From: Duncan <1i5t5.duncan@cox.net>
To: linux-btrfs@vger.kernel.org
Subject: Re: Missing half of available space (resend)
Date: Thu, 10 Dec 2015 03:27:36 +0000 (UTC) [thread overview]
Message-ID: <pan$b5bdb$fb739c20$bb4bafc8$41b6b38d@cox.net> (raw)
In-Reply-To: CAJCQCtSuEQRVsrLim9AD3di7Rxv2xA9FOd872pNXs-CRiFZpcw@mail.gmail.com
Chris Murphy posted on Wed, 09 Dec 2015 12:39:01 -0700 as excerpted:
> On Wed, Dec 9, 2015 at 10:28 AM, David Hampton
> <mailinglists@dhampton.net> wrote:
>> On Wed, 2015-12-09 at 16:48 +0000, Duncan wrote:
>>> David Hampton posted on Wed, 09 Dec 2015 01:30:09 -0500 as excerpted:
>>>
>>> > Seems I need to upgrade my tools. That command was added in 3.18
>>> > and I only have the 3.12 tools.
>>>
>>> Definitely so, especially because you're running raid6, which wasn't
>>> stable until 4.1 for both kernel and userspace. 3.12? I guess it did
>>> have the very basic raid56 support, but it's definitely nothing I'd
>>> trust, at that old not for btrfs in general, but FOR SURE not raid56.
>>
>> I've upgraded to the 4.2.0 kernel and the 4.0 btrfs-tools package.
>
> I think btrfs-progs 4.0 has a mkfs bug in it (or was that 4.0.1?)
Looking at the wiki, it was -progs 4.1.1 that had the mkfs.btrfs bug,
with 4.1.2 being the urgent bug-fix for it. (I just looked that up
myself for a different thread, a couple days ago, so had it fresh and
knew what I was looking for.)
https://btrfs.wiki.kernel.org/index.php/Changelog#btrfs-
progs_4.1.1_.28Jul_2015.29
> Anyway, even that is still old in Btrfs terms. I think Ubuntu needs to
> do better than this, or just acknowledge Btrfs is not supported, don't
> include btrfs-progs at all by default, and stop making it an install
> time option.
Three points related to list vs. distro release version support here:
1) In terms of this list, as a general rule there's currently four
recommended kernel series at any point in time. The latest two series
from either the current or LTS support series. On current, 4.4 is
getting close to release, so 4.3 and 4.2 are generally supported. On LTS,
again 4.4 will be an LTS series, with the previous two LTS series being
4.1 and 3.18, so those are generally supported. But 3.18 is getting a
bit long in the tooth now and 4.4 will be another LTS series, so people
still on it should already be planning their upgrade to at least 4.1.
For userspace (btrfs-progs), daily usage version isn't as critical as the
kernel, but if there's problems you'll probably want the newest, since it
has the latest fixes and the best chance for fixing things. As a general
rule of thumb, however, the kernel and userspace version numbers are
series-synced and developed at the same time with the same issues in
mind, so try to keep userspace version series at least equal to kernel
series, which if you're following kernel recommendations above, will
ensure that userspace doesn't get /too/ outdated and that it's roughly in
sync with your kernel. Newer userspace than kernel is fine. Which means
a 3.18 series userspace at the oldest, as well.
2) While that's the general rule, btrfs raid56 mode was only nominally
complete (earlier versions of raid56 had runtime completion, but recovery
tooling was incomplete, so you were effectively running a slow raid0 in
terms of recovery, that would be updated "for free" when the recovery
tools were complete) with 3.19 and there were major bugs into early 4.1.
So for raid56 mode you really *NEED* at least the latest LTS 4.1 kernel,
and following the userspace rule of thumb, the corresponding latest 4.1
series userspace, 4.1.2, with the bugfix for that mkfs.btrfs bug
mentioned above. That's the absolute minimum I'd consider reasonable
with raid56 mode, as before that, there ARE known bugs and it's simply
not worth the risk running btrfs raid56 mode on older, period.
However, with btrfs raid56 mode completion still so new, it really can't
be considered as stable as btrfs in general, with btrfs itself still
"stabilizing, but not fully stable or mature yet, so be sure you have
backups available if you value your data." I've long suggested that
btrfs raid56 mode would take a year (effectively five kernel series) to
stabilize after nominal code completion, before I'd consider it as stable
as btrfs in general. As it happens, that's 4.4, which is an LTS series
as well, so the timing is pretty sweet.
But that means as a practical matter, people running btrfs raid56 mode
have even more reason to be on the very latest, into the coming 4.4
series at least, because until then, it really can't be considered as
stable as btrfs in general. So while 4.1 is the absolute minimum I'd
consider running with raid56, you /really/ should be on the latest 4.3.x
stable kernel and 4.3.1 btrfs-progs userspace, and upgrade to 4.4 series
reasonably soon after it comes out as well. With 4.4 kernel being an LTS,
at that point you can relax a bit and stay on that series for awhile, if
desired.
3) Ubuntu unfortunately has a way of picking kernel versions to
standardize on, that aren't upstream kernel LTS versions. As a result,
while 3.18 for example, was an upstream LTS, the 3.19 ubuntu chose to
standardize on, was not. Fortunately, they've announced that they'll
standardize on the upcoming 4.4 LTS series for 2016.04, so hopefully
they've seen the light now and will sync a bit better with upstream LTS
than they have been doing.
The less distro-specific version of this point is that while the above
two latest current kernel and lts kernel series (and corresponding rule
of thumb userspace version syncing) recommendations stand for this list,
distros have their own kernel and userspace support rules, and to the
extent that they choose to support btrfs but diverge from the above
general list recommendations, users have a choice, they can either follow
their distro and presumably rely on the distro for primary btrfs support,
or follow list recommendations.
Meanwhile, it's not that the list won't try to support versions not in
sync with list recommendations, we'll certainly try, but that we'll have
a harder time doing so, because as those versions age, they get further
from the versions we're immediately familiar with. So we'll do our best,
but it's not going to be to the level we'd be able to support recommended
versions, and at times the best recommendation will be to upgrade to
something better supported and see what the results are.
As for level of distro support where their recommended versions differ
from those of this list and upstream, many of them do backport patches
from current where they consider appropriate, so in fact they _may_ have
a reasonably stable supported version, outside of upstream's recommended
versions. But the point is, that's their choice and their business, and
people on the list aren't likely to be particularly familiar with what
particular patches a distro has backported for their various releases, so
again, we won't be able to provide the level of support we could with
recommended versions as to some extent the distro packaged releases are
unknown quantities. Which is why in that case primary support really
should come from the distro. As to how well they actually provide it...
well, that's between the distro and its users, and isn't really a matter
this list is directly involved in, as we're btrfs kernel and userspace
upstream.
So in summary:
1) List primary support: Latest two upstream current and LTS kernel
release series, with userspace at least kernel-series-synced if not newer.
2) Btrfs raid56 mode is new and not yet as stable as btrfs in general.
LTS 4.1 series is the absolute minimum acceptable for base support,
current is preferred, thru at least the soon to be release 4.4 LTS, which
should stabilize raid56 to roughly the same level as btrfs in general.
3) Distros have their own release support models and users are free to
choose them for primary support over this list, but where distro support
series differ from those of this list, we'll still try to provide
support, but it'll be degraded compared to that provided to primary
supported release series, and the best list recommendation at times is
very likely to simply be, upgrade to something we can better support.
--
Duncan - List replies preferred. No HTML msgs.
"Every nonfree program has a lord, a master --
and if you use the program, he is your master." Richard Stallman
prev parent reply other threads:[~2015-12-10 3:27 UTC|newest]
Thread overview: 12+ messages / expand[flat|nested] mbox.gz Atom feed top
2015-12-09 5:02 Missing half of available space (resend) David Hampton
2015-12-09 5:27 ` Chris Murphy
2015-12-09 6:30 ` David Hampton
2015-12-09 16:48 ` Duncan
2015-12-09 17:28 ` David Hampton
2015-12-09 19:39 ` Chris Murphy
2015-12-09 19:56 ` Gareth Pye
2015-12-09 21:26 ` David Hampton
2015-12-09 21:28 ` Chris Murphy
2015-12-09 21:41 ` David Hampton
2015-12-09 21:59 ` Gareth Pye
2015-12-10 3:27 ` Duncan [this message]
Reply instructions:
You may reply publicly to this message via plain-text email
using any one of the following methods:
* Save the following mbox file, import it into your mail client,
and reply-to-all from there: mbox
Avoid top-posting and favor interleaved quoting:
https://en.wikipedia.org/wiki/Posting_style#Interleaved_style
* Reply using the --to, --cc, and --in-reply-to
switches of git-send-email(1):
git send-email \
--in-reply-to='pan$b5bdb$fb739c20$bb4bafc8$41b6b38d@cox.net' \
--to=1i5t5.duncan@cox.net \
--cc=linux-btrfs@vger.kernel.org \
/path/to/YOUR_REPLY
https://kernel.org/pub/software/scm/git/docs/git-send-email.html
* If your mail client supports setting the In-Reply-To header
via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line
before the message body.
This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.