The VPA's memory usage is proportional to the cluster size. Currently, cluster administrators must appropriately size the VPA's resource requests to have adequate capacity to monitor the pod metrics and relevant object resources. This epic seeks to use the VPA to right-size itself (or allow a cluster admin to configure it to do so).

Concerns which must be addressed:

• The admission webhook deployment will need to be configured so that during an update (i.e. pod restart), the new pod is created before the old one is removed. Otherwise the new pod will potentially come up with default resource request/limit values and the updater will continually view it as being in need of a restart.
• It might not be enough to take only resource requests into account. If we don't take limits into account, it will only be able to scale its own requests up to the provided limits. We should examine options related to scaling requests only (and allow the cluster admin to specify the limits it can potentially scale to) as well as scaling requests and limits.