I'm not in a good position to say for sure, as I've only used managed Kubernetes distributions, but I think it probably works out to less work than removing CPU

I'm not in a good position to say for sure, as I've only used managed Kubernetes distributions, but I think it probably works out to less work than removing CPU limits. Kernel upgrades are at least semi-routine, so most shops that run Kubernetes themselves are going to have a process for them. Conversely, removing CPU limits and migrating the critical path to a different set of tainted nodes is a substantial one-off change with a long tail of failure scenarios that need to be tested. Thus, I would expect that a kernel upgrade would be easier than doing what Buffer did.