nova hypervisor-show/stats reports incorrect values

[Ben Nemec]

When looking at hypervisor resources through either nova hypervisor-stats or nova hypervisor-show with compute nodes of differing sizes, I am getting incorrect/inconsistent values back for one of the hypervisors. For example, in an environment with one 32 GB compute node and one 16 GB node, I see the following when running nova hypervisor-show multiple times on the 32 GB node:

$ nova hypervisor-show 1 | grep memory_mb
| memory_mb | 15934 |
| memory_mb_used | 20992 |
$ nova hypervisor-show 1 | grep memory_mb
| memory_mb | 31906 |
| memory_mb_used | 20992 |

hypervisor-stats shows similar incorrect behavior:

$ nova hypervisor-stats | grep memory_mb
| memory_mb | 31868 |
| memory_mb_used | 34304 |
$ nova hypervisor-stats | grep memory_mb
| memory_mb | 63812 |
| memory_mb_used | 34304 |

From what I can tell, the same stats are being returned for both hypervisors, but which node's stats are being used randomly changes.

This particular environment is a two node devstack setup built today, but I've seen similar behavior in a three compute TripleO deployment using recent builds of Nova for at least a couple of weeks.

[Sarafraj Singh (sarafraj-singh)]

Please attach logs.