config drive on RBD leaves orphaned loopback device mounts

[David-wahlstrom]

Version: Juno (with https://review.openstack.org/#/c/123073/ backported)

Reproduce:
Create a VM with config-drive on ceph/rbd.

Expected result:
When the VM is created, part of the above patch copies the config drive onto RBD by way of using a loopback device to copy the contents. It is expected that once this is completed, the loopback device will be properly cleaned up. The end result being no loopback devices left open and the config drive being properly stored on RBD.

Actual result:
After the config drive is copied to RBD, a loopback device is left mounted to a deleted file:
/dev/loop0: [0807]:33682451 (/var/lib/nova/instances/0925ed53-16f1-48f2-a190-ab19706b80c6_del/disk.config (deleted))
The config drive is successfully copied to RBD, but eventually all the loopback devices are consumed, causing subsequent VM creations (on the hypervisor in question) to fail and requiring the hypervisor to be rebooted to clean them up.

[Michael Still (mikal)]

Juno is EOL. Can you re-create this issue on a more modern nova?

That said, copying the config drive to rbd is done with just a "rbd --import" in nova.virt.libvirt.storage.rbd_utils. There is a loop mount when the config drive is built if you're using the vfat format, but that is unmounted when we're done.

Can you confirm if you're using vfat as your config drive format please?

[Launchpad Janitor (janitor)]

[Expired for OpenStack Compute (nova) because there has been no activity for 60 days.]