don't save images on the hard drive

[eroux]

The current code is something like:

• for each pecha:
  • download all the images of a volume on the hard drive
  • read the images again to send them to GV
  • remove the images from the hard drive

It would be much more efficient to have something like

• for each pecha:
  • for each image
    • download from s3 in memory
    • send to GV and free the associated memory

or even better: stream the bytes from s3 to the GV request directly.

[ngawangtrinley]

If I remember correctly we download for conversion since some formats aren't supported by GV. @10zinten is that correct?

[eroux]

it doesn't change anything, the conversion with PIL happens in memory anyway, there's not need to write anything on the hard drive