get_available_resource/get_inventory should account for resources used by the NovaLink, VIOSes, and non-OS partitions
| Affects | Status | Importance | Assigned to | Milestone | |
|---|---|---|---|---|---|
| nova-powervm |
New
|
Undecided
|
Unassigned | ||
Bug Description
Today our 'reserved' amounts for VCPU, memory, and disk are zero. This does not reflect reality.
Ideally, we should be doing roughly the following:
Prerequisite: Gather a list of all partitions (LPAR and VIOS, including the NovaLink) that are *not* instances tracked by OpenStack. Hereinafter referred to as "non-OS partitions".
VCPU, memory: Sum 'min' amounts for all non-OS partitions. I don't remember whether/how inactive partitions reserve their allocations - take that into account too.
Disk:
- For the localdisk driver, go look at the configured VG, figure out which LVs are associated with non-OS partitions, and sum up their sizes.
- For the SSP driver, start by summing all image LUs. Find disk LUs associated with non-OS partitions and (I think) sum the sizes of the image LUs that back them; and be sure we're using a heavy overcommit ratio.
