While debugging https://issues.redhat.com/browse/OCPBUGS-15337, we discovered that VPA (the recommender) asks for all cpu and memory metrics of all containers in all namespaces. On clusters with many namespaces (~4000 in that case),"a lot" of queries can be initiated which can flood Prometheus.

Asking for node folks' input: https://redhat-internal.slack.com/archives/CK1AE4ZCK/p1690796890237489, see for more details.

(We could also see if prometheus-adapter can batch/schedule the queries better, but if there is not query for it to run it's even better)