Both Nova and Cinder publish the vpcus, memory and disk total metrics in the control plane as notifications whenever a new volume or vm is created, along with other useful metrics.

It might be better to retrieve metrics from there instead of coding new metrics on the poller.