snapshotting to a different volume group

Hiya,

I hope you could help me on this:

in short:

is it possible to create a snapshot LV of an original LV but in a different volume group?

or alternatively, what's the best way to bring up a LV in a VG that is missing a PV (as in a disk crash) if that LV has all its extents on still sane PVs?

in long:

basically, this is what I would like to do:

I have two disks, not as much as I would like, and I want:
- to have 100GB of data mirrored (using Linux software raid (mdadm)) on 2 disks so that it survives disk crashs
- to make a number of frequent snapshots of that data to be able to restore things in case of human errors.
- (and I also want to have some "striped" storage space for worthless data).

BUT, I don't want to have the snapshot volumes mirrored.

My initial idea was to have 2 volume groups: a "safe" one with only one /dev/md0 PV and an "unsafe" one with the striped LVs and the snapshots.

However I couldn't find how to create snapshots of the LVs in "safe" into the "unsafe" VG. It seems to me that it should be possible, shouldn't it?

Anyway, then I merged the two "safe" and "unsafe" VGs into one "vol" VG. Now I can make snapshots of the LV physically located on /dev/md0 into the "unsafe" area.

But if one of the disks fails, /dev/md0 will still be available but the "vol" VG won't be able to come up because of the other missing PVs.

If such a thing happen, I can happily live with the missing snapshots and "unsafe" storage for a while until I buy a replacement HD, but how should I do about making sure that the /dev/md0 filesystem is still available in the mean time?

Or is there a more straightforward way to achieve this that I haven't thought of?

TIA,
Stephane

No, not with LVM.

vgchange -P -ay

If you have 2 disks, you better mirror them to be completely HA on the spindle side of things and avoid any such config hassle, potentially leaving you with lost data.

Heinz

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

Heinz Mauelshagen                                 Red Hat GmbH
Consulting Development Engineer                   Am Sonnenhang 11
Storage Development                               56242 Marienrachdorf
                                                  Germany
[email protected]                            PHONE +49  171 7803392
                                                  FAX   +49 2626 924446

=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-

2008-04-18 10:24:36 +0200, Heinz Mauelshagen:

[...]

Hi Heinz,

thanks. I've just come accross zumastor while reading the ML archives. It looks promising for what I'm trying to achieve. I suspect it's gonna be the same wrt to snapshot though, but I'll give it a try.

Oh thanks. I missed that as the option was described in lvm(8) and not in vgchange(8). I'll see how to make it fit in the initrd scripts.

[...]

I didn't get that ("the spindle side of things"??). Do you mean I should mirror the whole disks? Note that they are not the same size. The thing is I have 400 GB of disk space available, I only need 100GB of data to be "safe". If I had 2 200GB drives and wanted to mirror everything, and have 50GB for snapshots, that would leave me with 150 GB for data. While as I need only 100GB of safe data and don't need the snapshots to be mirrored. I can have 250 GB of storage, 150 GB of which I don't care if it's lost (mirrors and snapshots).

Best regards,
Stephane

2008-04-18 10:24:36 +0200, Heinz Mauelshagen:

[...]

[...]

I just tried that with loop devices. Unfortunately, the LV comes up in read-only. And lvdisplay doesn't show the still sane LVs.

dmsetup info shows it as read-only and lvdisplay -P as read/write, btw:

lvdisplay -P:

--- Logical volume ---
LV Name /dev/test/fs
VG Name test
LV UUID
U03Sgj-dVYW-TBSs-fJHN-qXbm-y3vf-nN58mV
LV Write Access read/write

dmsetup info:

Name:              test-fs
State:             ACTIVE (READ-ONLY)
Read Ahead:        256
Tables present:    LIVE
Open count:        1
Event number:      0

Major, minor: 253, 4
Number of targets: 1
UUID:
LVM-CL1U6NmSQ0QxvXkg8Iy4BEJE4kzEAJokU03SgjdVYWTBSsfJHNqXbmy3vfnN58mV

By bypassing LVM2 and using dmsetup directly, would I theoretically be able to setup a snapshot outside of LVM2 managed volume groups, is it worth me trying to dig that way? Something like: having a LVM (or not) volume group with only /dev/md0 and creating snapshots of that elsewhere while leaving the VG containing /dev/md0 undisturbed?

TIA,
Stephane

2008-04-18 12:53:15 +0100, Stephane Chazelas:

[...]

[...]

[replying to myself]

OK, I get it.

When one creates a snapshot, he changes the original block device to be a "snapshot-origin" so that every write to the device is forked into:
- copying the original block to the snapshot device(s) COW(s)
- writing to the original device.

So, if I've got the COWs on a failed disk, I suspect this might degrade the performances or even stop the snapshot-origin from working altogether wouldn't it?

I'll keep testing...

Stephane