Currently, when ambra returns a file for download (pdf, xml, tif, supporting info) it Loads the whole file into memory and caches it.
This does not make sense for very large files.
We have one bio article with a number of ~250MB supporting info files.
We should find a way to avoid caching large files, and just return them directly from fedora without loading into memory on the ambra side.
I think it would be ideal to switch based on asset type. we want to cache thumbnails and equations on the article page, medium sized PNGs on the slideshow page, and XML files. These are the objects with type PNG, PNG_S, PNG_M, and XML on the fedora side. Everything else - including PNG_L - would be a straight download.
Alternately, we could set a filesize limit for the cache (only cache files smaller than 1MB or 10MB?)
this is a requirement for the plos bio/med ambra migration.