News for hackers. Stuff that matters.
Re:Debian 11 by nerdponx on Saturday August 14, 2021 @11:12AM (#28181257)
Is it that bad? I have been using Tumbleweed as my desktop OS for several months, but I don't really use the Btrfs features at all.
Re:Debian 11 by matmatmatmat on Saturday August 14, 2021 @01:12PM (#28182264)
Plenty of people, including me, have never lost a single bit to btrfs.

Now, is it super fast? (No.) Is it (or any other COW FS) the right choice for an SSD or database? (Probably not.) Is it the right choice for data that get read much more often than written and you'd like to be sure 10 years from now that you haven't lost any of it? (There's a pretty good argument to be made there, I think.)

Re:Debian 11 by ComputerGuru on Saturday August 14, 2021 @02:27PM (#28182876)
I’ve lost* data with just RAID1 thanks to btrfs bugs post-1.0 that corrupted the entire metadata on both disks. There was no good place to get support nor any instructions on attempting reconstruction of corrupted metadata at the time and I haven’t bothered with it since. Apparently I wasn't the only one that suffered such a loss and as I recall, it was blamed on OS-integrated CoW under certain circumstances but it was quite shortly after adopting it and not a particularly weird configuration so I swore it off and have been happily btrfs-free since.

I should have known better since during my initial evaluation in search of a better llvm for Linux, I set up a root non-raid btrfs volume comprised of multiple dissimilar disks and lost all the data after an unsafe shutdown (a kernel panic that may have been caused by btrfs in the first place) even though all the disks were still functioning fine. I was an early adopter of ZFS - first under OpenSolaris, then under OpenIndiana, then (and now) under FreeBSD, so I thought I understand what "initial stable release" meant but it is clear that what ZFS devs consider to be stable and what btrfs devs consider to be stable are leagues apart.

* I was able to use forensic tools and low-level fs-agnostic recovery methodologies to get some of the important stuff back, but the btrfs volumes were completely lost.

Re:Debian 11 by Dylan16807 on Saturday August 14, 2021 @09:28PM (#28185456)
Why would CoW+SSD be bad?
Re:Debian 11 by matmatmatmat on Sunday August 15, 2021 @12:08AM (#28186191)
COW filesystems have more write amplification than non-COW FS's, which will wear out the SSD more quickly.
Re:Debian 11 by Dylan16807 on Sunday August 15, 2021 @02:02AM (#28186551)
They'll have some write amplification at the filesystem level but should cause less amplification at the drive level.

Either way very few workloads get anywhere near wearing out an SSD, and the upsides of CoW features are almost always much higher than the risk of wearing out a drive. I'd say they fit just fine on an SSD.

Re:Debian 11 by alyandon on Saturday August 14, 2021 @11:44AM (#28181552)
There are features of btrfs that are currently considered experimental/unsafe to use like their raid-5 implementation.

I personally use btrfs raid-1 setups and have survived actual device failures without data loss. However, I also perform regular backups so I'm not overly concerned about "eat my data" bugs in a filesystem either.

Re:Debian 11 by 8fingerlouie on Monday August 16, 2021 @03:35AM (#28195792)
I was under the impression that the data eating RAID5/6 issues were patched "long ago", but due to the write hole issue, which isn't likely to get fixed anytime soon, which means your data is (probably) safe, except for whatever was being written during the "write hole", and your array may crash and become read-only.

The kernel wiki says the following :

    RAID56
    
    Some fixes went to 4.12, namely scrub and auto-repair 
    fixes. Feature marked as mostly OK for now.

    Further fixes to raid56 related code are applied each 
    release. The write hole is the last missing part, 
    preliminary patches have been posted but needed to be 
    reworked. The parity not checksummed note has been 
    removed.
Re:Debian 11 by alyandon on Wednesday August 18, 2021 @11:08AM (#28222116)
Then that's certainly some progress!