Heavily degrading performance when running PDFKit in a web worker

TL:DR: to generate a single PDF file, web workers seem to be much slower than executing PDFKit in the browser’s main thread. On top of that, execution times seem to grow much more quickly in web workers than the main thread as the content grows.

Hello,

I’m trying to generate large PDF files from a browser, and these consist mainly of tables. In order not to cause the main thread to do too much work, which would cause the application to stutter, I moved all the PDF generation logic (including PDFKit) into a web worker. However, I noticed a very sharp decrease in performance as generation times grew greatly.

To compare the performance of PDFKit in the main thread or a web worker, I created a simple script. It essentially creates a set number of pages and writes a table in each:

function generatePdf() {
    var doc = new PDFDocument({autoFirstPage: false});
    stream = doc.pipe(blobStream());
    stream.on("finish", function() {
        callback(stream.toBlob("application/pdf"));
    });

    var startTime = Date.now();
    var counter = 0;
    for (var page = 0; page < NUM_PAGES; page++) {
        doc.addPage({margin: 0, size: "A4"});
        var nextLineAt = TOP_MARGIN;
        while (nextLineAt < PAGE_HEIGHT - HORIZONTAL_MARGIN - 25) {
            var nextColumnAt = HORIZONTAL_MARGIN;
            while (nextColumnAt < PAGE_WIDTH - HORIZONTAL_MARGIN - 100) {
                counter++;
                doc.text("Cell number " + counter, nextColumnAt, nextLineAt);
                nextColumnAt += 100;
            }
            nextLineAt += 25;
        }
    }
    console.log(Date.now() - startTime);
    doc.end();
}

When running it with 100 pages on Chrome 68.0.3440.84, there was a noticeable difference between executing the task in the main or a worker thread, but as I kept on adding more pages, that difference skyrocketed:

Pages	Thread	Execution 1 (ms)	Execution 2 (ms)	Relative slowdown
100	Main	7429	7509	-
100	Worker	10125	10462	38%
250	Main	19028	19513	-
250	Worker	34485	35975	83%
500	Main	39963	38836	-
500	Worker	120310	120703	206%
1000	Main	78780	82270	-
1000	Worker	503469	gave up here	525%

I’m uploading the scripts I used for the benchmarks, as I couldn’t find a CDN link to PDFKit which I could use in a JSFiddle. There are two main files in there, main-thread.html and worker-thread.html. The main function (and the parameter governing the number of pages) can be found on generate-pdf.js.

Is there something I’m doing wrong? Can web workers’ performance be reasonably expected to degrade over time, or is there something in the library itself which can be causing this slowdown?

demo.zip

[jsoo1]

Hi there, I am experiencing the same problem. Will the faster version ever be merged into the main fork?

Update: Oh! I see the commits from the fix got merged with #859 . I still am seeing very heavy performance hits in webworkers. I will open a separate issue.

[alafr]

Can you try the same with this branch: https://github.com/CorvusCorrax/pdfkit/tree/removeStream ?