Compute node reports wrong amount of free memory

[Geza Gemes]

In nova/compute/resource_tracker.py in the definition of the _update_usage_from_instances function the amount of free memory reported by the hypervisor layer gets overwriten by the amount of configured host reserved memory:

resources['memory_mb_used'] = CONF.reserved_host_memory_mb

If the compute host uses more memory (not counting the VMs memory usage) than the reserved_host_memory, it results in erroneus free memory reporting by the compute host and consequently to suboptimal scheduling.

[Geza Gemes (geza-gemes)]

Fixed commited at: https://review.openstack.org/#/c/70835/

[Joe Gordon (jogo)]

see comments on review

[Geza Gemes (geza-gemes)]

The problem is valid, the proposed solution can be discussed. I will start a blueprint about it shortly.

[Joe Gordon (jogo)]

I am fine with treating this as a blueprint, but then we should track it as a BP not a bug.

[Sean Dague (sdague)]

We'll track it as a blueprint instead