As far as I understand, in ES7/ES2016 putting multiple await's in code will work similar to chaining .then() with promises, meaning that they will execute one after the other rather than in parallel. So, for example, we have this code:
await someCall();
await anotherCall();
Do I understand it correctly that anotherCall() will be called only when someCall() is completed? What is the most elegant way of calling them in parallel?
I want to use it in Node, so maybe there's a solution with async library?
EDIT: I'm not satisfied with the solution provided in this question: Slowdown due to non-parallel awaiting of promises in async generators, because it uses generators and I'm asking about a more general use case.
You can await on Promise.all():
await Promise.all([someCall(), anotherCall()]);
To store the results:
let [someResult, anotherResult] = await Promise.all([someCall(), anotherCall()]);
Note that Promise.all fails fast, which means that as soon as one of the promises supplied to it rejects, then the entire thing rejects.
const happy = (v, ms) => new Promise((resolve) => setTimeout(() => resolve(v), ms))
const sad = (v, ms) => new Promise((_, reject) => setTimeout(() => reject(v), ms))
Promise.all([happy('happy', 100), sad('sad', 50)])
.then(console.log).catch(console.log) // 'sad'
If, instead, you want to wait for all the promises to either fulfill or reject, then you can use Promise.allSettled. Note that Internet Explorer does not natively support this method.
const happy = (v, ms) => new Promise((resolve) => setTimeout(() => resolve(v), ms))
const sad = (v, ms) => new Promise((_, reject) => setTimeout(() => reject(v), ms))
Promise.allSettled([happy('happy', 100), sad('sad', 50)])
.then(console.log) // [{ "status":"fulfilled", "value":"happy" }, { "status":"rejected", "reason":"sad" }]
Note: If you use Promise.all actions that managed to finish before rejection happen are not rolled back, so you may need to take care of such situation. For example
if you have 5 actions, 4 quick, 1 slow and slow rejects. Those 4
actions may be already executed so you may need to roll back. In such situation consider using Promise.allSettled while it will provide exact detail which action failed and which not.
TL;DR
Use Promise.all for the parallel function calls, the answer behaviors not correctly when the error occurs.
First, execute all the asynchronous calls at once and obtain all the Promise objects. Second, use await on the Promise objects. This way, while you wait for the first Promise to resolve the other asynchronous calls are still progressing. Overall, you will only wait for as long as the slowest asynchronous call. For example:
// Begin first call and store promise without waiting
const someResult = someCall();
// Begin second call and store promise without waiting
const anotherResult = anotherCall();
// Now we await for both results, whose async processes have already been started
const finalResult = [await someResult, await anotherResult];
// At this point all calls have been resolved
// Now when accessing someResult| anotherResult,
// you will have a value instead of a promise
JSbin example: http://jsbin.com/xerifanima/edit?js,console
Caveat: It doesn't matter if the await calls are on the same line or on different lines, so long as the first await call happens after all of the asynchronous calls. See JohnnyHK's comment.
Update: this answer has a different timing in error handling according to the #bergi's answer, it does NOT throw out the error as the error occurs but after all the promises are executed.
I compare the result with #jonny's tip: [result1, result2] = Promise.all([async1(), async2()]), check the following code snippet
const correctAsync500ms = () => {
return new Promise(resolve => {
setTimeout(resolve, 500, 'correct500msResult');
});
};
const correctAsync100ms = () => {
return new Promise(resolve => {
setTimeout(resolve, 100, 'correct100msResult');
});
};
const rejectAsync100ms = () => {
return new Promise((resolve, reject) => {
setTimeout(reject, 100, 'reject100msError');
});
};
const asyncInArray = async (fun1, fun2) => {
const label = 'test async functions in array';
try {
console.time(label);
const p1 = fun1();
const p2 = fun2();
const result = [await p1, await p2];
console.timeEnd(label);
} catch (e) {
console.error('error is', e);
console.timeEnd(label);
}
};
const asyncInPromiseAll = async (fun1, fun2) => {
const label = 'test async functions with Promise.all';
try {
console.time(label);
let [value1, value2] = await Promise.all([fun1(), fun2()]);
console.timeEnd(label);
} catch (e) {
console.error('error is', e);
console.timeEnd(label);
}
};
(async () => {
console.group('async functions without error');
console.log('async functions without error: start')
await asyncInArray(correctAsync500ms, correctAsync100ms);
await asyncInPromiseAll(correctAsync500ms, correctAsync100ms);
console.groupEnd();
console.group('async functions with error');
console.log('async functions with error: start')
await asyncInArray(correctAsync500ms, rejectAsync100ms);
await asyncInPromiseAll(correctAsync500ms, rejectAsync100ms);
console.groupEnd();
})();
Update:
The original answer makes it difficult (and in some cases impossible) to correctly handle promise rejections. The correct solution is to use Promise.all:
const [someResult, anotherResult] = await Promise.all([someCall(), anotherCall()]);
Original answer:
Just make sure you call both functions before you await either one:
// Call both functions
const somePromise = someCall();
const anotherPromise = anotherCall();
// Await both promises
const someResult = await somePromise;
const anotherResult = await anotherPromise;
There is another way without Promise.all() to do it in parallel:
First, we have 2 functions to print numbers:
function printNumber1() {
return new Promise((resolve,reject) => {
setTimeout(() => {
console.log("Number1 is done");
resolve(10);
},1000);
});
}
function printNumber2() {
return new Promise((resolve,reject) => {
setTimeout(() => {
console.log("Number2 is done");
resolve(20);
},500);
});
}
This is sequential:
async function oneByOne() {
const number1 = await printNumber1();
const number2 = await printNumber2();
}
//Output: Number1 is done, Number2 is done
This is parallel:
async function inParallel() {
const promise1 = printNumber1();
const promise2 = printNumber2();
const number1 = await promise1;
const number2 = await promise2;
}
//Output: Number2 is done, Number1 is done
I've created a gist testing some different ways of resolving promises, with results. It may be helpful to see the options that work.
Edit: Gist content as per Jin Lee's comment
// Simple gist to test parallel promise resolution when using async / await
function promiseWait(time) {
return new Promise((resolve, reject) => {
setTimeout(() => {
resolve(true);
}, time);
});
}
async function test() {
return [
await promiseWait(1000),
await promiseWait(5000),
await promiseWait(9000),
await promiseWait(3000),
]
}
async function test2() {
return {
'aa': await promiseWait(1000),
'bb': await promiseWait(5000),
'cc': await promiseWait(9000),
'dd': await promiseWait(3000),
}
}
async function test3() {
return await {
'aa': promiseWait(1000),
'bb': promiseWait(5000),
'cc': promiseWait(9000),
'dd': promiseWait(3000),
}
}
async function test4() {
const p1 = promiseWait(1000);
const p2 = promiseWait(5000);
const p3 = promiseWait(9000);
const p4 = promiseWait(3000);
return {
'aa': await p1,
'bb': await p2,
'cc': await p3,
'dd': await p4,
};
}
async function test5() {
return await Promise.all([
await promiseWait(1000),
await promiseWait(5000),
await promiseWait(9000),
await promiseWait(3000),
]);
}
async function test6() {
return await Promise.all([
promiseWait(1000),
promiseWait(5000),
promiseWait(9000),
promiseWait(3000),
]);
}
async function test7() {
const p1 = promiseWait(1000);
const p2 = promiseWait(5000);
const p3 = promiseWait(9000);
return {
'aa': await p1,
'bb': await p2,
'cc': await p3,
'dd': await promiseWait(3000),
};
}
let start = Date.now();
test().then((res) => {
console.log('Test Done, elapsed', (Date.now() - start) / 1000, res);
start = Date.now();
test2().then((res) => {
console.log('Test2 Done, elapsed', (Date.now() - start) / 1000, res);
start = Date.now();
test3().then((res) => {
console.log('Test3 Done, elapsed', (Date.now() - start) / 1000, res);
start = Date.now();
test4().then((res) => {
console.log('Test4 Done, elapsed', (Date.now() - start) / 1000, res);
start = Date.now();
test5().then((res) => {
console.log('Test5 Done, elapsed', (Date.now() - start) / 1000, res);
start = Date.now();
test6().then((res) => {
console.log('Test6 Done, elapsed', (Date.now() - start) / 1000, res);
});
start = Date.now();
test7().then((res) => {
console.log('Test7 Done, elapsed', (Date.now() - start) / 1000, res);
});
});
});
});
});
});
/*
Test Done, elapsed 18.006 [ true, true, true, true ]
Test2 Done, elapsed 18.009 { aa: true, bb: true, cc: true, dd: true }
Test3 Done, elapsed 0 { aa: Promise { <pending> },
bb: Promise { <pending> },
cc: Promise { <pending> },
dd: Promise { <pending> } }
Test4 Done, elapsed 9 { aa: true, bb: true, cc: true, dd: true }
Test5 Done, elapsed 18.008 [ true, true, true, true ]
Test6 Done, elapsed 9.003 [ true, true, true, true ]
Test7 Done, elapsed 12.007 { aa: true, bb: true, cc: true, dd: true }
*/
In my case, I have several tasks I want to execute in parallel, but I need to do something different with the result of those tasks.
function wait(ms, data) {
console.log('Starting task:', data, ms);
return new Promise(resolve => setTimeout(resolve, ms, data));
}
var tasks = [
async () => {
var result = await wait(1000, 'moose');
// do something with result
console.log(result);
},
async () => {
var result = await wait(500, 'taco');
// do something with result
console.log(result);
},
async () => {
var result = await wait(5000, 'burp');
// do something with result
console.log(result);
}
]
await Promise.all(tasks.map(p => p()));
console.log('done');
And the output:
Starting task: moose 1000
Starting task: taco 500
Starting task: burp 5000
taco
moose
burp
done
(async function(){
function wait(ms, data) {
console.log('Starting task:', data, ms);
return new Promise(resolve => setTimeout(resolve, ms, data));
}
var tasks = [
async () => {
var result = await wait(1000, 'moose');
// do something with result
console.log(result);
},
async () => {
var result = await wait(500, 'taco');
// do something with result
console.log(result);
},
async () => {
var result = await wait(5000, 'burp');
// do something with result
console.log(result);
}
]
await Promise.all(tasks.map(p => p()));
console.log('done');
})();
await Promise.all([someCall(), anotherCall()]); as already mention will act as a thread fence (very common in parallel code as CUDA), hence it will allow all the promises in it to run without blocking each other, but will prevent the execution to continue until ALL are resolved.
another approach that is worth to share is the Node.js async that will also allow you to easily control the amount of concurrency that is usually desirable if the task is directly linked to the use of limited resources as API call, I/O operations, etc.
// create a queue object with concurrency 2
var q = async.queue(function(task, callback) {
console.log('Hello ' + task.name);
callback();
}, 2);
// assign a callback
q.drain = function() {
console.log('All items have been processed');
};
// add some items to the queue
q.push({name: 'foo'}, function(err) {
console.log('Finished processing foo');
});
q.push({name: 'bar'}, function (err) {
console.log('Finished processing bar');
});
// add some items to the queue (batch-wise)
q.push([{name: 'baz'},{name: 'bay'},{name: 'bax'}], function(err) {
console.log('Finished processing item');
});
// add some items to the front of the queue
q.unshift({name: 'bar'}, function (err) {
console.log('Finished processing bar');
});
Credits to the Medium article autor (read more)
You can call multiple asynchronous functions without awaiting them. This will execute them in parallel. While doing so, save the returned promises in variables, and await them at some point either individually or using Promise.all() and process the results.
You can also wrap the function calls with try...catch to handle failures of individual asynchronous actions and provide fallback logic.
Here's an example:
Observe the logs, the logs printed at the beginning of execution of the individual asynchronous functions get printed immediately even though the first function takes 5 seconds to resolve.
function someLongFunc () {
return new Promise((resolve, reject)=> {
console.log('Executing function 1')
setTimeout(resolve, 5000)
})
}
function anotherLongFunc () {
return new Promise((resolve, reject)=> {
console.log('Executing function 2')
setTimeout(resolve, 5000)
})
}
async function main () {
let someLongFuncPromise, anotherLongFuncPromise
const start = Date.now()
try {
someLongFuncPromise = someLongFunc()
}
catch (ex) {
console.error('something went wrong during func 1')
}
try {
anotherLongFuncPromise = anotherLongFunc()
}
catch (ex) {
console.error('something went wrong during func 2')
}
await someLongFuncPromise
await anotherLongFuncPromise
const totalTime = Date.now() - start
console.log('Execution completed in ', totalTime)
}
main()
// A generic test function that can be configured
// with an arbitrary delay and to either resolve or reject
const test = (delay, resolveSuccessfully) => new Promise((resolve, reject) => setTimeout(() => {
console.log(`Done ${ delay }`);
resolveSuccessfully ? resolve(`Resolved ${ delay }`) : reject(`Reject ${ delay }`)
}, delay));
// Our async handler function
const handler = async () => {
// Promise 1 runs first, but resolves last
const p1 = test(10000, true);
// Promise 2 run second, and also resolves
const p2 = test(5000, true);
// Promise 3 runs last, but completes first (with a rejection)
// Note the catch to trap the error immediately
const p3 = test(1000, false).catch(e => console.log(e));
// Await all in parallel
const r = await Promise.all([p1, p2, p3]);
// Display the results
console.log(r);
};
// Run the handler
handler();
/*
Done 1000
Reject 1000
Done 5000
Done 10000
*/
Whilst setting p1, p2 and p3 is not strictly running them in parallel, they do not hold up any execution and you can trap contextual errors with a catch.
This can be accomplished with Promise.allSettled(), which is similar to Promise.all() but without the fail-fast behavior.
async function Promise1() {
throw "Failure!";
}
async function Promise2() {
return "Success!";
}
const [Promise1Result, Promise2Result] = await Promise.allSettled([Promise1(), Promise2()]);
console.log(Promise1Result); // {status: "rejected", reason: "Failure!"}
console.log(Promise2Result); // {status: "fulfilled", value: "Success!"}
Note: This is a bleeding edge feature with limited browser support, so I strongly recommend including a polyfill for this function.
I create a helper function waitAll, may be it can make it sweeter.
It only works in nodejs for now, not in browser chrome.
//const parallel = async (...items) => {
const waitAll = async (...items) => {
//this function does start execution the functions
//the execution has been started before running this code here
//instead it collects of the result of execution of the functions
const temp = [];
for (const item of items) {
//this is not
//temp.push(await item())
//it does wait for the result in series (not in parallel), but
//it doesn't affect the parallel execution of those functions
//because they haven started earlier
temp.push(await item);
}
return temp;
};
//the async functions are executed in parallel before passed
//in the waitAll function
//const finalResult = await waitAll(someResult(), anotherResult());
//const finalResult = await parallel(someResult(), anotherResult());
//or
const [result1, result2] = await waitAll(someResult(), anotherResult());
//const [result1, result2] = await parallel(someResult(), anotherResult());
I vote for:
await Promise.all([someCall(), anotherCall()]);
Be aware of the moment you call functions, it may cause unexpected result:
// Supposing anotherCall() will trigger a request to create a new User
if (callFirst) {
await someCall();
} else {
await Promise.all([someCall(), anotherCall()]); // --> create new User here
}
But following always triggers request to create new User
// Supposing anotherCall() will trigger a request to create a new User
const someResult = someCall();
const anotherResult = anotherCall(); // ->> This always creates new User
if (callFirst) {
await someCall();
} else {
const finalResult = [await someResult, await anotherResult]
}
This question already has answers here:
Waiting for more than one concurrent await operation
(4 answers)
Closed 4 years ago.
How do I retrieve the result of a promise at a later time? In a test, I am retrieving an email before sending further requests:
const email = await get_email();
assert.equal(email.subject, 'foobar');
await send_request1();
await send_request2();
How can I send the requests while the slow email retrieval is going on?
At first, I considered awaiting the email later:
// This code is wrong - do not copy!
const email_promise = get_email();
await send_request1();
await send_request2();
const email = await email_promise;
assert.equal(email.subject, 'foobar');
This works if get_email() is successful, but fails if get_email() fails before the corresponding await, with a completely justified UnhandledPromiseRejectionWarning.
Of course, I could use Promise.all, like this:
await Promise.all([
async () => {
const email = await get_email();
assert.equal(email.subject, 'foobar');
},
async () => {
await send_request1();
await send_request2();
},
]);
However, it makes the code much harder to read (it looks more like callback-based programming), especially if later requests actually depend on the email, or there is some nesting going on. Is it possible to store the result/exception of a promise and await it at a later time?
If need be, here is a testcase with mocks that sometimes fail and sometimes work, with random timings. It must never output UnhandledPromiseRejectionWarning.
const wait = (ms) => new Promise(resolve => setTimeout(resolve, ms));
const send_request1 = () => wait(300), send_request2 = () => wait(200);
async function get_email() {
await wait(Math.random() * 1000);
if (Math.random() > 0.5) throw new Error('failure');
return {subject: 'foobar'};
}
const assert = require('assert');
async function main() {
// catch possible error
const email_promise = get_email().catch(e => e);
await send_request1();
await send_request2();
// wait for result
const email = await email_promise;
// rethrow eventual error or do whatever you want with it
if(email instanceof Error) {
throw email;
}
assert.equal(email.subject, 'foobar');
};
(async () => {
try {
await main();
} catch(e) {
console.log('main error: ' + e.stack);
}
})();
In case it's guaranteed that promise rejection will be handled later, a promise can be chained with dummy catch to suppress the detection of unhandled rejection:
try {
const email_promise = get_email();
email_promise.catch(() => {}); // a hack
await send_request1();
await send_request2();
const email = await email_promise;
assert.equal(email.subject, 'foobar');
} catch (err) {...}
The problem with this approach is that there are two concurrent routines but the code doesn't express this, this is a workaround for what is usually done with Promise.all. The only reason why this workaround is feasible is that there are only 2 routines, and one of them (get_email) requires to be chained with then/await only once, so a part of it (assert) can be postponed. The problem would be more obvious if there were 3 or more routines, or routines involved multiple then/await.
In case Promise.all introduces unwanted level of lambda nesting, this can be avoided by writing routines as named functions, even if they aren't reused anywhere else:
async function assertEmail() {
const email = await get_email();
assert.equal(email.subject, 'foobar');
}
async function sendRequests() {
await send_request1();
await send_request2();
}
...
try {
await Promise.all([assertEmail(), sendRequests()]);
} catch (err) {...}
This results in clean control flow and verbose but more intelligible and testable code.
So, I want to explain why we behave this way in Node.js:
// Your "incorrect code" from before
const email_promise = get_email(); // we acquire the promise here
await send_request1(); // if this throws - we're left with a mess
await send_request2(); // if this throws - we're left with a mess
const email = await email_promise;
assert.equal(email.subject, 'foobar');
That is, the reason we behave this way is to not deal with the "multiple rejections and possibly no cleanup" scenario. I'm not sure how you ended up with the long code for Promise.all but this:
await Promise.all([
async () => {
const email = await get_email();
assert.equal(email.subject, 'foobar');
},
async () => {
await send_request1();
await send_request2();
},
]);
Can actually be this:
let [email, requestData] = await Promise.all([
get_email(),
send_request1().then(send_request2)
]);
// do whatever with email here
It's probably what I would do.
I'm with Node.js and TypeScript and I'm using async/await.
This is my test case:
async function doSomethingInSeries() {
const res1 = await callApi();
const res2 = await persistInDB(res1);
const res3 = await doHeavyComputation(res1);
return 'simle';
}
I'd like to set a timeout for the overall function. I.e. if res1 takes 2 seconds, res2 takes 0.5 seconds, res3 takes 5 seconds I'd like to have a timeout that after 3 seconds let me throw an error.
With a normal setTimeout call is a problem because the scope is lost:
async function doSomethingInSeries() {
const timerId = setTimeout(function() {
throw new Error('timeout');
});
const res1 = await callApi();
const res2 = await persistInDB(res1);
const res3 = await doHeavyComputation(res1);
clearTimeout(timerId);
return 'simle';
}
And I cannot catch it with normal Promise.catch:
doSomethingInSeries().catch(function(err) {
// errors in res1, res2, res3 will be catched here
// but the setTimeout thing is not!!
});
Any ideas on how to resolve?
You can use Promise.race to make a timeout:
Promise.race([
doSomethingInSeries(),
new Promise((_, reject) => setTimeout(() => reject(new Error('timeout')), 11.5e3))
]).catch(function(err) {
// errors in res1, res2, res3 and the timeout will be caught here
})
You cannot use setTimeout without wrapping it in a promise.
Ok I found this way:
async function _doSomethingInSeries() {
const res1 = await callApi();
const res2 = await persistInDB(res1);
const res3 = await doHeavyComputation(res1);
return 'simle';
}
async function doSomethingInSeries(): Promise<any> {
let timeoutId;
const delay = new Promise(function(resolve, reject){
timeoutId = setTimeout(function(){
reject(new Error('timeout'));
}, 1000);
});
// overall timeout
return Promise.race([delay, _doSomethingInSeries()])
.then( (res) => {
clearTimeout(timeoutId);
return res;
});
}
Anyone errors?
The things that smells a bit to me is that using Promises as asynchronicity strategy will send us to allocate too many object that some other strategy needs but this is off-topic.
Problem with #Bergi answer that doSomethingInSeries continues executing even if you already rejected the promise. It is much better to cancel it.
LATEST ANSWER
You can try use AbortController for that. Check the old answer to see how to use it - api is similar.
Keep in mind that task is not cancelled immediately, so continuation (awaiting, then or catch) is not called exactly after timeout.
To guarantee that you can combine this and #Bergi approach.
OLD ANSWER
This is how it should look like:
async const doSomethingInSeries = (cancellationToken) => {
cancellationToken.throwIfCancelled();
const res1 = await callApi();
cancellationToken.throwIfCancelled();
const res2 = await persistInDB(res1);
cancellationToken.throwIfCancelled();
const res3 = await doHeavyComputation(res1);
cancellationToken.throwIfCancelled();
return 'simle';
}
Here is simple implementation:
const makeCancellationToken = (tag) => {
let cancelled = false;
return {
isCancelled: () => cancelled,
cancel: () => {
cancelled = true;
},
throwIfCancelled: () => {
if (cancelled) {
const error = new Error(`${tag ?? 'Task'} cancelled`);
error.cancelled = true;
throw error;
}
}
}
}
And finally usage:
const cancellationToken = makeCancellationToken('doSomething')
setTimeout(cancellationToken.cancel, 5000);
try {
await doSomethingInSeries(cancellationToken);
} catch (error) {
if (error.cancelled) {
// handle cancellation
}
}
Keep in mind that task is not cancelled immediately, so continuation (awaiting, then or catch) is not called exactly after 5 secs.
To guarantee that you can combine this and #Bergi approach.