Shared instance storage is a mixed bag. As long as it works, it reduces our vulnerability to the crash of a single virt host.

On the downside, if we rely on a single file system (as we did, historically, with Gluster), a single file-system bug can kill everything at once.

So, this is something we should look at but it needs to be tested extravagantly before we place any reliance on distributed instance storage again.

This task is now part of https://wikitech.wikimedia.org/wiki/Incident_documentation/20190213-cloudvps

Good technical overview of Ceph: https://www.suse.com/media/report/discover_cephfs.PDF