Btrfs RAID 5/6 Support Is "Mostly OK" With Linux 4.12

[phoronix]

Phoronix: Btrfs RAID 5/6 Support Is "Mostly OK" With Linux 4.12

I previously reported on Btrfs RAID 5/6 fixes for Linux 4.12 to work on fixing some potentially bad Btrfs RAID 5/6 problems. These changes for Linux 4.12 were enough to elevate the rating of this functionality...

http://www.phoronix.com/scan.php?page=news_item&px=Linux-4.12-Btrfs-RAID-Mostly-OK

[torsionbar28]

I'll stick with md raid. It's "fully ok".

[starshipeleven]

I'll stick with md raid. It's "fully ok".

they see me rollin', they hatin'

[nazar-pc]

I'll stick with md raid. It's "fully ok".

You should add "for you". Because md raid is as stupid as regular hardware raid and is not directly comparable to btrfs or zfs in terms of reliability, it is unable to correct any hardware errors. Btrfs 5/6 without all check sums is not very reliable too, but eventually it will be much better anyway.

[torsionbar28]

You should add "for you". Because md raid is as stupid as regular hardware raid and is not directly comparable to btrfs or zfs in terms of reliability, it is unable to correct any hardware errors.

Fully aware of the feature set delta. Features aren't worth much if they eat your data however. I'll take stability & reliability over features any day of the week. Sometimes simpler is better.

Not sure what you're referring to with "unable to correct any hardware errors". If my md raid-1 mirror encounters a hardware read error, it will read the good data from the other mirror member. Then it marks the faulty block as bad, and remaps it elsewhere on the drive. Is that not correcting for a hardware error? sure sounds like it is.

[starshipeleven]

Fully aware of the feature set delta. Features aren't worth much if they eat your data however. I'll take stability & reliability over features any day of the week. Sometimes simpler is better.

More like production-grade is better than in-development.

Not sure what you're referring to with "unable to correct any hardware errors". If my md raid-1 mirror encounters a hardware read error, it will read the good data from the other mirror member. Then it marks the faulty block as bad, and remaps it elsewhere on the drive. Is that not correcting for a hardware error? sure sounds like it is.

md raid and any other RAID responds to errors reported by the hardware itself, if the hardware does not report errors then it's all fine for it.

btrfs move all the responsibility for data integrity from the hardware's firmware ECC features to the OS instead, so it catches silent errors that the hardware let slip through, and is overall more consistent than firmwares.

[rubdos]

You know what... this might actually become my first kernel contribution. Fixing friggin' holes in RAID systems. "Mostly OK" wtf.

[starshipeleven]

"Mostly OK" wtf.

It's an improvement over "unstable".

[caligula]

Only btrfs feature I'm interested in is compression (mounted with -o compress). Software-based RAID isn't for me. If I could shrink F2FS partitions I'd dump EXT4 in a heartbeat on my Raspberry Pi though.

LZO? No you're not, you'd definitely want LZ4 which is about as efficient, yet doesn't limit the throughput (by being too CPU bound) on most hardware. Too bad only squashfs and ZFS support LZ4.

[johnc]

It's an improvement over "unstable".

Isn't that like saying a bucket with two holes is an improvement over a bucket with five holes? Is it really any more usable?