Welcome to OStack Knowledge Sharing Community for programmer and developer-Open, Learning and Share
Welcome To Ask or Share your Answers For Others

Categories

0 votes
1.1k views
in Technique[技术] by (71.8m points)

node.js - Possible to make an event handler wait until async / Promise-based code is done?

I am using the excellent Papa Parse library in nodejs mode, to stream a large (500 MB) CSV file of over 1 million rows, into a slow persistence API, that can only take one request at a time. The persistence API is based on Promises, but from Papa Parse, I receive each parsed CSV row in a synchronous event like so: parseStream.on("data", row => { ... }

The challenge I am facing is that Papa Parse dumps its CSV rows from the stream so fast that my slow persistence API can't keep up. Because Papa is synchronous and my API is Promise-based, I can't just call await doDirtyWork(row) in the on event handler, because sync and async code doesn't mix.

Or can they mix and I just don't know how?

My question is, can I make Papa's event handler wait for my API call to finish? Kind of doing the persistence API request directly in the on("data") event, making the on() function linger around somehow until the dirty API work is done?

The solution I have so far is not much better than using Papa's non-streaming mode, in terms of memory footprint. I actually need to queue up the torrent of on("data") events, in form of generator function iterations. I could have also queued up promise factories in an array and work it off in a loop. Any which way, I end up saving almost the entire CSV file as huge collection of future Promises (promise factories) in memory, until my slow API calls have worked all the way through.

async importCSV(filePath) {
    let parsedNum = 0, processedNum = 0;

    async function* gen() {
        let pf = yield;
        do {
            pf = yield await pf();
        } while (typeof pf === "function");
    };

    var g = gen();
    g.next();


    await new Promise((resolve, reject) => {
        try {
            const dataStream = fs.createReadStream(filePath);
            const parseStream = Papa.parse(Papa.NODE_STREAM_INPUT, {delimiter: ",", header: false});
            dataStream.pipe(parseStream);

            parseStream.on("data", row => {

                // Received a CSV row from Papa.parse()

                try {
                    console.log("PA#", parsedNum, ": parsed", row.filter((e, i) => i <= 2 ? e : undefined)
                    );
                    parsedNum++;

                    // Simulate some really slow async/await dirty work here, for example 
                    // send requests to a one-at-a-time persistence API

                    g.next(() => {  // don't execute now, call in sequence via the generator above
                        return new Promise((res, rej) => {
                            console.log(
                                "DW#", processedNum, ": dirty work START",
                                row.filter((e, i) => i <= 2 ? e : undefined)
                            );
                            setTimeout(() => {
                                console.log(
                                    "DW#", processedNum, ": dirty work STOP ",
                                    row.filter((e, i) => i <= 2 ? e : undefined)
                                );
                                processedNum++;
                                res();
                            }, 1000)
                        })
                    
                    });
                } catch (err) {
                    console.log(err.stack);
                    reject(err);                    
                }
            });
            parseStream.on("finish", () => {
                console.log(`Parsed ${parsedNum} rows`);
                resolve();
            });

        } catch (err) {
            console.log(err.stack);
            reject(err);                    
        }
    });
    while(!(await g.next()).done);
}

So why the rush Papa? Why not allow me to work down the file a bit slower -- the data in the original CSV file isn't gonna run away, we have hours to finish the streaming, why hammer me with on("data") events that I can't seem to slow down?

So what I really need is for Papa to become more of a grandpa, and minimize or eliminate any queuing or buffering of CSV rows. Ideally I would be able to completely sync Papa's parsing events with the speed (or lack thereof) of my API. So if it weren't for the dogma that async code can't make sync code "sleep", I would ideally send each CSV row to the API inside the Papa event, and only then return control to Papa.

Suggestions? Some kind of "loose coupling" of the event handler with the slowness of my async API is fine too. I don't mind if a few hundred rows get queued up. But when tens of thousands pile up, I will run out of heap fast.

See Question&Answers more detail:os

与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
Welcome To Ask or Share your Answers For Others

1 Answer

0 votes
by (71.8m points)

Why hammer me with on("data") events that I can't seem to slow down?

You can, you just were not asking papa to stop. You can do this by calling stream.pause(), then later stream.resume() to make use of Node stream's builtin back-pressure.

However, there's a much nicer API to use than dealing with this on your own in callback-based code: use the stream as an async iterator! When you await in the body of a for await loop, the generator has to pause as well. So you can write

async importCSV(filePath) {
    let parsedNum = 0;

    const dataStream = fs.createReadStream(filePath);
    const parseStream = Papa.parse(Papa.NODE_STREAM_INPUT, {delimiter: ",", header: false});
    dataStream.pipe(parseStream);

    for await (const row of parseStream) {
        // Received a CSV row from Papa.parse()
        const data = row.filter((e, i) => i <= 2 ? e : undefined);
        console.log("PA#", parsedNum, ": parsed", data);
        parsedNum++;
        await dirtyWork(data);
    }
    console.log(`Parsed ${parsedNum} rows`);
}

importCSV('sample.csv').catch(console.error);

let processedNum = 0;
function dirtyWork(data) {
    // Simulate some really slow async/await dirty work here,
    // for example send requests to a one-at-a-time persistence API
    return new Promise((res, rej) => {
        console.log("DW#", processedNum, ": dirty work START", data)
        setTimeout(() => {
             console.log("DW#", processedNum, ": dirty work STOP ", data);
             processedNum++;
             res();
        }, 1000);
    });
}

与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…
Welcome to OStack Knowledge Sharing Community for programmer and developer-Open, Learning and Share
Click Here to Ask a Question

...