[Labs-l] Migrating all bots boxes to BTRFS

[Mark Bergsma]

On Feb 14, 2013, at 10:19 AM, Petr Bena <benapetr at gmail.com> wrote:

> Also, don't be confused by this fs being flagged as beta. Most of its developers consider it stable enough for production and generally recommend to use it for any purpose.
> 
> 
> On Thu, Feb 14, 2013 at 10:17 AM, Petr Bena <benapetr at gmail.com> wrote:
> Hi,
> 
> I suggest to switch to BTRFS filesystem on all /dev/vdb devices on all boxes. I will start this experiment on bots-N (experimental) boxes for now. And if there would be no troubles I would start switching it on all production boxes as well.
> 
> BTRFS comes with tons of incredible features that none of existing linux filesystems can do. And I would really like to perform maintenance on these boxes without having to unmount its filesystem everytime when I need to change it (BTRFS support basically everything you need to unmount regular FS for ONLINE without any need to be unmounted, including defragmentation, resizing, mirroring (on level of file, directory or subvolume), making snaphots of files or directories and many more)

Yeah well, do make sure you have backups. I've been using btrfs for a couple years now on several systems, and while it mostly works well and is very cool, it has eaten my data a few times. Even recent versions. :)

What's funny is that the one box where I did expect trouble, a Wikimedia server used in production with a btrfs file system spanning 48 separate hard drives, worked great under load for 2 years.

-- 
Mark Bergsma <mark at wikimedia.org>
Lead Operations Architect
Wikimedia Foundation