Virtual Storage LVM volume is not needed when Ceph RBD is used for ephemeral volumes

[Dmitry Borodaenko]

Nailgun partitions manager allocates at least 5GB of disk space on compute nodes for Virtual Storage LVM volumes. When Ceph RBD storage backend is selected for ephemeral volumes, the Virtual Storage volume isn't used to store VMs, and the disk space allocated to it remains wasted.

[Arminder Singh Girgla (arminder)]

I think it uses it for caching the images.

[Dmitry Borodaenko (angdraug)]

It's not supposed to do that anymore, it doesn't need to cache the images when QEMU can access them directly via RBD. The only case left that I'm aware of is https://bugs.launchpad.net/nova/+bug/1262914 (image is still downloaded from Ceph when a snapshot is created from an instance).

[Vladimir Kozhukalov (kozhukalov)]

Ok. It is not about caching. It is about storing ephemeral VM images.

[Dmitry Borodaenko (angdraug)]

This bug has to be postponed to 5.1: it cannot be resolved until the bug mentioned in comment #2 above is addressed in Nova, and it wasn't addressed in Icehouse so far.

[Dima Shulyak (dshulyak)]

Bug in comment #2 still not resolved in nova, so i am postponing it till 6.0.

[Dima Shulyak (dshulyak)]

Bug is not fixed in upstream yet, i am posponing this bug for later

[Vladimir Sharshov (vsharshov)]

Can be safely moving to 8.0 as known-issue

[Dmitry Pyzhov (dpyzhov)]

I think wasting 5Gb of disk space is a bug, not a feature. And it is not a high priority.

[Anton Samoylov (asamoylov)]

Can't be fixed at the moment.

Nova image cache still uses /var/lib/nova/instances/_base/ dir for converting and uploading images to ceph.

root@node-2:~# nova-compute --version
13.0.0

During instance boot process:

1. downloading qcow2 image from glance to /var/lib/nova/instances/_base/
2. converting qcow2 image to raw
3. uploading raw image to ceph
4. image resizing in ceph according to instance's flavor.

There is a bug also related to this issue:
https://bugs.launchpad.net/nova/+bug/1420274