I have PDFs as large as 1GB on my platform but I have ensured they are "linearized". With your viewer:

• The viewer seems to be trying to load the whole file before attempting rendering
  • Pdf.js does support range requests but may need some configuring on your end
• If I try to load a PDF by assigning the :url but then “cancel” by unmounting the PDF viewer, the download continues in the background
  • That could mean the client sucking up the bandwidth without even realizing!

To address the first point, I recommend exposing props like

• disableAutoFetch
• disableStream
• rangeChunkSize
  which can get passed to pdfjs.getDocument and I can override these to not load the whole document unless needed.
  You may still need to be careful on loading page by page both in canvas and the thumbnail strip - I don't know if your viewer always renders all pages in advance.

To address the second point, pdfjs.getDocument returns a loading that can be cancelled on events unlike unmount

I've successfully tested this approach with this codehttps://github.com/Brief-Tech-Pte/localdoctools/tree/main/src/components/pdf-viewer albeit this is using pdf.js v5 and I think you guys are on v4.

This capability can be tested by loading a synthetic 100 page 50MB PDF and observing the viewer on loads what is required to view the pages being rendered. The linked repo above has this test case.