partitioning and subvol strategy with btrfs

Question

I'm new to btrfs and I'd like some advice on partitioning and subvolume strategy. The system is a light-duty web server, and assume it only has a single disk.

With ext filesystems, I've always created separate partitions for /, /var, swap (and possibly /boot and /home). For me, /var always contains all valuable web server data (e.g. MySQL databases) and none of the code. That allows me to easily move the data to a different system (move or copy /var) or re-install the OS without disrupting the data (reformat /) etc.

Using btrfs, I could do the same thing, use the same partitioning scheme, and have a separate btrfs filesystem on each partition. Or, I could have a single partition, and use btrfs subvolumes for /, /var and so forth. What would be the pros and cons of that?

It seems to me there may be some advantage in having /-only and /var-only snapshots, for example ("Restore all the data to a previous checkpoint" vs "restore all the code" vs "restore both"). Is that right, or does it only appear that way?

Bonus question: are there advantages to using lvm below the btrfs filesystem(s)?

Bonus question 2: how would your advice change if the system had two disks of the same size?

Any pointers to "this is what I did and how it has worked for me" articles would also be appreciated. I can find plenty of material on what I can do, but I'm failing to find much saying "this is what I tried and this is why it worked or didn't".

Solution 1

If you don’t have specific needs, use btrfs as you would use another filesystem. Separating /home is a good practice.

Personally, on home servers my only subvolume is /etc, so I can make snapshots of the configurations. This can be automated with tools like snapper.

Usually, there is little interest of restoring just a previous version of /var, because of the need to also restore /lib/stuff. It is an all-or-nothing situation.

Snapshots of /home can be VERY big, so the disk size management is soon a problem. It can be done without any problem, but keep an eye on the remaining space. Also, as snapshots can only be made on the same disk, they are not a solution for backups in case of disk failure. Think of them as something for the situations like "oups, I deleted this file two hours ago, but I still need it".

Bonus 1 : none. In fact, btrfs was designed in order to simplify the stack mdadm+lvm+fs. So it is really better to avoid it.

Bonus 2 : No, but make a RAID 1! Simple and efficient, your data will love you :)

Ninja bonus : you really probably want to have a good look at the btrfs wiki.

Solution 2

I've been revisiting this lately and thought I'd share a very well thought out post that proposes one partition with top-level directories containing subvolumes: https://bbs.archlinux.org/viewtopic.php?id=194491

TL;DR

subvolid=0
      ├── subvol_root
      │        └── /usr, /bin, /sbin, /.snapshots, etc
      ├── subvol_snapshots
      ├── subvol_home
      └── subvol_opt

Comments

[Imran Juma]

Great question! I wanted to ask almost the same. I'd like to use raid1 and luks encryption, but it is hard to find relevant info about the basics. E.g. whether I can install an op system on a single duplicated volume and can use subvolume for /swap, etc. without any drawback. Maybe I should read the manual, but I don't like long texts. :D

[JOduMonT]

btrfs support mirroring and other RAID'ish configuration. Even you should avoid to reproduce RAID6, you could easily make a mirror by simply adding a partition after the installation. You could find a nice how to here (complang.tuwien.ac.at/anton/btrfs-raid1.html)