Memory optimized promise blocking queue with concurrency control, specially designed to handle large data sets that must be consumed using streams.
Useful for rate-limiting async (or sync) operations that consume large data sets. For example, when interacting with a REST API or when doing CPU/memory intensive tasks.
If we use Bluebird.map()
for example, we are forced to load all the data in memory,
before being able to consume it - Out Of Memory Exception is right around the corner.
If we use p-queue (by the amazing sindresorhus) for example, we can utilize streams to avoid memory bloat, but we have no (easy) way to control the stream flow without hitting that Out Of Memory Exception.
The solution - a blocking queue that returns a promise that will be resolved when the added item gains an available slot in the queue, thus, allowing us to pause the stream consumption, until there is a real need to consume the next item - keeping us memory smart while maintaining concurrency level of data handling.
npm install promise-blocking-queue
Let's assume we have a very large (a couple of GBs) file called users.json
which contains a long list of users we want to add to our DB.
Also, let's assume that our DB instance it very cheap, and as such we don't want to load it too much, so we only want to handle
2 concurrent DB insert operations.
We can achieve a short scalable solution like so:
import * as JSONStream from 'JSONStream';
import * as fs from 'fs';
import * as es from 'event-stream';
import * as sleep from 'sleep-promise';
import { BlockingQueue } from 'promise-blocking-queue';
const queue = new BlockingQueue({ concurrency: 2 });
let handled = 0;
let failed = 0;
let awaitDrain: Promise<void> | undefined;
const readStream = fs.createReadStream('./users.json', { flags: 'r', encoding: 'utf-8' });
const jsonReadStream = JSONStream.parse('*');
const jsonWriteStream = JSONStream.stringify();
const writeStream = fs.createWriteStream('./results.json');
const addUserToDB = async (user) => {
try {
console.log(`adding ${user.username}`);
// Simulate long running task
await sleep((handled + 1) * 100);
console.log(`added ${user.username} #${++handled}`);
const writePaused = !jsonWriteStream.write(user.username);
if (writePaused && !awaitDrain) {
// Down stream asked to pause the writes for now
awaitDrain = new Promise((resolve) => {
jsonWriteStream.once('drain', resolve);
});
}
} catch (err) {
console.log(`failed ${++failed}`, err);
}
};
const handleUser = async (user) => {
// Wait until the down stream is ready to receive more data without increasing the memory footprint
if (awaitDrain) {
await awaitDrain;
awaitDrain = undefined;
}
return queue.enqueue(addUserToDB, user).enqueuePromise;
};
// Do not use async!
const mapper = (user, cb) => {
console.log(`streamed ${user.username}`);
handleUser(user)
.then(() => {
cb();
});
// Pause the read stream until we are ready to handle more data
return false;
};
const onReadEnd = () => {
console.log('done read streaming');
// If nothing was written, idle event will not be fired
if (queue.pendingCount === 0 && queue.activeCount === 0) {
jsonWriteStream.end();
} else {
// Wait until all work is done
queue.on('idle', () => {
jsonWriteStream.end();
});
}
};
const onWriteEnd = () => {
console.log(`done processing - ${handled} handled, ${failed} failed`);
process.exit(0);
};
jsonWriteStream
.pipe(writeStream)
.on('error', (err) => {
console.log('error wrtie streaming', err);
process.exit(1);
})
.on('end', onWriteEnd)
.on('finish', onWriteEnd);
readStream
.pipe(jsonReadStream)
.pipe(es.map(mapper))
.on('data', () => {
// Do nothing
})
.on('error', (err) => {
console.log('error read streaming', err);
process.exit(1);
})
.on('finish', onReadEnd)
.on('end', onReadEnd);
If users.json
is like:
[
{
"username": "a"
},
{
"username": "b"
},
{
"username": "c"
},
{
"username": "d"
}
]
Output will be:
streamed a
adding a
streamed b
adding b
streamed c // c now waits in line to start and streaming is paused until then
added a #1
adding c // c only gets handled after a is done
streamed d // d only get streamed after c has a spot in the queue
added b #2
adding d // d only gets handled after b is done
done read streaming
added c #3
added d #4
done processing - 4 handled, 0 failed
results.json
will be:
[
"a"
,
"b"
,
"c"
,
"d"
]
Returns a new queue
instance, which is an EventEmitter
subclass.
Type: object
Type: number
Default: Infinity
Minimum: 1
Concurrency limit.
BlockingQueue
instance.
Adds a sync or async task to the queue
Type: object
Type: Promise<void>
A promise that will be resolved when the queue has an available slot to run the task.
Used to realize that it is a good time to add another task to the queue.
Type: Promise<T>
A promise that will be resolved with the result of fn
.
Type: boolean
Indicates if the task has already started to run
Type: Function
Promise/Value returning function.
Type: any[]
The arguments to pass to the function
The number of promises that are currently running.
The number of promises that are waiting to run.
Emitted when the queue becomes empty. Useful if, for example, you add additional items at a later time.
Emitted when the queue becomes empty, and all promises have completed: queue.activeCount === 0 && queue.pendingCount === 0
.
The difference with empty
is that idle
guarantees that all work from the queue has finished.
empty
merely signals that the queue is empty, but it could mean that some promises haven't completed yet.
The library is based on p-limit and p-queue (by the amazing sindresorhus)
Promise Blocking Queue supports Node 6 LTS and higher.
All contributions are happily welcomed!
Please make all pull requests to the master
branch from your fork and ensure tests pass locally.