Ask Your Question

disk usage wrong, local vs. cinder?

asked 2016-03-29 08:24:06 -0500

daubler gravatar image


I'm confused with the different numbers of disk usage, and I'm unable to launch another instance. 3-node installation: 1 controller, 1 Block(cinder/LVM)/Network, 1 compute As admin on the dashboard under Hypervisors I see:

Local Disk Usage Used 81GB of 98GB

this is wrong, a df on the compute node shows:

/dev/mapper/rootvg-root   99G  1.9G   92G   2% /

It is absolutely no problem to make volumes in any size, cinder is working, but when it comes to launch an instance (boot from image (creates a new volume)), this fails with:

Failed to run task cinder.scheduler.flows.create_volume.ScheduleCreateVolumeTask;volume:create: No valid host was found. Exceeded max scheduling attempts 3 for volume None

in cinder/scheduler.log unless the desired volume fits into the "free" space (17G) reported by nova-compute.log:

Final resource view: name=compute-node phys_ram=257401MB used_ram=9216MB phys_disk=98GB used_disk=81GB

There are terabytes of free space in the volume-group of the block-node, so this shouldn't be a ressource-problem. For me it looks like nova doesn't know about all the free space cinder provides.

Please help atb daubler

edit retag flag offensive close merge delete


found a "workaround" to launch instances on new volumes. "disk_allocation_ratio=10.0" - a really dirty trick.

daubler gravatar imagedaubler ( 2016-03-30 07:26:30 -0500 )edit

4 answers

Sort by ยป oldest newest most voted

answered 2016-03-30 07:38:23 -0500

PZ gravatar image

Hello Daubler,

Did you mix the compute local storage and Cinder storage? Otherwise, why you shew the result of df on the compute node?

When you create a instance with "creates a new volume" option, the volume will be created on the Cinder storage node, by default, for my RDO installaion, it's "cinder-volume" volume.

Please run vgs on the Cinder storage node, then may you get some clue.

Regards, PZ

edit flag offensive delete link more


yes, that's why I'm puzzled - my VG is called opstvg, lvs:

 VG     #PV #LV #SN Attr   VSize   VFree
  opstvg   1   9   0 wz--n-   7.28t 7.13t
  rootvg   1   2   0 wz--n- 108.00g 4.00m
daubler gravatar imagedaubler ( 2016-03-30 08:14:53 -0500 )edit

making volumes is absolutely no problem, only when it comes to launching an instance with a new volume the "borders" of the local storage of the compute node (where never ever a volume will be created) prevent the creation of the volume... Strange, really strange.

daubler gravatar imagedaubler ( 2016-03-30 08:20:05 -0500 )edit

answered 2016-05-20 07:39:19 -0500

ZaphodB gravatar image

This is a bug in nova-compute which is currently being addressed. See my answer here: (

edit flag offensive delete link more

answered 2016-07-18 18:13:05 -0500

lost2 gravatar image

By local disk, you mean ephemeral?

Please check (

edit flag offensive delete link more

answered 2016-10-01 11:11:40 -0500

andy2411 gravatar image

Hi All,

I have just finished installing Liberty on a 2-node test system: 1-controller/neutron/nova/keystone/glance/cinder(not volume service) and the other machine as a block-storage node running the cinder-volume service.

All services are up and running.

The issue being experienced is as mentioned above, that is, nova is not seeing the storage-node when spinning up VMs. "DiskFilter returns 0".

I have tried disk_allocation and enabling libvirt options presented as answers, but they both do not help in resolving my issue.

My resolution at the moment is to also make the storage node a nova node enabling the libvirt options (as ZaphodB mentions) and I can finally create VMs.

I have scoured the internet and tried different options but the DiskFilter sees 0 hosts as it cannot satisfy the condition for disk and reports back "No valid Hosts found".

edit flag offensive delete link more

Get to know Ask OpenStack

Resources for moderators

Question Tools



Asked: 2016-03-29 08:24:06 -0500

Seen: 5,507 times

Last updated: Oct 01 '16