As a general rule, you should always keep in mind that you’re not really looking for a backup solution but rather a restore solution. So think about what you would like to be able to restore, and how you would accomplish that.

For my own use for example, I see very little value in backing up docker containers itself. They’re supposed to be ephemeral and easily recreated with build scripts, so I don’t use docker save or anything, I just make sure that the build code is safely tucked away in a git repository, which itself is backed up of course. In fact I have a weekly job that tears down and rebuilds all my containers, so my build code is tested and my containers are always up-to-date.

The actual data is in the volumes, so it just lives on a filesystem somewhere. I make sure to have a filesystem backup of that. For data that’s in use and which may give inconsistency issues, there are several solutions:

• docker stop your containers, create simple filesystem backup, docker start your containers.
• Do an LVM level snapshot of the filesystem where your volumes live, and back up the snapshot.
• The same but with a btrfs snapshot (I have no experience with this, all my servers just use ext4)
• If it’s something like a database, you can often export with database specific tools that ensure consistency (e.g. pg_dump, mongodump, mysqldump, … ), and then backup the resulting dump file.
• Most virtualization software have functionality that lets you to take snapshots of whole virtual disk images

As for the OS itself, I guess it depends on how much configuration and tweaking you have done to it and how easy it would be to recreate the whole thing. In case of a complete disaster, I intend to just spin up a new VM, reinstall docker, restore my volumes and then build and spin up my containers. Nevertheless, I still do a full filesystem backup of / and /home as well. I don’t intend to use this to recover from a complete disaster, but it can be useful to recover specific files from accidental file deletions.