ChatGPT解决这个技术问题 Extra ChatGPT

How do I await multiple promises in-parallel without 'fail-fast' behavior? [duplicate]

This question already has answers here: Wait until all promises complete even if some rejected (20 answers) Closed 3 years ago.

I'm using async/await to fire several api calls in parallel:

async function foo(arr) {
  const results = await Promise.all(arr.map(v => {
     return doAsyncThing(v)
  }))
  return results
}

I know that, unlike loops, Promise.all executes in-parallel (that is, the waiting-for-results portion is in parallel).

But I also know that:

Promise.all is rejected if one of the elements is rejected and Promise.all fails fast: If you have four promises which resolve after a timeout, and one rejects immediately, then Promise.all rejects immediately.

As I read this, if I Promise.all with 5 promises, and the first one to finish returns a reject(), then the other 4 are effectively cancelled and their promised resolve() values are lost.

Is there a third way? Where execution is effectively in-parallel, but a single failure doesn't spoil the whole bunch?

The other four are not cancelled, but their results are not propagated along the promise chain (IIUC).
If you want to avoid a specific failure mode rejecting the promise chain, then you can handle that failure in the sub-promise chain (using catch), thereby avoiding the fast-fail. Would this do what you want?
const noop = function(){} Promise.all( arr.map( v => doAsyncThing(v).catch(noop) ) ) turns an error into an undefined-value
@BenAston do you mean instead of return doAsyncThing(v), return doAsyncThing(v).catch(err => {return err}) ? ETA: I'm not sure what should go in the catch body--is there a way to pass the error object up w/o reject? So long as I can handle the reject()ed promises individually and still get the values from the resolve()d promises. re: cancelled vs not propagated, i am not certain I understand the difference. Both result in lost resolve() values, yeah?
@Thomas will the promises that were reject()ed then end up in the results array as undefined?

j
jarmod

While the technique in the accepted answer can solve your issue, it's an anti-pattern. Resolving a promise with an error isn't good practice and there is a cleaner way of doing this.

What you want to do, in pseudo-code, is:

fn task() {
  result-1 = doAsync();
  result-n = doAsync();

  // handle results together
  return handleResults(result-1, ..., result-n)
}

This can be achieved simply with async/await without the need to use Promise.all. A working example:

console.clear(); function wait(ms, data) { return new Promise( resolve => setTimeout(resolve.bind(this, data), ms) ); } /** * These will be run in series, because we call * a function and immediately wait for each result, * so this will finish in 1s. */ async function series() { return { result1: await wait(500, 'seriesTask1'), result2: await wait(500, 'seriesTask2'), } } /** * While here we call the functions first, * then wait for the result later, so * this will finish in 500ms. */ async function parallel() { const task1 = wait(500, 'parallelTask1'); const task2 = wait(500, 'parallelTask2'); return { result1: await task1, result2: await task2, } } async function taskRunner(fn, label) { const startTime = performance.now(); console.log(`Task ${label} starting...`); let result = await fn(); console.log(`Task ${label} finished in ${ Number.parseInt(performance.now() - startTime) } miliseconds with,`, result); } void taskRunner(series, 'series'); void taskRunner(parallel, 'parallel');

Note: You will need a browser which has async/await enabled to run this snippet.

This way you can use simply try/ catch to handle your errors, and return partial results inside parallel function.


Underappreciated. This is very clear exactly what is happening.
"anti-pattern: resolving a promise with an error isn't good practice" - where did you get that from? No, on the contrary, your parallel function is an antipattern that will potentially cause unhandled rejections.
And what's worse: this doesn't even solve the OPs problem. If task1 rejects before task2 fulfills, you still are failing fast.
where did you get that from? - Handling errors in then is anti-pattern because Promises has a dedicated way of dealing with errors in the catch branches so developers expect values in then and errors in catch. Mixing that up will lead to confusion for others who are not aware of your custom implementation details.
I downvoted because this answer does not make explicit the solution to the propagation of errors.
B
Ben Aston

ES2020 contains Promise.allSettled, which will do what you want.

Promise.allSettled([ Promise.resolve('a'), Promise.reject('b') ]).then(console.log)

Output:

[
  {
    "status": "fulfilled",
    "value": "a"
  },
  {
    "status": "rejected",
    "reason": "b"
  }
]

But if you want to "roll your own", then you can leverage the fact that using Promise#catch means that the promise resolves (unless you throw an exception from the catch or manually reject the promise chain), so you do not need to explicitly return a resolved promise.

So, by simply handling errors with catch, you can achieve what you want.

Note that if you want the errors to be visible in the result, you will have to decide on a convention for surfacing them.

You can apply a rejection handling function to each promise in a collection using Array#map, and use Promise.all to wait for all of them to complete.

Example

The following should print out:

Elapsed Time   Output

     0         started...
     1s        foo completed
     1s        bar completed
     2s        bam errored
     2s        done [
                   "foo result",
                   "bar result",
                   {
                       "error": "bam"
                   }
               ]

async function foo() { await new Promise((r)=>setTimeout(r,1000)) console.log('foo completed') return 'foo result' } async function bar() { await new Promise((r)=>setTimeout(r,1000)) console.log('bar completed') return 'bar result' } async function bam() { try { await new Promise((_,reject)=>setTimeout(reject,2000)) } catch { console.log('bam errored') throw 'bam' } } function handleRejection(p) { return p.catch((error)=>({ error })) } function waitForAll(...ps) { console.log('started...') return Promise.all(ps.map(handleRejection)) } waitForAll(foo(), bar(), bam()).then(results=>console.log('done', results))

See also.


return await is redundant fyi eslint.org/docs/rules/no-return-await
Why do you even use .then when you already use await? You could just remove all .then calls and it would work.
the output of your script does not include 'bar result' and 'bat result', but: done [ undefined, { "error": "bam" }, undefined ] is that intended?
Ya, I believe it should return new Promise so that we could get the correct results array: jsbin.com/ruralujame/edit?html,css,js,console,output
adding to David's comment, done [ undefined, { "error": "bam" }, undefined ] is returned. If you add return await and a then for bam, the results are displayed [Log] done – ["bar result", "bam result", "bat result"]. However I am unable to get the result done – ["bar result", {"error":"bam"}, "bat result"]. Is there a way to achieve this ?