how do you decide a k8s resource request is actually safe to lower?
retained text
say a workload uses a quarter of what it requests. cutting the request seems obvious, but a couple of things complicate it. if its HPA targets CPU utilization, lowering the CPU request can cause it to scale up even though actual CPU usage hasn’t changed. and after an OOMKill, current memory usage can look fine while missing what caused the previous container to get killed. especially if you’re lowering memory limits along with requests. both came up while we were building rightsizing into Radar (open source). for those doing this in prod, what other cases make you think twice before lowering requests?
document fetched 2026-10-15 00:40 · canonical https://reddit.com/r/kubernetes/comments/1x5ydk7/how_do_you_decide_a_k8s_resource_request_is