Elastic PyPI: your own Python Package Index on AWS Lambda

This is a really cute use for Lambda, and a great working example of the kinds of limitations one could expect to bump into.

I was recently asked to evaluate Lambda for use in hosting a bunch of existing web sites, and just from README.md, I'm glad I backed the hell away from that pandora's box before opening it.

This is a perfect use for Lambda (i.e., low-traffic, once-off tasks that don't require a full server), or at least this would be a good fit if not for immaturity issues like missing WWW-Authenticate.

But Lambda at scale is very expensive. A 100 request/second would cost more than $1,500/month, when even a single nano server could easily service that kind of load all by itself.

Back of napkin analysis: https://twitter.com/JamiesonBecker/status/802185522139582464

Lambda is great for tech demos, small or rare loads, triggered tasks, etc. Not so great as part of your core app pipeline.

Hosting a PyPI index on S3 is simpler, but I admire the effort here!

This is cool, but it makes me sad that it has to run on Python 2, as Lambda does not natively support Python 3.

(And no, I'm not counting tricks like calling into the underlying OS' Python 3)