> This will be a bit controversial, but Kubernetes, because if people use it via GKE, EKS etc. then I won't have to learn their organically grown solution to the same dozen-or-so operations problems, and I have yet to see one that isn't a hot mess in some way or other. Also anything running on top of Kubernetes won't be built the very old-fashioned vi-edits-on-server way, great for sanity.
I appreciate this perspective. It's sane. But my consistent experience has been that the complexity permitted by the Kubernetes configuration surface area tends to be larger, more complex, and ultimately more difficult to wrangle than the complexity introduced by (reasonably coherent) home-grown systems.
I would generally prefer to be responsible for a bespoke solution, with all of its warts and pathologies, than for a Kubernetes deployment. The problems of the former tend to be at least tractable; wrangling Kubernetes to a reasonable level of availability is in my experience only possible if you dedicate your career to it.
I appreciate this perspective. It's sane. But my consistent experience has been that the complexity permitted by the Kubernetes configuration surface area tends to be larger, more complex, and ultimately more difficult to wrangle than the complexity introduced by (reasonably coherent) home-grown systems.
I would generally prefer to be responsible for a bespoke solution, with all of its warts and pathologies, than for a Kubernetes deployment. The problems of the former tend to be at least tractable; wrangling Kubernetes to a reasonable level of availability is in my experience only possible if you dedicate your career to it.