Hacker News new | ask | show | jobs
by jeroenhd 490 days ago
Are these tweaks exclusively beneficial to mixed-media RAID setups, or do they also pose an advantage to single-device BTRFS setups?

I know BTRFS is the default FS for a bunch of distros now, but I haven't heard much about it in the RAID space. Most people seem to use ZFS, despite the massive memory pressure it adds, because of fear of bugs and reports of filesystem corruption years ago. Perhaps someone who uses BTRFS for RAID can comment on how well BTRFS RAID works these days?
5 comments
ThatPlayer 490 days ago
RAID1 has been fine for years, but hardly anyone is running that and RAID5/6 are still considered unstable even by the devs: https://btrfs.readthedocs.io/en/latest/Status.html#block-gro...

I ran into Btrfs RAID6 corruption myself about ten years ago. Though to btrfs's credit, I didn't lose any data: the filesystem only locked up read-only on me. But this time I'll take their word on unstable.

link
nisa 489 days ago
> RAID1 has been fine for years

It's never been RAID1 it's the odness of the pid decides which disk to read from.

link
forza_user 489 days ago
There is a patch for this. Hopefully it will be available in mainline kernels soon. https://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux....
link
bjoli 490 days ago
Btrfs raid 1 works great. I have 3 10tb drives in a raid1c2 for a total of 15tb of space. Btrfs raid1 allows you to use any combination of drives and sizes. If I would have one 12tb drive instead of one 10tb I would have had 16tb of usable space.

For raid5/6 I dont use the built in raid, but mdraid. I stopped waiting for btrfs raid56 to stabilise.

link
fenazego 478 days ago
I have been using btrfs in raid1 setups over the past years, and have not had any issues. Quite happy with the features (snapshotting, send/receive, adding drives to grow an fs)
link
forza_user 489 days ago
The allocator hints only makes sense with at least two devices. Btrfs is metadata heavy, so offloading metadata to faster devices can make a big performance difference. For example a HDD combined with a small SSD or NVMe disk, you would set "data preferred" on the HDD and "metadata preferred" on the SSD/NVMe disk.

One advantage with these patches compared to using dm-cache, lvmcache or bcache is that everything is native Btrfs and fully compatible with non-patched kernels (though the 'hints' won't apply for new writes). Btrfs allows for adding additional devices to existing filesystems, so no mkfs and data-shuffling is needed. If you do not like the result, simply set default hint (type 0). You can then "btrfs device remove" your previously added SSD/NVMe devices.

link
gmokki 489 days ago
Didn't ZFS also have disk corruption on Linux last year?

And a few years before that Ubuntu backports had Ubuntu specific ZFS corruption.

Basically the old truth still holds with checksummed filesystems too: raid is not backup

link