On Sun, 2008-08-31 at 11:19 +0200, Simen Timian Thoresen wrote:
Hi list,
I'm having one of those 'I'm stupid' -problems with LVM on CentOS 5.2. I've been working with traditional partitions until now, but I've finally been sold on the theoretical benefits of using LVM, but for now I only have a huge pile of broken filesystems to show for my efforts.
My scenario; I attach a disk, either over USB or iSCSI. I create a PV on this device, create a VG using the device, and slice off a tiny LV for tests.
Uh-oh! I hope these are for temporary usage only!
[root@kasse ~]# dmesg
<snip>
[root@kasse ~]# pvcreate /dev/sdg1 Physical volume "/dev/sdg1" successfully created [root@kasse ~]# vgcreate testgroup /dev/sdg1 Volume group "testgroup" successfully created [root@kasse ~]# lvcreate -L 200M testgroup -n testLV Logical volume "testLV" created
At this point, I have /dev/testgroup/testLV, which I can stick a fs on, mount, store files on, unmount, remount and recover the files from. Just as I'd expect.
[root@kasse ~]# lvdisplay --- Logical volume --- LV Name /dev/testgroup/testLV VG Name testgroup LV UUID 3KjDiZ-gTAD-sC9E-sOSA-772o-41Yk-ZKK7cx LV Write Access read/write LV Status available # open 0 LV Size 200.00 MB Current LE 50 Segments 1 Allocation inherit Read ahead sectors auto
- currently set to 256
Block device 253:1
I have not explicitly set the vg active, as I understand that it implicitly will be set active in non-clustered environments like mine. LVM works as I'd expect it to up until here in any case.
So - with the LV unmounted, I power my USB-device down, and then back up;
[root@kasse ~]# dmesg
<snip>
(note that my disk is now sdi, not sdg, and connected on scsi28 and not scsi27)
By now you realize that is a big problem?
If I now try to access the device, all falls appart; [root@kasse ~]# lvdisplay /dev/testgroup/testLV: read failed after 0 of 4096 at 209649664: Input/output error
<snip>
--- Logical volume --- LV Name /dev/testgroup/testLV VG Name testgroup LV UUID 3KjDiZ-gTAD-sC9E-sOSA-772o-41Yk-ZKK7cx LV Write Access read/write LV Status available # open 0 LV Size 200.00 MB Current LE 50 Segments 1 Allocation inherit Read ahead sectors auto
- currently set to 256
Block device 253:1
Trying to mount or otherwise access the LV also shows the problems;
scsi 27:0:0:0: rejecting I/O to dead device
<snip> Buffer I/O error on device dm-1, logical block 7 scsi 27:0:0:0: rejecting I/O to dead device scsi 27:0:0:0: rejecting I/O to dead device scsi 27:0:0:0: rejecting I/O to dead device scsi 27:0:0:0: rejecting I/O to dead device hfs: unable to find HFS+ superblock
So - LVM seems to believe my LV is still attached to scsi27, while it now actually sits on scsi28.
What am I missing here?
Use the pvdisplay command and you'll see.
# pvdisplay --- Physical volume --- PV Name /dev/sdb2 VG Name VolGroup00 PV Size 148.95 GB / not usable 11.37 MB Allocatable yes (but full) PE Size (KByte) 32768 Total PE 4766 Free PE 0 Allocated PE 4766 PV UUID eeqb0A-6EJV-VE3Q-paLi-kTXv-S46k-0xaeTy
Sit back a moment and think about it. You defined a PV on a specific device. That device is no longer available. LVM has to track the assignment, allocation and use of PVs. It must have a record of it somewhere.
# ls -l /etc/lvm total 44 drwx------ 2 root root 4096 May 25 00:41 archive drwx------ 2 root root 4096 May 25 00:41 backup drwx------ 2 root root 4096 Aug 30 08:38 cache -rw-r--r-- 1 root root 15911 May 25 00:41 lvm.conf
Use the man command to read up on more stuff, like "man lvm.conf" and "man lvm". Be sure and read some of the other refs at the bottom of the man pages.
How do I get LVM to use a persistent device instead of a transient path?
Normally you give a persistent device instead of a transient path. You didn't do that.
Generally speaking, LVM is designed to use fixed devices - those that always come up on the same address. Neither iscsi nor UHB devices do this reliably. However, on a reboot, there is a liklihood that they will be the same devices.
I've not tried to configure it on transient devices. There may be a way, but I've not looked into it.
If you make them their own VG, rather than adding the PV to another VG, you can use the export and import LVM commands, but then you really accomplish nothing more than what could be done by mounting the underlying devices in existing directory structures. No point in LVM then unless you anticipate having to grow those mounted structures while in use.
Even so, from my LFS days, I do recall a lot of flexibility in the udev process. Maybe there is something there that will allow the assigned device addresses to be fixed?
Do man for udev, udevd, udevinfo, udevmonitor. I'm guessing that you'll want to run some utility from udev that will (maybe) look at a a volume label (if one if available on the usb device - is there one on the iscsi device?) and assign a predetermined name if a certain label is found.
I've never used iscsi, so I'm even less help on that. Maybe there is something in its man pages that allow a persistent device name assignment.
Yours, -S
Sorry I couldn't offer more.