Mailinglist Archive: opensuse-factory (586 mails)

< Previous Next >
Re: [opensuse-factory] Problem with BTRFS in openSUSE Tumbleweed: No space left on device
On 8/30/16 4:35 AM, Richard Brown wrote:
On 30 August 2016 at 04:25, Jeff Mahoney <jeffm@xxxxxxxx> wrote:
On 8/29/16 5:42 PM, Lindsay Mathieson wrote:
On 30/08/2016 1:56 AM, Ronan Arraes Jardim Chagas wrote:
I just have the problem again. Now, it happens during the lunch time
when the machine was idle. Only the system processes were running. It
was not the first time that I saw this problem just after lunch when
the machine stayed idle for a long period (+- 1h). This time, only the
reboot worked, I could not run balance.

Given the debacle over RAID 5/6 and the ongoing issues with stability,
and supposedly a very messy codebase, is btrfs still a wise default for
installs?

Those are three separate issues. The first is, from our (SUSE)
perspective, isn't an issue. We don't support RAID 5/6, full stop. We
don't support device replace either, since those features aren't ready
for prime time. Those "not supported" rules would apply to openSUSE
too, but when I asked about adding the "allow_unsupported" option to
openSUSE years ago, we never reached a consensus.

Further to this point - anyone relying on any software RAID5/6
solution, or any hardware solution without either an NV cache or
battery backup, is fundamentally stupid and just asking to lose data.

btrfs' RAID 5/6 implementation may not be one of the best ones out
there, but all of them put data at risk thanks to the wonders of the
write hole.

Jeff, with Leap being a lot more aligned with SLE, maybe it's time to
resurrect the idea of implementing the same "not supported" rules
there? I would fully support, without hestitation, "allow_unsupported"
on Leap. On Tumbleweed I think there is room for more discussion, but
even there I think it's best for openSUSE and btrfs to reflect the
'sane' and 'safe' features by default - if people want to go off the
deepend, of course they still can.

Leap 42.2, being branched from the SLE12 SP2 kernel, does have the
allow_unsupported flag already.

As for Tumbleweed, I'd have no problem using it there too. I proposed
it in 2013 but, as I said, we never reached an agreement on it.

The allow_unsupported flag is what allows us to support btrfs as our
default file system. Are there shaky parts of btrfs? Absolutely.
That's why this flag exists -- so our users don't stumble into them by
accident. The fact that people are trying and/or even pointing to
raid5/6 as a problem for btrfs is a surprise to me simply because we
just don't allow it on SLES.

Late last night I proposed to some of the core btrfs developers that we
add a similar flag to the mainline kernel. The flag would have a
different name and possibly be per-mount instead of system-wide, but the
idea would be generally the same: Make it clear when a feature is
experimental and require explicit user action to enable it.

-Jeff

--
Jeff Mahoney
SUSE Labs

< Previous Next >