moving logical volumes to another system remotely - how?

linux-lvm Tomasz Chmielewski 5 participants 7 messages Feb 23, 2007 – Mar 2, 2007

Share & tools

Tomasz Chmielewski

author

2007-02-23 11:48 UTC #0 lore raw

I have a server which stores several LVM-2 logical volumes.

As this system is pretty loaded, I'd like to move some of the logical volumes to another machine. It has to be done remotely, so I can't do it as described in the LVM HOWTO (where one basically adds/replaces disks in one machine).

My common sense tells me that I should:

1. Unmount/not use the logical volumes on the source server
2. Make volumes of the same size on the target server
3. Copy it somehow over network

I'm not sure of 2 (make volumes of exactly the same size) and 3 (how to copy it all over network, if possible, using SSH only).

Signature · 2 lines

Tomasz Chmielewski
http://wpkg.org

[email protected]

2007-02-23 12:44 UTC #1 lore raw

Quoted 7 lines · Tomasz Chmielewski

I have a server which stores several LVM-2 logical volumes.
As this system is pretty loaded, I'd like to move some of the logical volumes to another machine. It has to be done remotely, so I can't do it as described in the LVM HOWTO (where one basically adds/replaces disks in one machine).

Quoted 8 lines

My common sense tells me that I should:
1. Unmount/not use the logical volumes on the source server
2. Make volumes of the same size on the target server
3. Copy it somehow over network
I'm not sure of 2 (make volumes of exactly the same size)

check your PE size, otherwise does what it says on the tin.

Quoted 2 lines

and 3 (how to
copy it all over network, if possible, using SSH only).

netcat and dd over forwarded port

Regards,
Paddy

Tomasz Chmielewski

author

2007-02-23 15:59 UTC #2 lore raw

Quoted 9 lines · [email protected] schrieb:

Quoted 8 lines · Tomasz Chmielewski
I have a server which stores several LVM-2 logical volumes.
As this system is pretty loaded, I'd like to move some of the logical volumes to another machine. It has to be done remotely, so I can't do it as described in the LVM HOWTO (where one basically adds/replaces disks in one machine).
My common sense tells me that I should:

Quoted 8 lines

Quoted 6 lines
1. Unmount/not use the logical volumes on the source server
2. Make volumes of the same size on the target server
3. Copy it somehow over network
I'm not sure of 2 (make volumes of exactly the same size)
check your PE size, otherwise does what it says on the tin.

Hmm, how? I just want to move some (not all) logical volumes.

I guess fdisk is a good idea?

# fdisk -l /dev/LVM2/ocsi1

Disk /dev/LVM2/ocsi1: 3221 MB, 3221225472 bytes

3221225472 / 1024 = 3145728

lvcreate -L3145728k -n ocsi1 LVM2

Hmm, hopefully, it's the right size?

Quoted 3 lines

Quoted 1 line
and 3 (how to copy it all over network, if possible, using SSH only).
netcat and dd over forwarded port

Thanks for the idea.

Signature · 2 lines

Tomasz Chmielewski
http://wpkg.org

Lars Ellenberg

2007-02-23 17:32 UTC #3 lore raw

/ 2007-02-23 16:59:23 +0100
\ Tomasz Chmielewski:

Quoted 27 lines

Quoted 9 lines · [email protected] schrieb:
Quoted 8 lines · Tomasz Chmielewski
I have a server which stores several LVM-2 logical volumes.
As this system is pretty loaded, I'd like to move some of the logical volumes to another machine. It has to be done remotely, so I can't do it as described in the LVM HOWTO (where one basically adds/replaces disks in one machine).
My common sense tells me that I should:
Quoted 7 lines
Quoted 6 lines
1. Unmount/not use the logical volumes on the source server
2. Make volumes of the same size on the target server
3. Copy it somehow over network
I'm not sure of 2 (make volumes of exactly the same size)
check your PE size, otherwise does what it says on the tin.
Hmm, how? I just want to move some (not all) logical volumes.
I guess fdisk is a good idea?
# fdisk -l /dev/LVM2/ocsi1
Disk /dev/LVM2/ocsi1: 3221 MB, 3221225472 bytes

Quoted 6 lines

3221225472 / 1024 = 3145728
lvcreate -L3145728k -n ocsi1 LVM2
Hmm, hopefully, it's the right size?

how about:
lvs --units k

Quoted 4 lines

Quoted 2 lines
Quoted 1 line
and 3 (how to copy it all over network, if possible, using SSH only).
netcat and dd over forwarded port
Thanks for the idea.

dd if=/dev/$vg/$lv bs=32M |

	buffer -S 10m -s 512k |
	gzip -1 |
ssh -T $target -- \
	"gunzip |
	buffer -s 512k |
	dd bs=32M of=/dev/$t_vg/$t_lv"

Signature · 3 lines

: Lars Ellenberg                            Tel +43-1-8178292-55 :
: LINBIT Information Technologies GmbH      Fax +43-1-8178292-82 :
: Vivenotgasse 48, A-1120 Vienna/Europe    http://www.linbit.com :

Rob Ostrander

2007-02-23 21:26 UTC #4 lore raw

[linux-lvm] workaround for RHEL4 + LVM2 inactive snapshot kernel panic

It seems RHEL4's latest kernel and device-mapper have a bug which causes a kernel panic when an a snapshot gets *filled*, marked as inactive and then removed. This is a known bug and fixed in later kernels but RHEL is a few versions behind... What would you suggest to get around this issue?
Is there any overhead for keeping inactive snapshot LVs around?

Tomasz Chmielewski

author

2007-03-02 11:43 UTC #5 lore raw

In reply to Lars Ellenberg

Quoted 31 lines · Lars Ellenberg

/ 2007-02-23 16:59:23 +0100
\ Tomasz Chmielewski:
Quoted 29 lines
Quoted 9 lines · [email protected] schrieb:
Quoted 8 lines · Tomasz Chmielewski
I have a server which stores several LVM-2 logical volumes.
As this system is pretty loaded, I'd like to move some of the logical volumes to another machine. It has to be done remotely, so I can't do it as described in the LVM HOWTO (where one basically adds/replaces disks in one machine).
My common sense tells me that I should:
Quoted 7 lines
Quoted 6 lines
1. Unmount/not use the logical volumes on the source server
2. Make volumes of the same size on the target server
3. Copy it somehow over network
I'm not sure of 2 (make volumes of exactly the same size)
check your PE size, otherwise does what it says on the tin.
Hmm, how? I just want to move some (not all) logical volumes.
I guess fdisk is a good idea?
# fdisk -l /dev/LVM2/ocsi1
Disk /dev/LVM2/ocsi1: 3221 MB, 3221225472 bytes
3221225472 / 1024 = 3145728

Quoted 7 lines

Quoted 3 lines
lvcreate -L3145728k -n ocsi1 LVM2
Hmm, hopefully, it's the right size?
how about:
lvs --units k

Indeed this one is more LVM-specific.
Didn't work for me though - my kernel oopsed a while ago, and all lvm commands stopped to work. When started, they were just in a "D" state.

The kernel oopsed after I created a snapshot, made it full (invalidated), and tried to remove it. The kernel I used was 2.6.17.8 running on Debian-ARM:

# lvremove /dev/LVM2/pdc-backup-new
  /dev/sda2: Checksum error
Do you really want to remove active logical volume "pdc-backup-new"? [y/n]: y
Segmentation fault

# dmesg
Unable to handle kernel paging request at virtual address 31376632
pgd = 8cce8000
[31376632] *pgd=00000000
Internal error: Oops: f3 [#1]
Modules linked in: iscsi_trgt bonding dm_snapshot dm_mirror loop
CPU: 0
PC is at exit_exception_table+0x44/0x70 [dm_snapshot]
LR is at exit_exception_table+0x40/0x70 [dm_snapshot]
pc : [<7f00bbe4>]    lr : [<7f00bbe0>]    Not tainted
sp : 8718dd24  ip : 31376632  fp : 8718dd48
r10: 804f14e0  r9 : 8718c000  r8 : 00000080
r7 : 90ba4218  r6 : 00000043  r5 : 8ee3577c  r4 : 31376632
r3 : 8e6e7000  r2 : 00000078  r1 : 8e6e7000  r0 : 804f14e0
Flags: Nzcv  IRQs on  FIQs on  Mode SVC_32  Segment user
Control: 397F  Table: ACCE8000  DAC: 00000015
Process lvremove (pid: 17844, stack limit = 0x8718c198)
Stack: (0x8718dd24 to 0x8718e000)
dd20:          8ee35740 8f643940 8f643944 90b9f020 00200200 00100100 8718dd70
dd40: 8718dd4c 7f00bcb0 7f00bbac 90b9f020 8abc4460 00000001 00000034 00000004
dd60: c134fd04 8718dd90 8718dd74 8016791c 7f00bc1c 8ee355c0 8abc4460 00000000
dd80: 8016aab0 8718ddac 8718dd94 80166888 801678a8 8cf55aa0 8abc4460 8718c000
dda0: 8718ddc4 8718ddb0 8016a284 80166828 90b98000 90b98000 8718ddd8 8718ddc8
ddc0: 8016aaf4 8016a208 000ae5d8 8718df44 8718dddc 8016a888 8016aabc 8be31468
dde0: 00000000 8718de00 8718ddf4 00000004 00000006 00000000 00004000 00000134
de00: 00000000 00000000 0000020c 00000000 00000000 0000fd26 00000000 00000000
de20: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
de40: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
de60: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
de80: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
dea0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
dec0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
dee0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
df00: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
df20: 8619d0a0 000ae5d8 c134fd04 00000005 8001fea4 2aadea7c 8718df5c 8718df48
df40: 800826a8 8016a5ac 8619d0a0 000ae5d8 8718df84 8718df60 80082920 8008264c
df60: ffffffff 8718dfac 8619d0a0 fffffff7 c134fd04 00000036 8718dfa4 8718df88
df80: 80082990 800826c4 00000000 0004537c 2aad66d0 2aad5274 00000000 8718dfa8
dfa0: 8001fd00 8008295c 0004537c 2aad66d0 00000005 c134fd04 000ae5d8 00001220
dfc0: 0004537c 2aad66d0 2aad5274 000b8808 0000004e 2aad66cc 2aadea7c 000ae5d8
dfe0: 2aadebec 7e987c3c 2aad44f8 2ac811e4 80000010 00000005 b7118d2a f2641dfd
Backtrace:
[<7f00bba0>] (exit_exception_table+0x0/0x70 [dm_snapshot]) from [<7f00bcb0>] (snapshot_dtr+0xa0/0xf4 [dm_snapshot])
[<7f00bc10>] (snapshot_dtr+0x0/0xf4 [dm_snapshot]) from [<8016791c>] (dm_table_put+0x80/0xe4)
[<8016789c>] (dm_table_put+0x0/0xe4) from [<80166888>] (dm_put+0x6c/0xcc)
 r7 = 8016AAB0  r6 = 00000000  r5 = 8ABC4460  r4 = 8EE355C0
[<8016681c>] (dm_put+0x0/0xcc) from [<8016a284>] (__hash_remove+0x88/0x9c)
 r6 = 8718C000  r5 = 8ABC4460  r4 = 8CF55AA0
[<8016a1fc>] (__hash_remove+0x0/0x9c) from [<8016aaf4>] (dev_remove+0x44/0x64)
 r5 = 90B98000  r4 = 90B98000
[<8016aab0>] (dev_remove+0x0/0x64) from [<8016a888>] (ctl_ioctl+0x2e8/0x3a8)
 r4 = 000AE5D8
[<8016a5a0>] (ctl_ioctl+0x0/0x3a8) from [<800826a8>] (do_ioctl+0x68/0x78)
[<80082640>] (do_ioctl+0x0/0x78) from [<80082920>] (vfs_ioctl+0x268/0x298)
 r5 = 000AE5D8  r4 = 8619D0A0
[<800826b8>] (vfs_ioctl+0x0/0x298) from [<80082990>] (sys_ioctl+0x40/0x60)
 r7 = 00000036  r6 = C134FD04  r5 = FFFFFFF7  r4 = 8619D0A0
[<80082950>] (sys_ioctl+0x0/0x60) from [<8001fd00>] (ret_fast_syscall+0x0/0x2c)
 r6 = 2AAD5274  r5 = 2AAD66D0  r4 = 0004537C
Code: e59c4000 ea000002 eb418145 e1a0c004 (e5944000)

So fdisk was the only way to read these partitions size - well, I also did "dd if=partition of=/dev/null" to measure the size :)

Quoted 4 lines

Quoted 3 lines
Quoted 2 lines
Quoted 1 line
and 3 (how to copy it all over network, if possible, using SSH only).
netcat and dd over forwarded port
Thanks for the idea.

Quoted 8 lines

dd if=/dev/$vg/$lv bs=32M |

	buffer -S 10m -s 512k |
	gzip -1 |
ssh -T $target -- \
	"gunzip |
	buffer -s 512k |
	dd bs=32M of=/dev/$t_vg/$t_lv"

I found netcat extremely slow (only 0.5-1 MB/s), perhaps same would be with SSH.

As I use iSCSI, I used such approach:

600 MHz ARM SAN <-iSCSI-> server <-iSCSI> 600 MHz Pentium mobile SAN

On "server" I just used:

dd if=iSCSI-ARM/part of=iSCSI-Pentium/part

With source/destination being the same size.

It was way faster, 6-8 MB/s (I had lots of gigabytes to copy).

Normal transfer rates between SAN <iSCSI> server are about 20-40 MB/s; perhaps it was that slow also because of an earlier kernel oops and a couple of lvm processes in "D" state.

Signature · 2 lines

Tomasz Chmielewski
http://wpkg.org

Bryn M. Reeves

2007-03-02 11:58 UTC #6 lore raw

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

Quoted 4 lines · Tomasz Chmielewski

The kernel oopsed after I created a snapshot, made it full (invalidated), and tried to remove it.
The kernel I used was 2.6.17.8 running on Debian-ARM:

<snip>

Quoted 7 lines

# dmesg
Unable to handle kernel paging request at virtual address 31376632
pgd = 8cce8000
[31376632] *pgd=00000000
Internal error: Oops: f3 [#1]
Modules linked in: iscsi_trgt bonding dm_snapshot dm_mirror loop
CPU: 0

<snip>

Quoted 4 lines

[<7f00bba0>] (exit_exception_table+0x0/0x70 [dm_snapshot]) from [<7f00bcb0>] (snapshot_dtr+0xa0/0xf4 [dm_snapshot]) [<7f00bc10>] (snapshot_dtr+0x0/0xf4 [dm_snapshot]) from [<8016791c>] (dm_table_put+0x80/0xe4)

This is a known issue in older kernels that was fixed upstream in 2.6.19. See:

http://www.uwsg.iu.edu/hypermail/linux/kernel/0612.1/2301.html
https://bugzilla.redhat.com/bugzilla/show_bug.cgi?id=204791

Kind regards,

Bryn.

-----BEGIN PGP SIGNATURE-----
Version: GnuPG v1.4.6 (GNU/Linux)
Comment: Using GnuPG with Fedora - http://enigmail.mozdev.org

iD8DBQFF6BF06YSQoMYUY94RAtZBAKC0+h3dCuTXr6lZ5M2eNfCfF8CPBgCgqMpY rQV9Bm+kHTBVRQoSDdUH7Nw=
=oa4U
-----END PGP SIGNATURE-----