SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Jeff Rollin]

Hi lug

>
> Here's hoping the problem is solved, thanks to all (especially Alex)
> for the help.
>
> Jeff.
>

I have successfully compiled a few things, so it looks like the
problem IS solved. Thanks again where due.

Ignore the following if you don't like minirants.

1. Frankly, I'm not impressed with Linux in this case*. /var is not a
"mission critical" filesystem in the sense that if it contains errors,
it can still be mounted and the errors don't necessarily mean the
system won't come up. How about some non-fatal error messages that
tell you you have errors on the disk and then go away - oh, and which
come up in X11? I have a feeling the reboots were caused by Linux
trying to cope with X11 being in control of the screen when the kernel
panicked, since twice when I had a VT displayed I got some error
messages. At worst what should happen in this sort of situation is
that you get brought down to single-user mode and get to repair the
error. As you know if you have read this thread before, the constant
reboots in response to this problem had me going down not only my own
garden path but several else's as too, metaphorically speaking.

*You read it here first (unless you read OSNews, anyway.)

2. In attempting to fix the problem, I installed Ubuntu to see whether
it's gcc would fail in the same manner as the gcc on Gentoo had done.
Much to my surprise, after successfully loading the distro onto my
second hard disk by not being too picky about options, and downloading
source of gcc and tar, I get the dread "gcc cannot create executables"
message when attempting to compile. (This had nothing to do with the
reboot problem I was having, since on Gentoo the compile would start,
get part way through, and the system would reboot. Also, I then loaded
up an old copy of Slackware on the same disk I had put Ubuntu, in the
same hardware configuration, and Slack's gcc worked.) This has not
improved my relationship with Ubuntu.

Jeff

--
Q: What will happen in the Aftermath?

A: Impossible to tell, since we're still in the Beforemath.

http://latedeveloper.org.uk


-- 
Q: What will happen in the Aftermath?

A: Impossible to tell, since we're still in the Beforemath.

http://latedeveloper.org.uk
-- 
gentoo-user@gentoo.org mailing list

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Boyd Stephen Smith Jr.]

[-- Attachment #1: Type: text/plain, Size: 1915 bytes --]

On Wednesday 28 March 2007, "Jeff Rollin" <jeff.rollin@gmail.com> wrote 
about 'SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - 
system reboots while compiling)':
> Ignore the following if you don't like minirants.
(My reply probably needs the same disclaimer.)

> 1. Frankly, I'm not impressed with Linux in this case*. /var is not a
> "mission critical" filesystem in the sense that if it contains errors,
> it can still be mounted and the errors don't necessarily mean the
> system won't come up.

By that definition, no filesystem I can think of is "mission critial", they 
will all withstand some damage and still let your system come up.  /var is 
*at least* as important as /usr -- I can easily recover the contents 
of /usr in case of critical failure, but reconstructing /var is damn near 
impossible.  Also, /usr can generally be very useful with just r/o access, 
while /var needs to be r/w to fill it's role.

Also, forcing a mount of a damaged filesystem is asking for trouble.  
Dangling inodes (or similar) can cause cascading failure; at best some 
processes will read garbage and crash (or, ideally, "magically" recover) 
at worst good data on the disk will be overwritten with bad. File locks on 
a damaged filesystem are meaningless since two files (not simply two 
dirents like with a hard link, but two unrelated files) might share disk 
sectors.

The system should definitely refuse to mount damaged file systems by 
default or *at the very least* mount them read-only.  I wouldn't mind and 
interactive prompt to force mounting a damaged filesystem, but I'd need a 
way to turn that off for unattended systems.

-- 
Boyd Stephen Smith Jr.                     ,= ,-_-. =. 
bss03@volumehost.net                      ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-' 
http://iguanasuicide.org/                      \_/     

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Jeff Rollin]

Hi Boyd.

On 28/03/07, Boyd Stephen Smith Jr. <bss03@volumehost.net> wrote:
> On Wednesday 28 March 2007, "Jeff Rollin" <jeff.rollin@gmail.com> wrote
> about 'SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help -
> system reboots while compiling)':
> > Ignore the following if you don't like minirants.
> (My reply probably needs the same disclaimer.)
>
> > 1. Frankly, I'm not impressed with Linux in this case*. /var is not a
> > "mission critical" filesystem in the sense that if it contains errors,
> > it can still be mounted and the errors don't necessarily mean the
> > system won't come up.
>
> By that definition, no filesystem I can think of is "mission critial", they
> will all withstand some damage and still let your system come up.  /var is
> *at least* as important as /usr -- I can easily recover the contents
> of /usr in case of critical failure, but reconstructing /var is damn near
> impossible.  Also, /usr can generally be very useful with just r/o access,
> while /var needs to be r/w to fill it's role.

That's true.

>
> Also, forcing a mount of a damaged filesystem is asking for trouble.
> Dangling inodes (or similar) can cause cascading failure; at best some
> processes will read garbage and crash (or, ideally, "magically" recover)
> at worst good data on the disk will be overwritten with bad. File locks on
> a damaged filesystem are meaningless since two files (not simply two
> dirents like with a hard link, but two unrelated files) might share disk
> sectors.

It IS a bad idea, but it's not like I "forced" a mount; the system
came up normally and functioned normally until it hit a damaged inode,
whereupon it crashed with nary an indication of what had gone wrong.

>
> The system should definitely refuse to mount damaged file systems by
> default or *at the very least* mount them read-only.

Agreed, definitely.

I wouldn't mind and
> interactive prompt to force mounting a damaged filesystem, but I'd need a
> way to turn that off for unattended systems.
>
> --
> Boyd Stephen Smith Jr.                     ,= ,-_-. =.
> bss03@volumehost.net                      ((_/)o o(\_))
> ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-'
> http://iguanasuicide.org/                      \_/
>
>

Jeff.

-- 
Q: What will happen in the Aftermath?

A: Impossible to tell, since we're still in the Beforemath.

http://latedeveloper.org.uk
-- 
gentoo-user@gentoo.org mailing list

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Boyd Stephen Smith Jr.]

[-- Attachment #1: Type: text/plain, Size: 1855 bytes --]

On Wednesday 28 March 2007, "Jeff Rollin" <jeff.rollin@gmail.com> wrote 
about 'Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - 
system reboots while compiling)':
> > > 1. Frankly, I'm not impressed with Linux in this case*. /var is not
> > > a "mission critical" filesystem in the sense that if it contains
> > > errors, it can still be mounted and the errors don't necessarily
> > > mean the system won't come up.
> >
> > [F]orcing a mount of a damaged filesystem is asking for trouble.
>
> It IS a bad idea, but it's not like I "forced" a mount; the system
> came up normally and functioned normally until it hit a damaged inode,
> whereupon it crashed with nary an indication of what had gone wrong.

Ah, yes, that's a problem.  What filesystem are you using?  I was fairly 
sure ext2/3 tries to detect damage (even while r/w mounted) and force a 
r/o re-mount or unmount.  [Not that that couldn't cause a freeze or 
reboot, but at least it's conservative.]

Reiserfs (and possibly others) is quite stupid, at least in this regard.  
After the filesystem is mounted it performs basically zero sanity checks, 
and always assumes the data provided by the block device is complete and 
accurate.  It can't handle a slowly failing HD, and will almost assuredly 
silently corrupt data on such a device.  This is one of the reasons some 
people strongly recommend against reiserfs.  I still use it, but my 
important data is on RAID6 (underneath LVM), so I can be fairly certain 
the data received by the filesystem is good.

/me is looking for a new favorite file system.

-- 
Boyd Stephen Smith Jr.                     ,= ,-_-. =. 
bss03@volumehost.net                      ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-' 
http://iguanasuicide.org/                      \_/     

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Jeff Rollin]

Hi Boyd

On 28/03/07, Boyd Stephen Smith Jr. <bss03@volumehost.net> wrote:
> On Wednesday 28 March 2007, "Jeff Rollin" <jeff.rollin@gmail.com> wrote
> about 'Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help -
> system reboots while compiling)':
> > > > 1. Frankly, I'm not impressed with Linux in this case*. /var is not
> > > > a "mission critical" filesystem in the sense that if it contains
> > > > errors, it can still be mounted and the errors don't necessarily
> > > > mean the system won't come up.
> > >
> > > [F]orcing a mount of a damaged filesystem is asking for trouble.
> >
> > It IS a bad idea, but it's not like I "forced" a mount; the system
> > came up normally and functioned normally until it hit a damaged inode,
> > whereupon it crashed with nary an indication of what had gone wrong.
>
> Ah, yes, that's a problem.  What filesystem are you using?  I was fairly
> sure ext2/3 tries to detect damage (even while r/w mounted) and force a
> r/o re-mount or unmount.  [Not that that couldn't cause a freeze or
> reboot, but at least it's conservative.]

The filesystem I am currently using on that partition is Reiserfs3, as
you suspected judging from what you say later. However, I am also
currently in the process of migrating all my reiserfs filesystems to
XFS. I have heard that it is unstable on X86 but I have not had any
problems with it (touchwood) to date except insofar as it is only
possible to grow xfs volumes, not shrink them. openSUSE (or is it
Novell?) certainly seem to be less keen on Reiserfs than they used to
be, judging from reports from about the beginning of the year.

>
> Reiserfs (and possibly others) is quite stupid, at least in this regard.
> After the filesystem is mounted it performs basically zero sanity checks,
> and always assumes the data provided by the block device is complete and
> accurate.  It can't handle a slowly failing HD, and will almost assuredly
> silently corrupt data on such a device.  This is one of the reasons some
> people strongly recommend against reiserfs.  I still use it, but my
> important data is on RAID6 (underneath LVM), so I can be fairly certain
> the data received by the filesystem is good.

Yes, for this and other reasons I am moving away from it as I said.
Hopefully this disk is not "slowly failing", but sometimes we get
powercuts here and I suspect the damage occured on one of those
occasions.

>
> /me is looking for a new favorite file system.

I have had no complaints with ext3, but do you have any thoughts on XFS?

Jeff
-- 
Q: What will happen in the Aftermath?

A: Impossible to tell, since we're still in the Beforemath.

http://latedeveloper.org.uk
-- 
gentoo-user@gentoo.org mailing list

[gentoo-user] Re: SOLVED: Recover from LVM errors? (Was: Re: Help - system reboots while compiling)

[Remy Blank]

[-- Attachment #1: Type: text/plain, Size: 121 bytes --]

Boyd Stephen Smith Jr. wrote:
> /me is looking for a new favorite file system.

<troll>
ZFS?
</troll>

-- Remy


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 187 bytes --]

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Alan McKinnon]

On Wednesday 28 March 2007, Jeff Rollin wrote:
> openSUSE (or is it
> Novell?) certainly seem to be less keen on Reiserfs than they used to
> be, judging from reports from about the beginning of the year.

It's unlikely that Novell has issues with the current technical quality 
of reiserfs, which is as good now as it's been for years, and that's 
pretty good.

Much more likely is that Hans Reiser IS NameSys, and with his recent 
troubles there's a big question mark and uncertainty as to whether 
NameSys can deliver on support and maintenance.

Plus, I see very little ongoing effort to get reiser to support enormous 
volume sizes. There's reiser4, but I doubt a decent fsck tool for it 
will ever see the light of day. Meanwhile development on ext4 proceeds, 
and those who need >16TB filesystems will have one shortly.

All in all, the odds are tipping in favour of ext4

alan


-- 
Optimists say the glass is half full,
Pessimists say the glass is half empty,
Developers say wtf is the glass twice as big as it needs to be?

Alan McKinnon
alan at linuxholdings dot co dot za
+27 82, double three seven, one nine three five
-- 
gentoo-user@gentoo.org mailing list

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Boyd Stephen Smith Jr.]

[-- Attachment #1: Type: text/plain, Size: 1433 bytes --]

On Wednesday 28 March 2007, Alan McKinnon <alan@linuxholdings.co.za> wrote 
about 'Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - 
system reboots while compiling)':
> All in all, the odds are tipping in favour of ext4

I don't need quite such large filesystems as my largest is just under 4TB, 
and my system will probably max out around 7TB, but I need a filesystem 
that is maintained (Namesys had basically abandoned reiserfs in favor of 
reiser4 well before Hans' current troubles started), and has good 
all-around performance characteristics (I have both large source trees, 
invloving a multitude of directories and small-ish files AND a 
video "library" containing very large files in my /home).  I would also 
like to see some support for the "tail packing" of resiserfs -- It's not 
that important, but last I checked one saved over 100MB by the portage 
tree on reiserfs AND mini-benchmarks like emerge --sync and 
find '/usr/portage' > /dev/null actually ran faster than ext3.

That said, I'm very encouraged about ext4, and will probably migrate some 
unimportant data over to that filesystem in near future and perform my own 
bonnie++ tests.

-- 
Boyd Stephen Smith Jr.                     ,= ,-_-. =. 
bss03@volumehost.net                      ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-' 
http://iguanasuicide.org/                      \_/     

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Re: SOLVED: Recover from LVM errors? (Was: Re: Help - system reboots while compiling)

[Boyd Stephen Smith Jr.]

[-- Attachment #1: Type: text/plain, Size: 1556 bytes --]

On Wednesday 28 March 2007, Remy Blank <remy.blank@pobox.com> wrote 
about '[gentoo-user]  Re: SOLVED: Recover from LVM errors? (Was: Re: 
Help - system reboots while compiling)':
> Boyd Stephen Smith Jr. wrote:
> > /me is looking for a new favorite file system.
>
> <troll>
> ZFS?
> </troll>

You say troll, I say possibility; I'll certainly consider it.

However, the demos that I've seen about ZFS stress how easy it is to 
administer, and all the LVM-style features it has.  Personally, 
I've /very/ comfortable with LVM and am of the opinion that such features 
don't actually belong at the "filesystem" layer.  [Though, VxFs seems to 
incorporate some of the same features as well.]  So, ZFS doesn't look that 
promising.

I need to good general purpose filesystem, what matters most to be is:
1) Online growing of the filesystem, with LVM I use this a lot, I won't 
consider a filesystem I can't grow while it is in active use.
2) Journaling or other techniques (FFS from the *BSD world does something 
they don't like to call journaling) that reduce the frequency of full 
fscks.
3) All-round performance, and I don't mind it using extra CPU time or 
memory to make filesystem performance better, I have both to spare.
4) Storage savings (like tail packing or transparent compression)

-- 
Boyd Stephen Smith Jr.                     ,= ,-_-. =. 
bss03@volumehost.net                      ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-' 
http://iguanasuicide.org/                      \_/     

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Alan McKinnon]

On Wednesday 28 March 2007, Boyd Stephen Smith Jr. wrote:
> On Wednesday 28 March 2007, Alan McKinnon <alan@linuxholdings.co.za>
> wrote about 'Re: SOLVED: Recover from LVM errors? (Was: Re:
> [gentoo-user] Help -
>
> system reboots while compiling)':
> > All in all, the odds are tipping in favour of ext4
>
> I don't need quite such large filesystems as my largest is just under
> 4TB, and my system will probably max out around 7TB, but I need a
> filesystem that is maintained (Namesys had basically abandoned
> reiserfs in favor of reiser4 well before Hans' current troubles
> started), and has good all-around performance characteristics (I have
> both large source trees, invloving a multitude of directories and
> small-ish files AND a video "library" containing very large files in
> my /home).  I would also like to see some support for the "tail
> packing" of resiserfs -- It's not that important, but last I checked
> one saved over 100MB by the portage tree on reiserfs AND
> mini-benchmarks like emerge --sync and
> find '/usr/portage' > /dev/null actually ran faster than ext3.

In my experience that's a representative data set for a Linux geek :-) 
Mine's very similar and I too find that reiser3 performs better all 
round. Unpacking 35000 smallish files in a kernel tree is no small 
task...

> That said, I'm very encouraged about ext4, and will probably migrate
> some unimportant data over to that filesystem in near future and
> perform my own bonnie++ tests.

How do you plan to get around the decidedly non-trivial task of getting 
a decent fsck on a filesystem where plugins handle the metadata?

Don't get me wrong, I think reiser4 is a good idea, and it's well 
thought out. But everything comes at a price, and in this case it's 
fsck

alan

-- 
Optimists say the glass is half full,
Pessimists say the glass is half empty,
Developers say wtf is the glass twice as big as it needs to be?

Alan McKinnon
alan at linuxholdings dot co dot za
+27 82, double three seven, one nine three five
-- 
gentoo-user@gentoo.org mailing list

[gentoo-user] Re: SOLVED: Recover from LVM errors? (Was: Re: Help - system reboots while compiling)

[Remy Blank]

[-- Attachment #1: Type: text/plain, Size: 1818 bytes --]

Boyd Stephen Smith Jr. wrote:
>> <troll>
>> ZFS?
>> </troll>
> 
> You say troll, I say possibility; I'll certainly consider it.

Actually, I would be very interested in using ZFS for my data.

The "troll" was more about the fact that the ZFS license was explicitly
designed to be GPL-2 incompatible, hence preventing it from being
included into Linux (it would require a clean-room rewrite from the specs).

> However, the demos that I've seen about ZFS stress how easy it is to 
> administer, and all the LVM-style features it has.  Personally, 
> I've /very/ comfortable with LVM and am of the opinion that such features 
> don't actually belong at the "filesystem" layer.

I haven't made the step to LVM and am still using a plain old RAID-1
mirror. I'm not that comfortable adding one more layer to the data path,
and one more difficulty in case of hard disk failure.

> I need to good general purpose filesystem, what matters most to be is:
> 1) Online growing of the filesystem, with LVM I use this a lot, I won't 
> consider a filesystem I can't grow while it is in active use.
> 2) Journaling or other techniques (FFS from the *BSD world does something 
> they don't like to call journaling) that reduce the frequency of full 
> fscks.
> 3) All-round performance, and I don't mind it using extra CPU time or 
> memory to make filesystem performance better, I have both to spare.
> 4) Storage savings (like tail packing or transparent compression)

I completely agree with 1) and 2), and 3) and 4) are nice to haves. What
I like in ZFS is the data integrity check, i.e. every block gets a
checksum, and it can auto-repair in a RAID-Z configuration, something
that RAID-1 cannot.

So I would add:
5) Reliable data integrity checks and self-healing capability.

-- Remy


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Boyd Stephen Smith Jr.]

[-- Attachment #1: Type: text/plain, Size: 1671 bytes --]

On Thursday 29 March 2007 02:19:57 Alan McKinnon wrote:
> On Wednesday 28 March 2007, Boyd Stephen Smith Jr. wrote:
> > That said, I'm very encouraged about ext4, and will probably migrate
> > some unimportant data over to that filesystem in near future and
> > perform my own bonnie++ tests.
>
> How do you plan to get around the decidedly non-trivial task of getting
> a decent fsck on a filesystem where plugins handle the metadata?
>
> Don't get me wrong, I think reiser4 is a good idea, and it's well
> thought out. But everything comes at a price, and in this case it's
> fsck


I think you misread me.  I'm interested in ext4, and disappointed in 
NameSys' 
handling of reiser4.  I love the *idea* of reiser4, but being able the 
resize 
the filesystem is *mandatory* for my setup, and I don't get that with 
reiser4.  However, forward movement on a resizer (and other, currently 
vaporware, filesystem features/utilties) had been completely abandoned to 
the 
effort of getting reiser4 mainlined, well before Hans' legal troubles 
started.  I feel this was/is a mistake; I have no problem running 
mm-sources 
when it has a feature I desire. But with the filesystem as it is I can't 
actually use it for more than testing.

I've heard some (but not enough) about ext4 and it seems promising.  I'm 
keeping my eye on it, and will probably throw some "production" data on it 
before it's mainlined.

-- 
Boyd Stephen Smith Jr.                     ,= ,-_-. =. 
bss03@volumehost.net                      ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-' 
http://iguanasuicide.org/                      \_/     

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

Re: [gentoo-user] Re: SOLVED: Recover from LVM errors? (Was: Re: Help - system reboots while compiling)

[Boyd Stephen Smith Jr.]

[-- Attachment #1: Type: text/plain, Size: 3019 bytes --]

On Thursday 29 March 2007 03:09:33 Remy Blank wrote:
> Boyd Stephen Smith Jr. wrote:
> >> <troll>
> >> ZFS?
> >> </troll>
> >
> > You say troll, I say possibility; I'll certainly consider it.
>
> Actually, I would be very interested in using ZFS for my data.
>
> The "troll" was more about the fact that the ZFS license was explicitly
> designed to be GPL-2 incompatible, hence preventing it from being
> included into Linux (it would require a clean-room rewrite from the 
specs).
>
> > However, the demos that I've seen about ZFS stress how easy it is to
> > administer, and all the LVM-style features it has.  Personally,
> > I've /very/ comfortable with LVM and am of the opinion that such 
features
> > don't actually belong at the "filesystem" layer.
>
> I haven't made the step to LVM and am still using a plain old RAID-1
> mirror. I'm not that comfortable adding one more layer to the data path,
> and one more difficulty in case of hard disk failure.
>
> > I need to good general purpose filesystem, what matters most to be is:
> > 1) Online growing of the filesystem, with LVM I use this a lot, I won't
> > consider a filesystem I can't grow while it is in active use.
> > 2) Journaling or other techniques (FFS from the *BSD world does 
something
> > they don't like to call journaling) that reduce the frequency of full
> > fscks.
> > 3) All-round performance, and I don't mind it using extra CPU time or
> > memory to make filesystem performance better, I have both to spare.
> > 4) Storage savings (like tail packing or transparent compression)
>
> I completely agree with 1) and 2), and 3) and 4) are nice to haves. What
> I like in ZFS is the data integrity check, i.e. every block gets a
> checksum, and it can auto-repair in a RAID-Z configuration, something
> that RAID-1 cannot.

RAID-3?/5/6 can self-repair like this, but the checksumming is done at the 
stripe, rather than inode level.  Since I use HW RAID-6 across 10 drives, 
I'm 
not that concerned with this done at the filesystem level.  Even without 
the 
extra disks, you can use SW RAID across partitions on a single (or small 
number of) disk(s).  [(Ab)uses of SW RAID like this are not something I'd 
always recommend, but can provide the integrity checks you desire.]

Also, EVMS provides a BBR (bad block relocatation) target, that can work 
around isolated disk failures.

> 5) Reliable data integrity checks and self-healing capability.

Overall, I see this as something I'd rather see done at the block device 
level, instead of the filesystem level.  Surely, a filesystem should not 
shy 
away from sanity checks that can be done with little overhead besides CPU 
time, but adding a checksum to each block might be a little overkill.

-- 
Boyd Stephen Smith Jr.                     ,= ,-_-. =. 
bss03@volumehost.net                      ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-' 
http://iguanasuicide.org/                      \_/     

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]

[gentoo-user] Re: SOLVED: Recover from LVM errors? (Was: Re: Help - system reboots while compiling)

[Remy Blank]

[-- Attachment #1: Type: text/plain, Size: 804 bytes --]

Boyd Stephen Smith Jr. wrote:
> RAID-3?/5/6 can self-repair like this, but the checksumming is done at the 
> stripe, rather than inode level. 

AFAIK, RAID-5 doesn't self-heal except for the specific case where a bad
block is detected by the hardware, so the RAID driver knows which drive
has the bad stripe. If the parity is just inconsistent, there is no way
of knowing which drive's stripe should be reconstructed.

RAID-6 OTOH should be able to do that.

> Surely, a filesystem should not shy 
> away from sanity checks that can be done with little overhead besides CPU 
> time, but adding a checksum to each block might be a little overkill.

As long as performance is OK, I am willing to sacrifice the space for
the per-block checksum.

BTW, 10 drives? Nice setup!

-- Remy


[-- Attachment #2: OpenPGP digital signature --]
[-- Type: application/pgp-signature, Size: 189 bytes --]

Re: SOLVED: Recover from LVM errors? (Was: Re: [gentoo-user] Help - system reboots while compiling)

[Hemmann, Volker Armin]

On Donnerstag, 29. März 2007, Boyd Stephen Smith Jr. wrote:

> I think you misread me.  I'm interested in ext4, and disappointed in
> NameSys'
> handling of reiser4.  I love the *idea* of reiser4, but being able the
> resize
> the filesystem is *mandatory* for my setup, and I don't get that with
> reiser4.  However, forward movement on a resizer (and other, currently
> vaporware, filesystem features/utilties) had been completely abandoned to
> the
> effort of getting reiser4 mainlined, well before Hans' legal troubles
> started.  I feel this was/is a mistake; I have no problem running
> mm-sources
> when it has a feature I desire. But with the filesystem as it is I can't
> actually use it for more than testing.

That is your point of view.

For others, 'Resizing' is something that is not needed and never used. But 
being in mainline is mandatory! -mm Kernels are full of experimental stuff, 
highly unstable and very buggy. Nothing you can really trust. So as long as 
something does not show up in the Linus' Kernel, it is not usable. And that 
does not cover the testing something gets in Linus' kernel, that is simply 
not there with -mm kernels.

--
gentoo-user@gentoo.org mailing list

Re: [gentoo-user] Re: SOLVED: Recover from LVM errors? (Was: Re: Help - system reboots while compiling)

[Boyd Stephen Smith Jr.]

[-- Attachment #1: Type: text/plain, Size: 1947 bytes --]

On Thursday 29 March 2007, Remy Blank <remy.blank@pobox.com> wrote 
about '[gentoo-user]  Re: SOLVED: Recover from LVM errors? (Was: Re: 
Help - system reboots while compiling)':
> Boyd Stephen Smith Jr. wrote:
> > Surely, a filesystem should not shy
> > away from sanity checks that can be done with little overhead besides
> > CPU time, but adding a checksum to each block might be a little
> > overkill.
>
> As long as performance is OK, I am willing to sacrifice the space for
> the per-block checksum.

Yeah, if It's a CRC32 that's only 4 bytes out of a 4k block.  .1% space 
overhead is paltry compared to the space cost of RAID-3/5/6.

Even if it's something longer with error correction as well as detection, 
like a Hamming code, I imagine it could be *very* useful.  Most 
checksums/checkdigits [e.g. CRC16/CRC32 or using any Crypo hash as a 
checksum] only do error detection, but the theory behind error correction 
has been around nearly as long, it's just more "expensive".  More layers 
of redundancy are generally a good thing.

> BTW, 10 drives? Nice setup!

The machine's hostname is "monster" for a reason. 2x Dual-Core Opteron 
275s, 2x NVidia 7800GTX (overclocked by BFG), 4G RAM, 10x 500G Hitachi's 
in 2x Chenbro 5-in-3 enclosures (in RAID6 = ~4TB usable space), 2x 74G 
Raptors in software RAID-0, Dell 1905FP + Dell 2407WFP, 7.1 sound, SATA 
DVD+/-RW drive, basically everything I could ever need.  Built it myself 
(well, with the help of my geek friends as well) in the 1st half of 2005, 
although I've added some to it since then (drives and monitor).

There's a pic and blog post about it on the drupal installation @ my 
domain, listed in my .sig.

-- 
Boyd Stephen Smith Jr.                     ,= ,-_-. =. 
bss03@volumehost.net                      ((_/)o o(\_))
ICQ: 514984 YM/AIM: DaTwinkDaddy           `-'(. .)`-' 
http://iguanasuicide.org/                      \_/     

[-- Attachment #2: Type: application/pgp-signature, Size: 189 bytes --]