Right now reportlab is making PDFs most of the time. However when one file gets several large images (125 files with a total on disk size of 7MB), we end up running out of memory and crashing trying to build a PDF that should ultimately be smaller than 39MB. The problem stems from:
elif mode not in ('L','RGB','CMYK'):
im = im.convert('RGB')
self.mode = 'RGB'
Where nice b&w (bitonal) images are converted to RGB and when you have images with sizes in the 2595×3000, they consume a lot of memory. (Not sure why they consume 2GB, but that point is moot. When we add them to reportlab our entire python memory footprint is about 50MB, when we call
doc.build(elements, canvasmaker=canvasmaker)
Memory usage skyrockets as we go from bitonal PNGs to RGB and then render them onto the page.
While I try to see if I can figure out how to inject bitonal images into reportlab PDFs, I thought I would see if anyone else had an idea of how to fix this problem either in reportlab or with another tool.
We have a working PDF maker using PODOFO in C++, one of my possible solutions is to write a script/outline for that tool that will simply generate the PDF in a subprocess and then return that via a file or stdout.
Short of redoing PIL you are out of luck. The Images are converted internally in PIL to 24 bit color TIFs. This is not something you can easily change.
We switched to Podofo and generate the PDF outside of python.