kube resource optimization challenge

hey folks, we're seeing some weird memory spikes in our k8s cluster, even after applying limits. i'm trying to fine-tune requests and limits for a critical microservice, but cgroup stats don't always align with kubectl top. for example, sometimes we get things like this:

$ kubectl top pod my-critical-service-xyz
NAME                       CPU(cores)   MEMORY(bytes)
my-critical-service-xyz    150m         850Mi

$ cat /sys/fs/cgroup/memory/memory.usage_in_bytes
1024000000

that's a pretty big diff. any best practices for robust resource optimization when cgroupv2 and container runtimes seem to report differently? thanks in advance!