When I create an audit using workload_stabilization and instance_ram_usage as metric as with:

openstack optimize audit create -g workload_balancing -s workload_stabilization -p metrics='[ "instance_ram_usage" ]'

The standard deviation is wrongly calculated.

In my case, I am using two compute nodes with 7678 MB of ram memory each one.

One of the nodes is using ~ 3638 MB and the other ~ 2100 MB

Based on those values the standard deviation should be ~ 0.1 but the audit is reporting:

Standard deviation for instance_ram_usage is 9.797657178882261e-05.

I suspect there is a problem with the units it's using internally for ram

How to reproduce

In a RHOSO deployment with watcher service deployed, create multiple VMs in the same host until the memory usage in different computes is enough that should trigger migrations.

Create an audit with workload_balancing as goal and workload_stabilization as strategy. The metrics parameter should be [ "instance_ram_usage" ]. It should create a non-empty migration plan.

Validation

A new tempest test for workload_stabilization using ram metrics with fake data at least, should be added.