Comment by sgarland

IMO, the prime offender is simply not understanding fundamentals. From simple things like “a network call is orders of magnitude slower than a local disk, which is orders of magnitude slower than RAM…” (and moreover, not understanding that EBS et al. are networked disks, albeit highly specialized and optimized), or doing insertions to a DB by looping over a list and writing each row individually.

I have struggled against this long enough that I don’t think there is an easy fix. My current company is the first I’ve been at that is taking it seriously, and that’s only because we had a spate of SEV0s. It’s still not easy, because a. I and the other technically-minded people have to find the problems, then figure out how to explain them b. At its heart, it’s a culture war. Properly normalizing your data model is harder than chucking everything into JSON, even if the former will save you headaches months down the road. Learning how to profile code (and fix the problems) may not be exactly hard, but it’s certainly harder than just adding more pods to your deployment.