What's the status of BTRFS RST instead of linux md raid5?

What's the status of BTRFS RST instead of linux md raid5?

[Marc MERLIN]

So, I'm aware that the built in btrfs raid5 support that's been there
for 10-ish years is not recommended due to hard to fix or unfixable
flaws.

But with raid5, once I have a checksum mismatch, btrfs could easily run
all combinations of the blocks to find out which one can be removed to
yield a correct checksum for that data block, but obviously those layers
don't talk, so nothing happens outside of btrfs telling me the checksum
is bad, and there is no good way to rebuild anything, especially when
it's bit rot in a read only snapshot where I cannot reasonably delete or
rewrite the file without some surgery.

Gemini tells me that RST with the future that fixes this so I can use
raid5 or raid6 directly in btrfs where everything talks to one another,
but I can't find good docs or status on RST.
Even this list has very little about RST outside of some old patches.

So, is RST kind of ready and documented, or is it still very
experimental and "read the source, luke" ?

Thanks,
Marc
-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
 
Home page: http://marc.merlins.org/  

Re: What's the status of BTRFS RST instead of linux md raid5?

[Johannes Thumshirn]

On 4/9/26 1:10 AM, Marc MERLIN wrote:
> So, I'm aware that the built in btrfs raid5 support that's been there
> for 10-ish years is not recommended due to hard to fix or unfixable
> flaws.
>
> But with raid5, once I have a checksum mismatch, btrfs could easily run
> all combinations of the blocks to find out which one can be removed to
> yield a correct checksum for that data block, but obviously those layers
> don't talk, so nothing happens outside of btrfs telling me the checksum
> is bad, and there is no good way to rebuild anything, especially when
> it's bit rot in a read only snapshot where I cannot reasonably delete or
> rewrite the file without some surgery.
>
> Gemini tells me that RST with the future that fixes this so I can use
> raid5 or raid6 directly in btrfs where everything talks to one another,
> but I can't find good docs or status on RST.
> Even this list has very little about RST outside of some old patches.
>
> So, is RST kind of ready and documented, or is it still very
> experimental and "read the source, luke" ?


RST is still considered experimental and the development on the RAID5/6 
support is unfortunately delayed due to different more urgent fires that 
have to be extinguished.

Note while RST can handle non zoned devices, it's primary focus always 
was always to enable RAID on zoned block devices. Which it does for 
RAID0, RAID1 and RAID10.

As for RAID5 I have a prototype doing writes and reads for the trivial 
cases (i.e. full stripe write/read).


My apologies to everyone waiting for this. I haven't abandoned the 
project, I just need more time to get it done.


If you're looking for documentation please have a look here: 
https://github.com/kdave/btrfs-progs/blob/devel/Documentation/dev/design-raid-stripe-tree.txt

Re: What's the status of BTRFS RST instead of linux md raid5?

[Marc MERLIN]

On Mon, Apr 13, 2026 at 01:37:30PM +0200, Johannes Thumshirn wrote:
> 
> RST is still considered experimental and the development on the RAID5/6
> support is unfortunately delayed due to different more urgent fires that
> have to be extinguished.
> 
> Note while RST can handle non zoned devices, it's primary focus always was
> always to enable RAID on zoned block devices. Which it does for RAID0, RAID1
> and RAID10.
> 
> As for RAID5 I have a prototype doing writes and reads for the trivial cases
> (i.e. full stripe write/read).
> 
> 
> My apologies to everyone waiting for this. I haven't abandoned the project,
> I just need more time to get it done.
> 
> If you're looking for documentation please have a look here: https://github.com/kdave/btrfs-progs/blob/devel/Documentation/dev/design-raid-stripe-tree.txt

Thank you so much for the status update, *much* appreciated.
As for when, this is all unpaid work I assume, so it gets done when it
gets done, no worries.

For what it's worth, I've ben doing mdadm raid5 with btrfs since 2014,
and unfortuantely it's failed in too many ways, almost each time causing
a btrfs filesystem that was impossible to recover (usually ro mount was
ok, but impossible to get back in read/write state, which is very
inconvenient for a 20TB+ filesystem dueto how long it takes to restore
backups)

As luck would have it, this just happened to me again
https://yhbt.net/lore/linux-btrfs/advYoKqUCUmD4oO6@merlins.org/

I know I don't know the deep insides of the btrfs code, but it's
unfortunately failed enough times on top of mdadm raid5 that I can't
wait to switch to RST when it's ready.  It's been 12 years already, so I
can wait a bit longer still :)

Marc
-- 
"A mouse is a device used to point at the xterm you want to type in" - A.S.R.
 
Home page: http://marc.merlins.org/                       | PGP 7F55D5F27AAF9D08