We can now use Vertical Pod Autoscalers to mostly set-and-forget per-container resource requests and limits.
How
Mostly, I adjusted the base resource requests of each service based on current usage, then updated the resource limits to preserve the ratio between them and the requests, then added VPAs to the production configuration of each service, with scaling limits being a bit higher than the current resource limits.
Why
We can now use Vertical Pod Autoscalers to mostly set-and-forget per-container resource requests and limits.
How
Mostly, I adjusted the base resource requests of each service based on current usage, then updated the resource limits to preserve the ratio between them and the requests, then added VPAs to the production configuration of each service, with scaling limits being a bit higher than the current resource limits.