django-zip-stream | Django extension to assemble ZIP | Runtime Evironment library

I've got a proxy written in Django which receives requests for certain files. After deciding whether the user is allowed to see the file the proxy gets the file from a remote service and serves it to the user. There's a bit more to it but this is the gist.

This setup works great for single files, but there is a new requirement that the users want to download multiple files together as a zip. The files are sometimes small, but can also become really large (100MB plus) and it can be anywhere from 2 up to 1000 files simultaneously. This can become really large, and a burden to first get all those files, zip them and then serve them in the same request.

I read about the possibility to create "streaming zips"; a way to open a zip and then start sending the files in that zip until you close it. I found a couple php examples and in Python the django-zip-stream extension. They all assume locally stored files and the django extension also assumes the usages of nginx.

There are a couple things I wonder about in my situation:

1. I don't have the files locally stored. I can get them with an async/await structure and serve them simultaneously. That would mean I always have two files in memory (the one I'm currently serving, and the next one I'm getting from the source server).
2. Unfortunately I don't have control over the web servers which will serve this. I can of course put an nginx container in front of it, but I don't think nginx could serve from files I store in Python vars because I get them from the source server.
3. Whether I'm doing this in Python or let it be zipped in nginx, I presume the needed CPU cycles for this would be substantial.

Does anybody know whether streaming zips are a good idea with my setup of very large remote files? I'm a bit afraid that many requests will easily DOS our servers because of CPU or memory limits.

I can also build a queue which zips the files and sends an email to the user, but if possible I'd like to keep the application as stateless as possible.

All tips are welcome!