When building my image (centos 7 image with bits for Sahara [openstack data processing]) for qcow2, it succeeds as expected, but when I add -t tar to generate a docker loadable archive, I get the following.
In case it matters, this is being done on a Fedora 20 machine
tar: ./dev/log: socket ignored
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 10 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 9 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 8 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 7 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 6 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 5 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 4 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 3 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 2 more seconds...
losetup: /dev/loop6: detach failed: No such device or address
/dev/loop6 may be busy, sleeping up to 1 more seconds...
Gave up trying to detach /dev/loop6
If I add ./dev/log to the excludes in the "sudo tar...." line, the tar warning goes away, but the losetup failures remain. It looks like the losetup failure is eventually generating an exit code of 1, which is unfortunate.
This probably isn't super high priority because my tar archive does get built and it works, but it seems like something that should be handled.
I believe the problem is that we're treating all losetup failures as temporary.