Comment 7 for bug 1739646

I believe instance_type.disk == 0 is paleolithic in OpenStack terms. It recall it causing me headaches round about Havana, at least. Pretty sure it just affects everybody.

I believe the relevant test here is in compute.API._check_requested_image, which intentionally skips the flavor disk size check for volumes. This suggests a kludgy non-code workaround, which is to set the flavor disk size to 1 instead of 0. This would give users up to 1GB of local disk space per instance. This is 1GB more than the operator intends, but it's unlikely to be a practical attack.

I would also be in favour of the config setting to disable the size 0 special behaviour. I think the combination of this and pointing out the kludge should cover most use cases until we fix this properly.