Comment 16 for bug 1349711

It is hard to say for sure. I would not know the details of EC2 guest setup. The block device could be local but more likely is some form of network attachment (iscsi, ndb). And btrfs is still a newer kid on the block. So there still might be surprises just because of that.

You description was about what we wanted. Something that might be used to come up with a reproducer which in turn could be run on bare-metal or a different hypervisor to prove or rule out a tie with EC2/Xen. One detail I did not see was whether the btrfs was created with special options (what mkfs.btrfs line was used). So we can get the as close as possible (minus the cloudy details).

For your production systems. Is using btrfs a hard requirement or could those at least work around by getting moved to ext4 or xfs? At least until the btrfs issue is understood. Which could take a bit (at least it is like reading a crystal ball to make any estimate).