So what? Sure they would scale better when using Erlang, Java or Go... but sometimes it is wiser to finish building something than making the best ultrascalable system. If you are really successful you will find ways to scale.
Facebook is a good example because of all the money they've thrown at a problem they shouldn't have. First was a custom PHP interpretor then a compiler and now hack. If they didn't have nearly unlimited money to throw at it would things have ended differently?
Language choice is one of the easiest choices to make. Pick a fast one out of the box if you plan to get big. It's not like the faster languages take orders of magnitude longer to write code in, the effect is minimal at best.
I don't think you understand how uWSGI works.
> And Python is a terrible choice for something at their scale.
Maybe they could use whatever youtube uses for their frontend instead?
...or gunicorn. Or twisted / hendrix. Or any WSGI container with process consciousness in the past 5+ years.
Java/Golang/Rust/C++/C# are all about 100X faster than Python. Wouldn't it be more reasonable to use one of those than mess with the runtome to squeeze out 20% more performance?
It's absolutely not the same. The JVM uses GC for all object collection. Python uses reference counting and only relies on GC to clean up cycles.
Forking threads for web pages is so old school...And Python is a terrible choice for something at their scale.
Just redo the hosting bit in Java or golang and call it a day. If their UI code is sufficiently isolated from the back end it's not a huge deal.
Instagram is a pretty small application feature-wise, a few devs could probably do it in a couple months