Dangerous Promise.all
Promise.all is a great way to parallelize multiple operations. It’s especially useful when we need to make parallel requests to multiple services.
However, there’s a hidden danger. We’ll see an example in this task and explore how to avoid it.
Let’s say we have a connection to a remote service, such as a database.
There’re two functions: connect() and disconnect().
When connected, we can send requests using database.query(...) – an async function which usually returns the result but also may throw an error.
Here’s a simple implementation:
let database;
function connect() {
database = {
async query(isOk) {
if (!isOk) throw new Error('Query failed');
}
};
}
function disconnect() {
database = null;
}
// intended usage:
// connect()
// ...
// database.query(true) to emulate a successful call
// database.query(false) to emulate a failed call
// ...
// disconnect()Now here’s the problem.
We wrote the code to connect and send 3 queries in parallel (all of them take different time, e.g. 100, 200 and 300ms), then disconnect:
// Helper function to call async function `fn` after `ms` milliseconds
function delay(fn, ms) {
return new Promise((resolve, reject) => {
setTimeout(() => fn().then(resolve, reject), ms);
});
}
async function run() {
connect();
try {
await Promise.all([
// these 3 parallel jobs take different time: 100, 200 and 300 ms
// we use the `delay` helper to achieve this effect
delay(() => database.query(true), 100),
delay(() => database.query(false), 200),
delay(() => database.query(false), 300)
]);
} catch(error) {
console.log('Error handled (or was it?)');
}
disconnect();
}
run();Two of these queries happen to be unsuccessful, but we’re smart enough to wrap the Promise.all call into a try..catch block.
However, this doesn’t help! This script actually leads to an uncaught error in console!
Why? How to avoid it?
The root of the problem is that Promise.all immediately rejects when one of its promises rejects, but it do nothing to cancel the other promises.
In our case, the second query fails, so Promise.all rejects, and the try...catch block catches this error.Meanwhile, other promises are not affected – they independently continue their execution. In our case, the third query throws an error of its own after a bit of time. And that error is never caught, we can see it in the console.
The problem is especially dangerous in server-side environments, such as Node.js, when an uncaught error may cause the process to crash.
How to fix it?
An ideal solution would be to cancel all unfinished queries when one of them fails. This way we avoid any potential errors.
However, the bad news is that service calls (such as database.query) are often implemented by a 3rd-party library which doesn’t support cancellation. Then there’s no way to cancel a call.
As an alternative, we can write our own wrapper function around Promise.all which adds a custom then/catch handler to each promise to track them: results are gathered and, if an error occurs, all subsequent promises are ignored.
function customPromiseAll(promises) {
return new Promise((resolve, reject) => {
const results = [];
let resultsCount = 0;
let hasError = false; // we'll set it to true upon first error
promises.forEach((promise, index) => {
promise
.then(result => {
if (hasError) return; // ignore the promise if already errored
results[index] = result;
resultsCount++;
if (resultsCount === promises.length) {
resolve(results); // when all results are ready - successs
}
})
.catch(error => {
if (hasError) return; // ignore the promise if already errored
hasError = true; // wops, error!
reject(error); // fail with rejection
});
});
});
}This approach has an issue of its own – it’s often undesirable to disconnect() when queries are still in the process.
It may be important that all queries complete, especially if some of them make important updates.
So we should wait until all promises are settled before going further with the execution and eventually disconnecting.
Here’s another implementation. It behaves similar to Promise.all – also resolves with the first error, but waits until all promises are settled.
function customPromiseAllWait(promises) {
return new Promise((resolve, reject) => {
const results = new Array(promises.length);
let settledCount = 0;
let firstError = null;
promises.forEach((promise, index) => {
Promise.resolve(promise)
.then(result => {
results[index] = result;
})
.catch(error => {
if (firstError === null) {
firstError = error;
}
})
.finally(() => {
settledCount++;
if (settledCount === promises.length) {
if (firstError !== null) {
reject(firstError);
} else {
resolve(results);
}
}
});
});
});
}Now await customPromiseAllWait(...) will stall the execution until all queries are processed.
This is a more reliable approach, as it guarantees a predictable execution flow.
Lastly, if we’d like to process all errors, we can use either use Promise.allSettled or write a wrapper around it to gathers all errors in a single AggregateError object and rejects with it.
// wait for all promises to settle
// return results if no errors
// throw AggregateError with all errors if any
function allOrAggregateError(promises) {
return Promise.allSettled(promises).then(results => {
const errors = [];
const values = [];
results.forEach((res, i) => {
if (res.status === 'fulfilled') {
values[i] = res.value;
} else {
errors.push(res.reason);
}
});
if (errors.length > 0) {
throw new AggregateError(errors, 'One or more promises failed');
}
return values;
});
}