Btrfs has been deprecated?

Posted Aug 1, 2017 19:43 UTC (Tue) by dany (guest, #18902)
In reply to: Btrfs has been deprecated? by ubhofmann
Parent article: Red Hat Enterprise Linux 7.4 released

From experience (btrfs+directIO), I bet on data loss/corruption scenarios. Can be something else of course.

Btrfs has been deprecated?

Posted Aug 1, 2017 19:54 UTC (Tue) by jhoblitt (subscriber, #77733) [Link] (3 responses)

In the interim xfs has become a first class citizen, moving from a rhel add-on to supported in the base distro. I presume xfs has quite a large installation base (I have certainly used it heavily in the past for large filesystems) among rhel customers. It might be as simple as limiting the number of supported filesystems.

Btrfs has been deprecated?

Posted Aug 1, 2017 20:01 UTC (Tue) by Paf (subscriber, #91811) [Link]

More than that, XFS is currently the default option for the root filesystem and the other created-by-default file systems/partitions.

Btrfs has been deprecated?

Posted Aug 2, 2017 9:50 UTC (Wed) by danieldk (subscriber, #27876) [Link] (1 responses)

And AFAIK they are adding support for CoW:

https://lkml.org/lkml/2016/10/12/176

Btrfs has been deprecated?

Posted Aug 2, 2017 20:53 UTC (Wed) by nix (subscriber, #2304) [Link]

reflinking via CoW's been working well for a while now, since 4.9 or thereabouts.

Btrfs has been deprecated?

Posted Aug 1, 2017 22:51 UTC (Tue) by flussence (guest, #85566) [Link] (10 responses)

More specifically, I'd guess it's due to Btrfs' RAID being unfinished and dangerous for several years: https://btrfs.wiki.kernel.org/index.php/Status

It's an okay FS on my desktop, and I guess it works for Facebook where they can afford to do RAID6 at the building level, but this sucks hard for anyone wanting to use it on a normal-sized server.

Btrfs has been deprecated?

Posted Aug 1, 2017 23:15 UTC (Tue) by neilbrown (subscriber, #359) [Link] (8 responses)

> I'd guess it's due to Btrfs' RAID being unfinished and dangerous for several years:

That may be a good reason to disable btrfs RAID support. It doesn't seem like a good reason to dump btrfs altogether. If you want RAID6, use md/raid6 (or "hardware RAID") underneath btrfs.

Btrfs has been deprecated?

Posted Aug 2, 2017 2:18 UTC (Wed) by smckay (guest, #103253) [Link]

How many btrfs devs does Red Hat employ? How many xfs devs? How many Red Hat customers are using xfs vs. btrfs, and what customer delta would they get from promoting btrfs to supported? In short, what's the cost/benefit on including btrfs in the set of configurations they support?

I know the answer to exactly zero of these questions but I suspect their answers are relevant to RH's decision to dump btrfs.

Btrfs has been deprecated?

Posted Aug 2, 2017 21:25 UTC (Wed) by wx (guest, #103979) [Link] (6 responses)

> If you want RAID6, use md/raid6 (or "hardware RAID") underneath btrfs.

No, don't do that.

In an md raid, a disk failure will cause the contents of the _entire_ disk to be resynced as md does not differentiate between used and unused disk space - only the filesystem one layer up has that info. With today's disks a resync takes ages and puts a lot of stress on the other disks in the array significantly increasing the risk of additional drive failures in the worst possible situation: in an already degraded state.

Customers who understand and care about these issues are already using ZFS (not necessarily under Linux) and have no reason to switch. At the end of the day there's just no business case for Red Hat in btrfs besides catering to customers who absolutely insist on using Linux but are afraid of legal trouble wrt the CDDL (which Canonical decided would not happen a while back and which hasn't happened to Canonical yet, at least not on public record).

Btrfs has been deprecated?

Posted Aug 2, 2017 23:37 UTC (Wed) by quotemstr (subscriber, #45331) [Link] (4 responses)

Discard ought to be provide exactly the right information to md, right?

Btrfs has been deprecated?

Posted Aug 3, 2017 8:30 UTC (Thu) by Wol (subscriber, #4433) [Link] (3 responses)

Sorry, I don't get it.

md-raid works at the partition level. It knows nothing, cares nothing, about the file system above it so even if you sent a discard to md it wouldn't have a clue what to do with it.

Cheers,
Wol

Btrfs has been deprecated?

Posted Aug 8, 2017 15:39 UTC (Tue) by meuh (guest, #22042) [Link] (2 responses)

FS -> (DM ->) MD -> HDD

If FS initiate TRIM request (discard), beside forwarding the request to HDD, MD can record that one or more stripe are not storing useful information any more (such flag would be invalided by a write on the stripe). Going to the status of each stripes during recover, MD can process only the useful part of the RAID array.

Btrfs has been deprecated?

Posted Aug 9, 2017 7:45 UTC (Wed) by matthias (subscriber, #94967) [Link] (1 responses)

This would work for large TRIM requests. However, TRIM requests are at block granularity. To properly track, which stripes have valid data, MD would need to store for each block whether it has valid data. Otherwise, it cannot tell whether some TRIM request has removed the last used block from a stripe.

Btrfs has been deprecated?

Posted Aug 13, 2017 14:47 UTC (Sun) by Wol (subscriber, #4433) [Link]

And, not being a storage device, MD does not see this as its responsibility.

Yes, it might be a nice addition, but it's added complexity, and it would need somebody to step up to the plate and implement it (plus it would require mucking about with the layout - there may or may not be somewhere to store the fact that stripes are empty or in use).

Cheers,
Wol

Btrfs has been deprecated?

Posted Aug 3, 2017 8:29 UTC (Thu) by Wol (subscriber, #4433) [Link]

So how on earth does ZFS get around that issue? And anyway, no a mdraid does not necessarily cause that problem at all! Yes I would certainly advise against using raid-5 for a large (5 or more) disks array, because losing a disk puts the array in a state of no redundancy.

But firstly if the failed disk is even partially recoverable, that will massively reduce the stress on the array. (As always, we need developers - getting that to work seamlessly is on the hit list of jobs needing doing :-(

Secondly, if you've been scrubbing the array like you're supposed to, you know very well that the array is good and will have no trouble coping with a rebuild.

And thirdly, how on earth is ZFS supposed to know what is on the failed disk, other than by doing a scan of the rest of the disks, stressing them in exactly the same way as md-raid!?

(And yes, md-raid does scan the entire underlying partition like you suggest, but there are easy - if controversial - ways round that.)

Cheers,
Wol

Btrfs has been deprecated?

Posted Aug 21, 2017 17:05 UTC (Mon) by anton (subscriber, #25547) [Link]

We recently played around with Btrfs and RAID1, and what we saw looked somewhat rough (it's relatively easy to get the filesystem into an unrecoverable read-only state; then you have to copy the data over), but we did not experience data loss. You can read about our experience here.