[linux-lvm] Creating snapshot on / hangs machine

[linux-lvm] Creating snapshot on / hangs machine

[Kai Leibrandt]

Hi all,

I am not sure if this should even be possible, but in order to take 
consistent backups I would like to take a snapshot of my / filesystem, and 
then use dump for the backup, but this consistently hangs my machine. My box 
is Fedora Core 2, kernel 2.6.8-1.521, lvm2-2.00.15-2, the filesystems are:

/boot: ext3 on /dev/md0
/: ext3 on /dev/VG00/RootLV on /dev/md1
/export: ext3 on /dev/VG00/ExportLV on /dev/md1
swap: swap on /dev/VG00/SwapLV on /dev/md1

VG00 has 500Mb free, and the snapshots I create are of the right size 
(lvcreate -L500M -s -n BackupLV /dev/VG00/ExportLV for instance). Taking 
snapshots  on /export is fine, and I have done some stress testing on this, 
repeatedly creating snapshots, dumping the snapshot to file, removing 
snapshot, etc. As soon as I create a snapshot on / however, the machine 
hangs, although X still works (well, the mouse still moves), it is no longer 
possible to open a new shell, switch consoles, or even <CTRL>C the lvcreate.

I have tried this with ext2 as well as with xfs, including doing xfs_freeze 
before taking the snapshot, so I don't think it's a filesys related matter.

Is there anything I am missing here? Is it even supported to do / snapshots?

Many thanks for any help,

Kai.

_________________________________________________________________
Express yourself instantly with MSN Messenger! Download today it's FREE! 
http://messenger.msn.com/

Re: [linux-lvm] Creating snapshot on / hangs machine

[Andrew W. Donoho]

[-- Attachment #1: Type: text/plain, Size: 1567 bytes --]


On Oct 14, 2004, at 11:31, Kai Leibrandt wrote:
> As soon as I create a snapshot on / however, the machine hangs, 
> although X still works (well, the mouse still moves), it is no longer 
> possible to open a new shell, switch consoles, or even <CTRL>C the 
> lvcreate.
>
> I have tried this with ext2 as well as with xfs, including doing 
> xfs_freeze before taking the snapshot, so I don't think it's a filesys 
> related matter.
>
> Is there anything I am missing here? Is it even supported to do / 
> snapshots?


Kai,

	I reported a similar problem starting with my root partition when I 
added a second drive. After a bit of skullduggery, I believe that LVM 
activities on the root partition particularly stress the memory system. 
Or, at least, when I slowed the memory system down on my x86-64 
machine, I no longer get crashes like you describe above.

	What convinced me that this might be a problem was when I tried 
compiling a kernel to turn on preemptive scheduling and gcc quit about 
halfway through. The build system listed a comment that when gcc quits 
there is probably a memory problem. After I slowed down the memory 
system, I was able to do a continuous "clean and compile" over night 
with a 3.0+ load average. I have not had any other crashes since then.

Andrew

P.S. I just made these changes two days ago. I was just beginning to 
feel comfortable with my newfound stability. I hope this helps you.

____________________________________
Andrew W. Donoho
awd@DDG.com, PGP Key ID: 0x81D0F250
+1 (512) 453-6652 (o), +1 (512) 750-7596 (m)



[-- Attachment #2: Type: text/enriched, Size: 1617 bytes --]



On Oct 14, 2004, at 11:31, Kai Leibrandt wrote:

<excerpt>As soon as I create a snapshot on / however, the machine
hangs, although X still works (well, the mouse still moves), it is no
longer possible to open a new shell, switch consoles, or even <<CTRL>C
the lvcreate.


I have tried this with ext2 as well as with xfs, including doing
xfs_freeze before taking the snapshot, so I don't think it's a filesys
related matter.


Is there anything I am missing here? Is it even supported to do /
snapshots?

</excerpt>


Kai,


	I reported a similar problem starting with my root partition when I
added a second drive. After a bit of skullduggery, I believe that LVM
activities on the root partition particularly stress the memory
system. Or, at least, when I slowed the memory system down on my
x86-64 machine, I no longer get crashes like you describe above. 


	What convinced me that this might be a problem was when I tried
compiling a kernel to turn on preemptive scheduling and gcc quit about
halfway through. The build system listed a comment that when gcc quits
there is probably a memory problem. After I slowed down the memory
system, I was able to do a continuous "clean and compile" over night
with a 3.0+ load average. I have not had any other crashes since then.


Andrew


P.S. I just made these changes two days ago. I was just beginning to
feel comfortable with my newfound stability. I hope this helps you.


<fontfamily><param>Helvetica</param>____________________________________

Andrew W. Donoho

awd@DDG.com, PGP Key ID: 0x81D0F250

+1 (512) 453-6652 (o), +1 (512) 750-7596 (m)


</fontfamily>

Re: [linux-lvm] Creating snapshot on / hangs machine

[Clint Byrum]

On Thu, 2004-10-14 at 13:16 -0500, Andrew W. Donoho wrote:
> 
> 
> ______________________________________________________________________
> 
> 
> On Oct 14, 2004, at 11:31, Kai Leibrandt wrote:
>         As soon as I create a snapshot on / however, the machine
>         hangs, although X still works (well, the mouse still moves),
>         it is no longer possible to open a new shell, switch consoles,
>         or even <CTRL>C the lvcreate.
>         
>         I have tried this with ext2 as well as with xfs, including
>         doing xfs_freeze before taking the snapshot, so I don't think
>         it's a filesys related matter.
>         
>         Is there anything I am missing here? Is it even supported to
>         do / snapshots?
> 
> 
> Kai,
> 
> I reported a similar problem starting with my root partition when I
> added a second drive. After a bit of skullduggery, I believe that LVM
> activities on the root partition particularly stress the memory
> system. Or, at least, when I slowed the memory system down on my
> x86-64 machine, I no longer get crashes like you describe above. 
> 


Weird.. I thought the problem was that LVM was causing the atime on
the /dev entries to be updated, which required write access, but there
was a lock on the filesystem while the snapshot was created.