Detailed bug description:
When I create a new volume from a very large QCOW image (virtual size) the root file system of a controller might be overfilled because the cinder downloads the full size image to the /var/lib/cinder/conversion/ directory
Steps to reproduce:
Install one controller and two ceph/compute nodes. The controller has to have the minimal possible size of a root filesystem (20GB). Ceph size should be enough to keep about 30GB. Create a QCOW2 image with large virtual size and upload it to Glance:
root@node-2:/var/lib/nova/tmp# qemu-img info ubuntu_qcow_20GB.qcow2
image: ubuntu_qcow_20GB.qcow2
file format: qcow2
virtual size: 20G (21474836480 bytes)
disk size: 835M
cluster_size: 65536
Format specific information:
compat: 1.1
lazy refcounts: false
refcount bits: 16
corrupt: false
Then try to create a new volume from this image:
root@node-1:~# cinder create --image-id 4a5c016b-dff0-4e52-b640-bd63528911a6 --name volume_from_image_20GB 20
During the conversion Cinder will download the whole image to the /var/lib/cinder/conversion/ directory and it might overfill the root directory of a controller node
Expected results:
Root filesystem should not be overfilled
Reproducibility:
Fuel 9.1
Workaround:
1. To keep images in Raw format in Ceph so that Cinder does not download them but perform snapshots in Ceph storage
2. To forcefully separate /var from the root file system
Impact:
One of controllers might stop because its root file system is overfilled
Description of the environment:
MOS 9.1
Upstream bug filed: https:/ /bugs.launchpad .net/cinder/ +bug/1653225