpoc/

Iteration with .reduce() that Resolves With Array with delay

And, if you want to insert a small delay between operations:

function delay(t, v) {
    return new Promise(function(resolve) {
        setTimeout(resolve.bind(null, v), t);
    });
}

function processArrayWithDelay(array, t, fn) {
   var results = [];
   return array.reduce(function(p, item) {
       return p.then(function() {
           return fn(item).then(function(data) {
               results.push(data);
               return delay(t, results);
           });
       });
   }, Promise.resolve());
}

processArray(arr, 200, processItem).then(function(result) {
    // all done here
    // array of data here in result
}, function(reason) {
    // rejection happened
});

Iteration with Bluebird Promise Library

The Bluebird promise library has a lot of concurrency controlling features built right in. For example, to sequence iteration through an array, you can use Promise.mapSeries().

Promise.mapSeries(arr, function(item) {
    // process each individual item here, return a promise
    return processItem(item);
}).then(function(results) {
    // process final results here
}).catch(function(err) {
    // process array here
});

Or to insert a delay between iterations:

Promise.mapSeries(arr, function(item) {
    // process each individual item here, return a promise
    return processItem(item).delay(100);
}).then(function(results) {
    // process final results here
}).catch(function(err) {
    // process array here
});

Using ES7 async/await

If you\'re coding in an environment that supports async/await, you can also just use a regular for loop and then await a promise in the loop and it will cause the for loop to pause until a promise is resolved before proceeding. This will effectively sequence your async operations so the next one doesn\'t start until the previous one is done.

async function processArray(array, fn) {
    let results = [];
    for (let i = 0; i < array.length; i++) {
        let r = await fn(array[i]);
        results.push(r);
    }
    return results;    // will be resolved value of promise
}

// sample usage
processArray(arr, processItem).then(function(result) {
    // all done here
    // array of data here in result
}, function(reason) {
    // rejection happened
});

---

FYI, I think my processArray() function here is very similar to Promise.map() in the Bluebird promise library which takes an array and a promise producing function and returns a promise that resolves with an array of resolved results.

---

@vitaly-t - Here some some more detailed comments on your approach. You are welcome to whatever code seems best to you. When I first started using promises, I tended to use promises only for the simplest things they did and write a lot of the logic myself when a more advanced use of promises could do much more of it for me. You use only what you are fully comfortable with and beyond that, you\'d rather see your own code that you intimately know. That\'s probably human nature.

I will suggest that as I understood more and more of what promises can do for me, I now like to write code that uses more of the advanced features of promises and it seems perfectly natural to me and I feel like I\'m building on well tested infrastructure that has lots of useful features. I\'d only ask that you keep your mind open as you learn more and more to potentially go that direction. It\'s my opinion that it\'s a useful and productive direction to migrate as your understanding improves.

Here are some specific points of feedback on your approach:

You create promises in seven places

As a contrast in styles, my code has only two places where I explicitly create a new promise - once in the factory function and once to initialize the .reduce() loop. Everywhere else, I\'m just building on the promises already created by chaining to them or returning values within them or just returning them directly. Your code has seven unique places where you\'re creating a promise. Now, good coding isn\'t a contest to see how few places you can create a promise, but that might point out the difference in leverage the promises that are already created versus testing conditions and creating new promises.

Throw-safety is a very useful feature

Promises are throw-safe. That means that an exception thrown within a promise handler will automatically reject that promise. If you just want the exception to become a rejection, then this is a very useful feature to take advantage of. In fact, you will find that just throwing yourself is a useful way to reject from within a handler without creating yet another promise.

Lots of Promise.resolve() or Promise.reject() is probably an opportunity for simplification

If you see code with lots of Promise.resolve() or Promise.reject() statements, then there are probably opportunities to leverage the existing promises better rather than creating all these new promises.

Cast to a Promise

If you don\'t know if something returned a promise, then you can cast it to a promise. The promise library will then do it\'s own checks whether it is a promise or not and even whether it\'s the kind of promise that matches the promise library you\'re using and, if not, wrap it into one. This can save rewriting a lot of this logic yourself.

Contract to Return a Promise

In many cases these days, it\'s completely viable to have a contract for a function that may do something asynchronous to return a promise. If the function just wants to do something synchronous, then it can just return a resolved promise. You seem to feel like this is onerous, but it\'s definitely the way the wind is blowing and I already write lots of code that requires that and it feels very natural once you get familiar with promises. It abstracts away whether the operation is sync or async and the caller doesn\'t have to know or do anything special either way. This is a nice use of promises.

The factory function can be written to create one promise only

The factory function can be written to create one promise only and then resolve or reject it. This style also makes it throw safe so any exception occuring in the factory function automatically becomes a reject. It also makes the contract to always return a promise automatic.

While I realize this factory function is a placeholder function (it doesn\'t even do anything async), hopefully you can see the style to consider it:

function factory(idx) {
    // create the promise this way gives you automatic throw-safety
    return new Promise(function(resolve, reject) {
        switch (idx) {
            case 0:
                resolve(\"one\");
                break;
            case 1:
                resolve(\"two\");
                break;
            case 2:
                resolve(\"three\");
                break;
            default:
                resolve(null);
                break;
        }
    });
}

If any of these operations were async, then they could just return their own promises which would automatically chain to the one central promise like this:

function factory(idx) {
    // create the promise this way gives you automatic throw-safety
    return new Promise(function(resolve, reject) {
        switch (idx) {
            case 0:
                resolve($.ajax(...));
            case 1:
                resole($.ajax(...));
            case 2:
                resolve(\"two\");
                break;
            default:
                resolve(null);
                break;
        }
    });
}

Using a reject handler to just return promise.reject(reason) is not needed

When you have this body of code:

    return obj.then(function (data) {
        result.push(data);
        return loop(++idx, result);
    }, function (reason) {
        return promise.reject(reason);
    });

The reject handler is not adding any value. You can instead just do this:

    return obj.then(function (data) {
        result.push(data);
        return loop(++idx, result);
    });

You are already returning the result of obj.then(). If either obj rejects or if anything chained to obj or returned from then .then() handler rejects, then obj will reject. So you don\'t need to create a new promise with the reject. The simpler code without the reject handler does the same thing with less code.

---

Here\'s a version in the general architecture of your code that tries to incorporate most of these ideas:

function factory(idx) {
    // create the promise this way gives you automatic throw-safety
    return new Promise(function(resolve, reject) {
        switch (idx) {
            case 0:
                resolve(\"zero\");
                break;
            case 1:
                resolve(\"one\");
                break;
            case 2:
                resolve(\"two\");
                break;
            default:
                // stop further processing
                resolve(null);
                break;
        }
    });
}


// Sequentially resolves dynamic promises returned by a factory;
function sequence(factory) {
    function loop(idx, result) {
        return Promise.resolve(factory(idx)).then(function(val) {
            // if resolved value is not null, then store result and keep going
            if (val !== null) {
                result.push(val);
                // return promise from next call to loop() which will automatically chain
                return loop(++idx, result);
            } else {
                // if we got null, then we\'re done so return results
                return result;
            }
        });
    }
    return loop(0, []);
}

sequence(factory).then(function(results) {
    log(\"results: \", results);
}, function(reason) {
    log(\"rejected: \", reason);
});

Working demo: http://jsfiddle.net/jfriend00/h3zaw8u8/

Some comments about this implementation:

1. Promise.resolve(factory(idx)) essentially casts the result of factory(idx) to a promise. If it was just a value, then it becomes a resolved promise with that return value as the resolve value. If it was already a promise, then it just chains to that promise. So, it replaces all your type checking code on the return value of the factory() function.

2. The factory function signals that it is done by returning either null or a promise who\'s resolved value ends up being null. The above cast maps those two conditions to the same resulting code.

3. The factory function catches exceptions automatically and turns them into rejects which are then handled automatically by the sequence() function. This is one significant advantage of letting promises do a lot of your error handling if you just want to abort processing and feed the error back on the first exception or rejection.

4. The factory function in this implementation can return either a promise or a static value (for a synchronous operation) and it will work just fine (per your design request).

5. I\'ve tested it with a thrown exception in the promise callback in the factory function and it does indeed just reject and propagate that exception back to reject the sequence promise with the exception as the reason.

6. This uses a similar method as you (on purpose, trying to stay with your general architecture) for chaining multiple calls to loop().

回答2:

Promises represent values of operations and not the operations themselves. The operations are already started so you can\'t make them wait for one another.

Instead, you can synchronize functions that return promises invoking them in order (through a loop with promise chaining for instance), or using the .each method in bluebird.

回答3:

You can\'t simply run X async operations and then want them to be resolved in an order.

The correct way to do something like this is to run the new async operation only after the one before was resolved:

doSomethingAsync().then(function(){
   doSomethingAsync2().then(function(){
       doSomethingAsync3();
       .......
   });
});

Edit
Seems like you want to wait for all promises and then invoke their callbacks in a specific order. Something like this:

var callbackArr = [];
var promiseArr = [];
promiseArr.push(doSomethingAsync());
callbackArr.push(doSomethingAsyncCallback);
promiseArr.push(doSomethingAsync1());
callbackArr.push(doSomethingAsync1Callback);
.........
promiseArr.push(doSomethingAsyncN());
callbackArr.push(doSomethingAsyncNCallback);

and then:

$.when(promiseArr).done(function(promise){
    while(callbackArr.length > 0)
    {
       callbackArr.pop()(promise);
    }
});

The problems that can occur with this is when one or more promises fail.

回答4:

Although quite dense, here\'s another solution that will iterate a promise-returning function over an array of values and resolve with an array of results:

function processArray(arr, fn) {
    return arr.reduce(
        (p, v) => p.then((a) => fn(v).then(r => a.concat([r]))),
        Promise.resolve([])
    );
}

Usage:

const numbers = [0, 4, 20, 100];
const multiplyBy3 = (x) => new Promise(res => res(x * 3));

// Prints [ 0, 12, 60, 300 ]
processArray(numbers, multiplyBy3).then(console.log);

Note that, because we\'re reducing from one promise to the next, each item is processed in series.

It\'s functionally equivalent to the \"Iteration with .reduce() that Resolves With Array\" solution from @jfriend00 but a bit neater.

回答5:

I suppose two approaches for handling this question:

1. Create multiple promises and use the allWithAsync function as follow:

let allPromiseAsync = (...PromisesList) => {
return new Promise(async resolve => {
    let output = []
    for (let promise of PromisesList) {
        output.push(await promise.then(async resolvedData => await resolvedData))
        if (output.length === PromisesList.length) resolve(output)
    }
}) }
const prm1= Promise.resolve(\'first\');
const prm2= new Promise((resolve, reject) => setTimeout(resolve, 2000, \'second\'));
const prm3= Promise.resolve(\'third\');

allPromiseAsync(prm1, prm2, prm3)
    .then(resolvedData => {
        console.log(resolvedData) // [\'first\', \'second\', \'third\']
    });

1. Use the Promise.all function instead:

  (async () => {
  const promise1 = new Promise(resolve => {
    setTimeout(() => { resolve() }, 2500)
  })

  const promise2 = new Promise(resolve => {
    setTimeout(() => { resolve() }, 5000)
  })

  const promise3 = new Promise(resolve => {
    setTimeout(() => { resolve() }, 1000)
  })

  const promises = [promise1, promise2, promise3]

  await Promise.all(promises)

  console.log(\'This line is shown after 8500ms\')
})()

回答6:

In my opinion, you should be using a for loop(yes the only time I would recommend a for loop). The reason is that when you are using a for loop it allows you to await on each of the iterations of your loop where using reduce, map or forEach with run all your promise iterations concurrently. Which by the sounds of it is not what you want, you want each promise to wait until the previous promise has resolved. So to do this you would do something like the following.

const ids = [0, 1, 2]
const accounts = ids.map(id => getId(id))
const accountData = async() => {
   for await (const account of accounts) {
       // account will equal the current iteration of the loop
       // and each promise are now waiting on the previous promise to resolve! 
   }
}

// then invoke your function where ever needed
accountData()

And obviously, if you wanted to get really extreme you could do something like this:

 const accountData = async(accounts) => {
    for await (const account of accounts) {
       // do something
    }
 }

 accountData([0, 1, 2].map(id => getId(id)))

This is so much more readable than any of the other examples, it is much less code, reduced the number of lines needed for this functionality, follows a more functional programming way of doing things and is using ES7 to its full potential!!!!

Also depending on your set up or when you are reading this you may need to add the plugin-proposal-async-generator-functions polyfill or you may see the following error

@babel/plugin-proposal-async-generator-functions (https://git.io/vb4yp) to the \'plugins\' section of your Babel config to enable transformation.