While serving time in HPC I saw a set of ETL jobs that were written using awk and xargs that beat the pants off a lot of in house code on both a performance and a simplicity standpoint. You lose a bit of "safety" when you take that approach (yes, I know you can do it in awk), but if you tightly control the inputs you don't need as many guard rails.

It's both unsurprising and amazing to me that a well crafted shell script can be highly performant.