Cells capacity reports can be wrong when multiple flavors have the same disk/ram size

[Andrew Laski]

mysql> select memory_mb,memory_mb_used,free_ram_mb,free_disk_gb from compute_nodes where free_ram_mb > 8000;
+-----------+----------------+-------------+--------------+
| memory_mb | memory_mb_used | free_ram_mb | free_disk_gb |
+-----------+----------------+-------------+--------------+
| 131026 | 121640 | 9386 | 810 |
| 131026 | 121621 | 9405 | 990 |
| 131026 | 121636 | 9390 | 790 |
+-----------+----------------+-------------+--------------+
3 rows in set (0.00 sec)

which indicates three hosts that can handle an 8GB instance, but the capacity reported in the cell log shows:

capacities: {'ram_free': {'units_by_mb': {'8192': 6,

What's happening is that two flavors have memory_mb set to 8192 so the slots are being counted twice.

[OpenStack Infra (hudson-openstack)]

Fix proposed to branch: master
Review: https://review.openstack.org/71372

[OpenStack Infra (hudson-openstack)]

Reviewed: https://review.openstack.org/71372
Committed: https://git.openstack.org/cgit/openstack/nova/commit/?id=3720fd17b060de6f14e537e6c003e923c67c65a0
Submitter: Jenkins
Branch: master

commit 3720fd17b060de6f14e537e6c003e923c67c65a0
Author: Andrew Laski <email address hidden>
Date: Wed Feb 5 15:07:23 2014 -0500

    Count memory and disk slots once in cells state manager

    If multiple instance types in a cell had the same memory_mb value or
    root_gb + ephemeral_gb added to the same value, then capacity counts
    would be inaccurate since slots would get counted multiple times.
    Rather than counting based on instance type it is now based on the set
    of memory or disk values pulled from instance types.

    Change-Id: Iaded7fa69dadb4aa0335ef1569975bdfcb2767c6
    Closes-bug: #1276728