Hi, I spoke with the Stackdriver team, and there are two sets of metrics – one comes from inside the guest, and one comes from outside. We think you may be able to get what you want from this set 1 (the other set is this set 2).Brian Kessler commented
From the documentation it is not clear what data is returned in  The description says "Agent memory in use" which implies it is the memory used by the agent process and not the memory in use on the instance. Additionally, "The Stackdriver Monitoring agent does not support the following VM instances:
Google App Engine flexible environment instances.
Google Container Engine node instances."
And Google Container Engine node instances are specifically one of our major use cases. There we may be scheduling many different sized containers onto a node and want to know that it is being properly utilized with respect to memory as each container has a requested memory that it requires in order to be scheduled on an instance.Brian Kessler supported this idea ·