Btrfs Gets A Big Improvement For More Robust RAID1 In Linux 5.5

[Charlie68]

Yup. Here's another one who's actually using btrfs on 6xSSD array and on 5xHDD array.

waxhead, you forgot btrfs-RAID10.

I haven't forgotten, I just don't need it!

[m132]

does it still use its laughable

PID % Number_of_drives

instead of balancing the reads between all drives like ZFS?

[pal666]

I'm not aware that btrfs deals with drive failures automatically. Any quoto for this?

scrub automatically repairs corrupted blocks, i'm not sure how it will handle drive failure though. in any case my explanation of purpose of raif6 still stands even in case of manually triggered repair

[AndyChow]

I'm not aware that btrfs deals with drive failures automatically. Any quoto for this? As far as I know, if a drive fails, you just loose a copy, so instead of RAID1C3 you would end up with a normal RAID1... So time to replace the failed drive with a new one!

Yeah, it doesn't. You need to add a new drive, then remove the failed one. But, hey, this might have changed, but good luck doing it in practice. Because last time I had a btrfs raid1 have a disk failure, I couldn't mount the array back. So I couldn't even add a new disk, let alone remove the failed one.

One of the devs of the projects did a good analysis of the problem, took a couple hours of his time to help me out and refused payment for his troubles, but something like the leaf of the last tree of all the last available snapshots was corrupted, so I could not recover a single file.

But of course I had a few backups in ext4 with full checksum metadata. A few month later, some kernel version silently corrupted many files on my ext4 live and backup arrays. That whole blk-mq bug.

tl/dr: always keep multiple copies of important data on different drives, filesystems, and even geographical locations if you can afford it. And keep checksum lists, because when it hits the fan, you need to be sure that if you have multiple copies of the same files, but they don't have the same checksum, you can figure out which copy is the one that doesn't have bit rot.

[foobaz]

What are the pros and cons of Btrfs RAID10 as compared to RAID1? One difference is that RAID10 requires at least 4 disks, instead of 2, so that's a drawback. I imagine the benefit is speed, right? RAID10 must be faster than RAID1 on the same disks. If I have four or more disks, is there any reason to use RAID1 instead of RAID10? Is either one more resilient than the other or do they have the exact same failure cases?

[Zucca]

What are the pros and cons of Btrfs RAID10 as compared to RAID1? One difference is that RAID10 requires at least 4 disks, instead of 2, so that's a drawback. I imagine the benefit is speed, right? RAID10 must be faster than RAID1 on the same disks. If I have four or more disks, is there any reason to use RAID1 instead of RAID10? Is either one more resilient than the other or do they have the exact same failure cases?

With btrfs you can change the layout on-the-fly from RAID1 to RAID10 for example... as long as you have enough space. So you can always test out which is best for you.
Although it seems that there isn't that much difference.
See this also: https://www.phoronix.com/scan.php?pa...nux55-ssd-raid

[foobaz]

Hey thanks for the good info. It looks like overall, RAID 10 is a minor performance benefit over RAID 1, but only if you have exactly 4 disks. I need that many disks anyway to get the amount of storage I need, so I think I'll give it a shot.

[AndyChow]

Make sure to have backups because it's a new feature and could still be buggy

That's what I love about btrfs. Here you enable higher redundancy, and I half expect it to destroy itself even more spectacularly.