OpenStack Compute (nova)

Hypervisor API returns CPUs as VCPUs

Bug #1241343 reported by Sam Morrison on 2013-10-18

This bug report is a duplicate of: Bug #1202965: The name VCPUs (total) of Hypervisors is confusing. Edit Remove

This bug affects 3 people

Affects		Status	Importance	Assigned to	Milestone
	OpenStack Compute (nova)	New	Undecided	Guangya Liu (Jay Lau)

Bug Description

The stats returns in the hypervisor API are incorrect.

Eg it returns vcpu when they are actually cpus. (or cores to be precise, I think....)

Eg you can have: vcpus = 64 and vcpus_used = 67 (when over allocating)
vcpus should take into account the over allocation ratio.

This is also the case for memory.

Here is an example:

u'current_workload': 0,
u'disk_available_least': 2476,
u'free_disk_gb': 3087,
u'free_ram_mb': -17108,
u'hypervisor_type': u'QEMU',
u'hypervisor_version': 1000000,
u'local_gb': 5357,
u'local_gb_used': 2270,
u'memory_mb': 257836,
u'memory_mb_used': 274944,
u'running_vms': 26,
u'vcpus': 64,
u'vcpus_used': 67

Tags:

Revision history for this message

Guangya Liu (Jay Lau) (jay-lau-513) wrote on 2013-10-18:

Its an interesting problem, we did encounter this many times. I think it is a good time to do some enhancement in Icehouse.

Not only cpu and memory, but disk also have such issue.

A possbile solution was that add some new fields in DB to record the cpu/memory/disk that have been multipled by allocation ratio, but not update vcpus/memory_mb/free_disk_gb directly.

I think that it might be better to log a blueprint to trace this, as this is more like to be a feature