data DUP

data DUP

[Roger Binns]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

Is there any particular reason why I can't use DUP for data?

When I try to set it with balance there is a kernel message:

  btrfs: dup for data is not allowed

The glossary at https://btrfs.wiki.kernel.org/index.php/Glossary says:

  Regular data cannot be assigned DUP level.

It is somewhat baffling that code and documentation exists to prevent this!

My current use case is an older hard drive I am putting backups on.  Since
my data fits in less than half of the drive, and drives get bad sectors
(this has had several it has reallocated) using DUP would be useful.

(I realise making two partitions and RAID-1 with them would work, which
makes the DUP restriction even sillier.)

Roger
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iEYEARECAAYFAlFy978ACgkQmOOfHg372QTCXwCdGuWbRFY8cZoWNFZb1k3mFeZM
nrkAoK9c5KddsJ1R57pAR1Lk89fPSo9Q
=ENfT
-----END PGP SIGNATURE-----

Re: data DUP

[Hugo Mills]

[-- Attachment #1: Type: text/plain, Size: 1259 bytes --]

On Sat, Apr 20, 2013 at 01:17:06PM -0700, Roger Binns wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> Is there any particular reason why I can't use DUP for data?

   Technically, no. Performance is likely to suck if you use
rotational disks, and you may find some SSDs deduplicate blocks,
making it fairly pointless for those devices.

> When I try to set it with balance there is a kernel message:
> 
>   btrfs: dup for data is not allowed

   What kernel and userspace versions are you using? I thought the
restriction had been removed at some point (but possibly I'm just
misremembering it).

> The glossary at https://btrfs.wiki.kernel.org/index.php/Glossary says:
> 
>   Regular data cannot be assigned DUP level.
> 
> It is somewhat baffling that code and documentation exists to prevent this!

   I think it did get fixed at some point. The documentation doesn't
always get updated everywhere -- it can be quite hard to find every
single implication of a feature when that feature gets changd/updated.

   Hugo.

-- 
=== Hugo Mills: hugo@... carfax.org.uk | darksatanic.net | lug.org.uk ===
  PGP key: 65E74AC0 from wwwkeys.eu.pgp.net or http://www.carfax.org.uk
        --- UNIX: Spanish manufacturer of fire extinguishers. ---        

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: data DUP

[Roger Binns]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 20/04/13 13:48, Hugo Mills wrote:> On Sat, Apr 20, 2013 at 01:17:06PM
- -0700, Roger Binns wrote:
>> Is there any particular reason why I can't use DUP for data?
> 
> Technically, no. Performance is likely to suck if you use rotational
> disks, and you may find some SSDs deduplicate blocks, making it fairly
> pointless for those devices.

Wanting dup is because I care about resilience in the face of errors over
performance, so I don't actually care how bad performance sucks.  I only
put data on SSDs that I can afford to lose (usually backed up to
Dropbox/github/spinning disks).

>> When I try to set it with balance there is a kernel message:
>> 
>> btrfs: dup for data is not allowed
> 
> What kernel and userspace versions are you using? I thought the 
> restriction had been removed at some point (but possibly I'm just 
> misremembering it).

Whatever Ubuntu 12.10 ships with.  Kernel package is 3.5.0.27.43 and
btrfs-tools is 0.19+20120328-7ubuntu1.  Note the message came from the
kernel so it would appear to be solely to blame for refusing my request.

> it can be quite hard to find every single implication of a feature when
> that feature gets changd/updated.

I'm more amused that someone went to the trouble of putting in kernel
detection and messages plus updating the documentation in order to prevent
using DUP for data!

Roger
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iEYEARECAAYFAlFzBWEACgkQmOOfHg372QT5twCeI5KOgHYT1wcPNPtni9TYKQaG
aeIAmQGw/nL/ziTDAu4gUJIFPwCLI5td
=ABih
-----END PGP SIGNATURE-----

Re: data DUP

[Hugo Mills]

[-- Attachment #1: Type: text/plain, Size: 1743 bytes --]

On Sat, Apr 20, 2013 at 02:15:14PM -0700, Roger Binns wrote:
> -----BEGIN PGP SIGNED MESSAGE-----
> Hash: SHA1
> 
> On 20/04/13 13:48, Hugo Mills wrote:> On Sat, Apr 20, 2013 at 01:17:06PM
> - -0700, Roger Binns wrote:
> >> When I try to set it with balance there is a kernel message:
> >> 
> >> btrfs: dup for data is not allowed
> > 
> > What kernel and userspace versions are you using? I thought the 
> > restriction had been removed at some point (but possibly I'm just 
> > misremembering it).
> 
> Whatever Ubuntu 12.10 ships with.  Kernel package is 3.5.0.27.43 and
> btrfs-tools is 0.19+20120328-7ubuntu1.  Note the message came from the
> kernel so it would appear to be solely to blame for refusing my request.

   OK, I'm not 100% certain, but that's probably too early for the fix
I was thinking of. You should upgrade anyway -- there's been a number
of serious bugs in btrfs fixed since then. Ubuntu has new kernels in a
PPA at [1].

> > it can be quite hard to find every single implication of a feature when
> > that feature gets changd/updated.
> 
> I'm more amused that someone went to the trouble of putting in kernel
> detection and messages plus updating the documentation in order to prevent
> using DUP for data!

   It'd have been a mere couple of lines of code, and probably there to 
match historical usage of mkfs (which I believe has also been dealt with).

   Hugo.

[1] https://launchpad.net/~yofel/+archive/btrfs

-- 
=== Hugo Mills: hugo@... carfax.org.uk | darksatanic.net | lug.org.uk ===
  PGP key: 65E74AC0 from wwwkeys.eu.pgp.net or http://www.carfax.org.uk
          --- A linked list is still a binary tree.  Just a ---          
                          very unbalanced one.                           

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 828 bytes --]

Re: data DUP

[Roger Binns]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 20/04/13 14:23, Hugo Mills wrote:
> You should upgrade anyway -- there's been a number of serious bugs in
> btrfs fixed since then.

13.04 is imminent so I'll pick up a newer kernel as part of that anyway.
(Also Tanglu which I hope to move to intends to use the Ubuntu kernel anyway.)

In any event I am not worried.  Bug fixes get backported. The probability
of hitting any serious bug is low (others would likely be victims first)
and worst case I have backups (snapshots, other machines, Google/Dropbox,
DVDs & hard drives at other people's houses etc).

There are two major reasons I switched from ext4.  The first is that
everything is online, including adding and removing devices, checking data
integrity etc.

The second is that data not silently lost.  I had some bad sectors develop
on ext4 spinning disk and the only way to properly recover was to do
offline checks that would have taken ~ 24 hours!  Finding out which
filenames were involved was far too much effort.

My philosophy is that we have machines with billions of processor cycles
per second.  They can figure things out for themselves without requiring
me to baby sit them!

I also run btrfs in a variety of configurations - raid0, raid1, sata, usb,
hdd, ssd, single device, multi-device, bare, dmcrypt, machines on all the
time, laptop, frequent suspend/resume, frequent power on/off.  I've never
experienced any problems with btrfs and use scrub for reassurance.

Roger

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iEYEARECAAYFAlFzHkwACgkQmOOfHg372QSBBwCfYYJQm9l9ts9eWOvZIRUlTCWJ
KsMAoLVsd1fRdV+T7KO7nVVGuFCGYN5a
=Y4by
-----END PGP SIGNATURE-----

Re: data DUP

[Calvin Walton]

[-- Attachment #1: Type: text/plain, Size: 1235 bytes --]

On Sat, 2013-04-20 at 16:01 -0700, Roger Binns wrote:
> On 20/04/13 14:23, Hugo Mills wrote:
> > You should upgrade anyway -- there's been a number of serious bugs in
> > btrfs fixed since then.
> 
> 13.04 is imminent so I'll pick up a newer kernel as part of that anyway.
> (Also Tanglu which I hope to move to intends to use the Ubuntu kernel anyway.)
> 
> In any event I am not worried.  Bug fixes get backported. The probability
> of hitting any serious bug is low (others would likely be victims first)
> and worst case I have backups (snapshots, other machines, Google/Dropbox,
> DVDs & hard drives at other people's houses etc).

Unfortunately, bugfixes in btrfs have tended to be *not* backported;
aside from a few special cases, btrfs changes only go into new kernel
3.x releases. A few of the Linux distributions that officially support
btrfs (Possibly Oracle, Red Hat, Suse? You would have to contact them
for details...) may be backporting fixes themselves.

As a result, we recommend that btrfs users should generally run whatever
the latest upstream kernel release is - in particular, you should try
the latest kernel before reporting bugs in btrfs.

-- 
Calvin Walton <calvin.walton@kepstin.ca>

[-- Attachment #2: smime.p7s --]
[-- Type: application/x-pkcs7-signature, Size: 4027 bytes --]

Re: data DUP

[Roger Binns]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 27/04/13 14:40, Calvin Walton wrote:
> Unfortunately, bugfixes in btrfs have tended to be *not* backported; 
> aside from a few special cases, ...

Your efforts to scare me are admirable, but have failed :-)

As btrfs development has progressed, the probability of a random user like
me hitting bugs keeps decreasing.  This is a reflection of the maturity of
the code base, the increasing number of users, improved test suites, more
eyes on the code, more diversity in uses etc. As far as I can see,
backported bugfixes are made when the probability of a bug being
encountered is significantly higher than the current probabilities, and
that is why they are rare.

As for the severity of the rarely hit bugs, the COW nature of the data
means there is unlikely to be corruption, and if there is then of the most
recent activity.  Additionally the checksums mean it is possible to
proactively verify (online) that unexpected corruption hasn't been
creeping in.

And if all that fails, I have multiple layers of backups.

Roger

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.11 (GNU/Linux)

iEYEARECAAYFAlF8X14ACgkQmOOfHg372QT5EQCfSbzZBRdm1KyljpoNpP5kHy15
iFoAn2eyair54JQdhnL2dfvqAJrVMWe4
=9zYA
-----END PGP SIGNATURE-----

Re: data DUP

[Alex Elsayed]

Roger Binns wrote:

> On 27/04/13 14:40, Calvin Walton wrote:
>> Unfortunately, bugfixes in btrfs have tended to be *not* backported;
>> aside from a few special cases, ...
> 
> Your efforts to scare me are admirable, but have failed :-)
> 
> As btrfs development has progressed, the probability of a random user like
> me hitting bugs keeps decreasing.  This is a reflection of the maturity of
> the code base, the increasing number of users, improved test suites, more
> eyes on the code, more diversity in uses etc. As far as I can see,
> backported bugfixes are made when the probability of a bug being
> encountered is significantly higher than the current probabilities, and
> that is why they are rare.
> 
> As for the severity of the rarely hit bugs, the COW nature of the data
> means there is unlikely to be corruption, and if there is then of the most
> recent activity.  Additionally the checksums mean it is possible to
> proactively verify (online) that unexpected corruption hasn't been
> creeping in.
> 
> And if all that fails, I have multiple layers of backups.

I would be *very* hard-pressed to see it as an attempt to scare you - he's 
just saying what is (very consistently) said to others on this list: When 
using btrfs, run a recent kernel :P.

Honestly, even leaving aside the lack of backporting, there are other 
benefits to a recent kernel - things like cross-subvolume reflinks, btrfs 
device replace support being far more efficient than 
add/balance/remove/balance, and a bunch more.

Re: data DUP

[Roger Binns]

-----BEGIN PGP SIGNED MESSAGE-----
Hash: SHA1

On 27/04/13 19:53, Alex Elsayed wrote:
> When using btrfs, run a recent kernel :P.

Every software developer says that of what they produce.  Newer is almost
always better in many different axes.

> Honestly, even leaving aside the lack of backporting, there are other 
> benefits to a recent kernel - things like cross-subvolume reflinks,
> btrfs device replace support being far more efficient than 
> add/balance/remove/balance, and a bunch more.

Those are all "features", none of which I use or have had to use yet.

If it will make you feel better I did upgrade some systems today to the
most recent Ubuntu release which meant going from kernel 3.5 to 3.8.

Roger
-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.12 (GNU/Linux)

iEYEARECAAYFAlF83dUACgkQmOOfHg372QQeDACgz0oBnrYeg6fO5tFxUy9qonE9
HYIAoJWjT8z2sJ356YAph1NAyLKhcEBz
=ySAg
-----END PGP SIGNATURE-----

Re: data DUP

[David Sterba]

On Sat, Apr 20, 2013 at 01:17:06PM -0700, Roger Binns wrote:
> Is there any particular reason why I can't use DUP for data?

IIRC separate DUP for metadata and data is not possible due to some
technical reason (I don't remember nor can find the mail where this was
mentioned).

DUP data is possible only in the --mixed mode so this

 mkfs.btrfs --mixed -d dup -m dup /dev/ice

works.

david