Performance issues when rendering large PDFs

[jesusgp22]

This might be a good question for pdf.js community itself but how does rendering large PDFs can be better handled with react-pdf?

pdf.js suggests not rendering more than 25 pages at a time:
https://github.com/mozilla/pdf.js/wiki/Frequently-Asked-Questions#allthepages

I even had to add this to my component to keep react from trying re-create the virtual DOM of the Document:

    shouldComponentUpdate(nextProps, nextState) {
        if(nextProps.file !== this.props.file
            || nextState.numPages !== this.state.numPages
            || nextState.width !== this.state.width){
            return true
        }
        return false
    }

The problem is that I also need to dynamically set the width of the document on user interacting so I can't save myself from re-creating the virtual DOM after width changes, any way I can achieve this with your lib?

[michaeldzjap]

@jesusgp22 You probably want to use some kind of virtualization library for displaying PDF's with a lot of pages, like react-virtualized for instance. Maybe this is useful to you.

[jesusgp22]

Hey, thank you so much for your answer, I'll def check this out

You might want to add a note about this on react-pdf documentation to help others with the same performance issues or even in the future add this as a core feature for large docs.

[jesusgp22]

Following up on this @michaeldzjap I am watching some presentations on react-virtualized and it will break text search feature, is this a trade off that I can't get around?

[michaeldzjap]

I am not familiar with the text search feature I have to admit. But I suspect that it relies on the text layer for each page to be rendered in order to be able to find all the relevant results for a specific search query (e.g. a word that could be located anywhere in the document). The whole point of virtualizing a collection of elements (Page components in the case of react-pdf) is to not render them all at the same time.

I don't think there is an easy way around this unfortunately. A solution could be to keep a virtual representation of a text layer of each page in memory (like how React does this for HTML elements) and search through that instead. Might be possible.

[jesusgp22]

That's an interesting approach, I am guessing this will most likely break browser text search feature anyway, in any case I think it is ok to implement this using just a regular search box element. Now the questions are:

• How can I extract the text from the pdf to keep a "virtual" copy of the whole text layer I can search from
• After getting a list of results from the text how can I implement a feature to seek these features in the document (guessing I will need to map results to scrollbar coordinates accordingly)

[michaeldzjap]

How can I extract the text from the pdf to keep a "virtual" copy of the whole text layer I can search from

I think you would need to dig into pdf.js for this, relying on the react-pdf api probably is not enough. You can get the text content for a page using this apparently:

page.getTextContent().then(function(textContent) { ... });

After getting a list of results from the text how can I implement a feature to seek these features in the document (guessing I will need to map results to scrollbar coordinates accordingly)

Yes, that is a tricky one... You'd know the page number. Maybe it should be a 2 step operation or something. 1 - Search through the virtual text layers for a query. Keep a result of all pages that match. 2 - For each page in the result of step 1 see if it is rendered, if it is you can probably find the word rather easily, because I think each word is rendered as a separate HTML element in a text layer. If the page is not rendered yet, scroll to it with react-virtualized so that it will be rendered and then again find the HTML element that contains the first occurrence of the word/query in the text layer element tree.

Something like the above. I might think too simplistic about this, I haven't actually tried it myself. But this is how I would approach things initially I think.

[jesusgp22]

I was wondering if the biggest performance issue was rendering the text layer or the canvas, in case rendering the canvas is an issue, it might be possible to ask pdf.js to only render the text layer?
I know this is not possible with the current react-pdf API

[wojtekmaj]

@jesusgp22 Nope, you can toggle text content and annotations on/off, but canvas is currently not behind a flag. I don't see a good reason against having it toggleable, though :)

[wojtekmaj]

I think you would need to dig into pdf.js for this, relying on the react-pdf api probably is not enough.

@michaeldzjap Any reason for this? Documents's onLoadSuccess should return all pdf properties and methods, and Page's onLoadSuccess should return all page properties and methods.

If you use Document, you can get the number of pages, iterate through all of them with pdf.getPage(pageNumber) and run the code you pasted on getPage()'s results.

[michaeldzjap]

@wojtekmaj Yes, my wording was rather poor. What I meant is that pdf.getPage(), page.getTextContent() etc. all are pdf.js related rather than react-pdf specific. So although of course all those methods are perfectly well accessible through the react-pdf API, they really belong to the underlying pdf.js API.

If you use Document, you can get the number of pages, iterate through all of them with pdf.getPage(pageNumber) ...

Yes. This is exactly what I do to cache all document page widths and height on initial load when using react-pdf together with react-virtualized.

[jesusgp22]

Thank you both for this amazing discussion 👍

[MarcoNicolodi]

We are also having trouble loading long PDFs. We are loading a 17mb PDFs and the application crashes, and since we have customers with 100mb+ PDFs, crashing is not an option.

This example which is also a react wrapper to PDF.js seem to work for us. It tricks PDF.js to load only the current visible page and the ten previous pages. It looks like it has something to do with the wrapper div's styles, because when you change some of the styles it loses it lazy loading behaviour.

I couldnt reproduce this trick to your lib. But we liked react-pdf so much that we are still trying to adapt this lazy load trick to it.

We like the fact that your lib has no default toolbox and that it has mapped its props to pdf.js handlers/configs, so we can develop our customized toolbox.

So we would be glad to see it working better with long pdfs, maybe using this trick that yurydelendik/pdfjs-react uses (thats a shame that I couldnt reproduce it with your lib! )

[jesusgp22]

@MarcoNicolodi I found that react-virtualized worked really bad with react-pdf I implemented the aproach to only render a few pages but to make things work you have to render a div that has the dimensions of the pages you don't render

you can 100% integrate this with react-pdf using the document object that is returned by react-pdf and use getPage and page.getViewport methods to get the page dimensions

I built my own algorithm to detect what pages are visible and I run it everytime the user scrolls or if a resize event happens.

[wojtekmaj]

Hey everyone,
I'd like to remind you that it was never React-PDF's intention to provide the users with fully-fledged PDF reader. Instead, this is only a tool to make it. While I have a plan of creating React-PDF-based PDF reader, I'm far from it. Mozilla is working on it for years and they seem to never be done. I think it would go similar way ;)

There is some good news too, though. If I can suggest something, onRenderSuccess callback that you can define for <Page> components can be your powerful friend. You can use it to, for example, force pages to be rendered one by one:

import React, { Component } from 'react';
import { Document, Page } from 'react-pdf/build/entry.webpack';

import './Sample.less';

export default class Sample extends Component {
  state = {
    file: './test.pdf',
    numPages: null,
    pagesRendered: null,
  }

  onDocumentLoadSuccess = ({ numPages }) =>
    this.setState({
      numPages,
      pagesRendered: 0,
    });

  onRenderSuccess = () =>
    this.setState(prevState => ({
      pagesRendered: prevState.pagesRendered + 1,
    }));

  render() {
    const { file, numPages, pagesRendered } = this.state;

    /**
     * The amount of pages we want to render now. Always 1 more than already rendered,
     * no more than total amount of pages in the document.
     */
    const pagesRenderedPlusOne = Math.min(pagesRendered + 1, numPages);

    return (
      <div className="Example">
        <header>
          <h1>react-pdf sample page</h1>
        </header>
        <div className="Example__container">
          <div className="Example__container__document">
            <Document
              file={file}
              onLoadSuccess={this.onDocumentLoadSuccess}
            >
              {
                Array.from(
                  new Array(pagesRenderedPlusOne),
                  (el, index) => {
                    const isCurrentlyRendering = pagesRenderedPlusOne === index + 1;
                    const isLastPage = numPages === index + 1;
                    const needsCallbackToRenderNextPage = isCurrentlyRendering && !isLastPage;

                    return (
                      <Page
                        key={`page_${index + 1}`}
                        onRenderSuccess={
                          needsCallbackToRenderNextPage ? this.onRenderSuccess : null
                        }
                        pageNumber={index + 1}
                      />
                    );
                  },
                )
              }
            </Document>
          </div>
        </div>
      </div>
    );
  }
}

Of course you can do much more - add placeholders, check on scroll which pages need rendering, keep info on whether all pages so far were rendered... I believe in your creativity ;) And if I can be of any help regarding API, please let me know!