The purpose of cinder image caching is to cache a copy of each glance image on the cinder backend, so that subsequent volume creates from that image do not need to download the image from cache. However, when concurrent requests are made to create volumes from a 'new' glance image we end up with multiple copies of the same glance image in the backend.
In general this is not a problem, and it is noted in the spec:
"It is also possible that multiple of the same images are cached at the same time, this is considered ‘OK’, eventually extra ones will be evicted if space is needed."
However, if a customer creates a new glance image, and then creates many volumes from that image, we can end up with many copies of the image, which is not desirable. For example, we've seen users create 30 volumes concurrently, and end up with 30 downloads and 30 copies of the new image in the cinder cache. As well as wasting space and causing confusion, the create times for the volumes can be very long, because of the time to download 30 copies of the image.
Fix proposed to branch: master
Review: https:/