[Discuss] VM backups

[Richard Pieri]

This can be a significant problem with virtualization. If you back up
container files then you can easily restore the entire VM but you can't
easily restore individual files. If you back up from within the VM then
you can easily restore individual files but restoring the complete VM is
largely impossible.

The problem here isn't that doing backups is cumbersome. It's that
treating the VM as a single monolithic entity is a poor practice. A
typical Unix system has separate file systems for OS, configuration data
(/etc), user data (/home), logs (/var) etc. If you apply that to virtual
disks then you can have a vdisk for the OS, a vdisk for configuration, a
vdisk for transient data and however many vdisks for user data that you
need. This gives you the ability to tailor backup solutions for each
kind of data and the associated vdisks:

OS and transient data vdisks aren't backed up. There are master copies
that can be dropped into place as needed. If keeping logs is a
requirement then a central log server is a good way to handle them.

Configuration data vdisks aren't backed up per se. A configuration
management system or version control system makes for a better way to
manage configuration data.

Home directories are backed daily up from within the guests themselves.
This is the only sane way to manage single file restores. Virtual disks
may be cloned on a weekly or monthly basis to expedite disaster recovery.

Other data should be handled on a case by case basis.


I don't do this. All of my VMs are tiny things less than 5G each. I use
rsnapshot to perform nightly backups of important bits of data to a
central file server. The fact that they're all virtual machines is
irrelevant to my backup procedures.

-- 
Rich P.