[rbanffy]

Python is an excellent choice for that. What I tend to do is build with Python and, if (and that's a big if) the thing needs to be orders of magnitude faster, then we profile the app to find the hot spots and optimize those, either in Python, with some better approach, or in another language still being called from Python. A rewrite of a whole service from scratch is relatively rare in my experience (unless the original is really terrible and impossible to maintain).

[itsmemattchung]

I take a similar approach. Python allows me to rapidly prototype and ... gets the job done — fast (in terms of developer productivity). Once performance becomes a main criteria, it's not so difficult to fork/exec some Go/Rust/C/C++ executable.

[rbanffy]

I'm thinking about using MPI for that - it should minimize the fork/exec time. Another approach is the extensive use of queues (which is a good thing anyway) and process the queue messages using a more optimized backend.