tape recovery of just one disk?

[Howard Marshall]

I have a 10.20 system that I have started getting scsi errors on the boot disk and I would like to replace it. The system was se up with not only the OS but also the application disks in the root vg (I didn't set it up that way). Unfortunately it’s a constant production machine, light work but always doing something so I its preferable to have as little down time as possible, you know, especially for preventative maintenance.

In the assets column I have a backup machine that I can take up and down pretty much at will and I have the replacement disk.

Is there any way I can use ignite make tape recovery or some similar tool to install the vg and lvol layout info, the boot info, and just the file systems for the boot disk and just shut the production machine down and swap out the disk drive.

I have cleared the lvols that cross more than just the one disk so that shouldn't be an issue.

[Mel Burslan]

if you use make_tape_recovery with -x inc_entire=vg00, you should be able to get all of vg00 contents to the recovery image regarless how many disks they are spread across.

A caveat emptor, I am not sure if inc_entire option of make_tape_recovery is available on 10.20 or not.

HTH

[Patrick Wallek]

NO, unfortunately there is not really a way to do that. Ignite does let you exclude certain directories, but when you rebuild the VG00, the other non-OS LV's in VG00 will go away. There is no way I know of to import other LVs.

What type of disk is it? What type of system? Is there any mirroring done? Do you have any slots available to add a disk?

[Pete Randall · Accepted solution]

You can also use the " -x exclude=dir" syntax to exclude your data directories and just get the OS image. Check the man page for your version.


Pete

[Devender Khatana]

Hi,

As stated above you can exclude particular mount points in ignite backup. So exclude the mount points for LVOLs which contains application data.

Also if you are recovering system, it will be a good time to move these data LVOLs out of vg00. Create or import other VG's and then create some new LVOLs/file systems and mount them on the same path where your current vg00 data LVOLs are mounted. As the applications only referes to mount point and not anything else, there should not be any problems in doing this.

But take caution if you have database running with raw devices. It will not be possible in that case.

HTH,
Devender

[Howard Marshall]

I have a recovery tape that includes the entire vg00 but that is three disks and a great deal of data that will take a lot of down time to restore from tape and that is what I am trying to avoid by wanting to tape it up on the production machine and re-create the boot disk on the backup then just shut the prod server down, switch the disks, bring prod back up.

I am sure that this is possible, it is after all, just 1s and 0s. Unfortunately I do not have mirroring or I would do it that way.

My original though was to add the disk to the standby machine, make it bootable, duplicate the file systems from prod, copy the data across the network, switch drives, boot in lm mode export the vg then import vg00 with all the disks but the system gets very upset when the number of disks in the exported vg don't match and the vgids do not match and I don't know how to specify the vgid or how to add phantom disks to the one I create on the other machine.

I think I am going to try to use the exclude list but I have little confidence in it sense the system was not set up (in my opinion) correctly.

The application programs were not installed in their own logical volumes, they were just added to var or usr or opt and that file system was increased to accommodate it. What it comes down to is I have a lot of application stuff in my root vg and it doesn't take very good advantage of LVM

H

[Mike Shilladay]

Morning Howard,

There is an entirely different way of doing this, by using the dd command

dd if= of= bs=2048

Have not tried it with a boot disk, nor going via a tape drive, but I do not see why it will not work.

I have successfully taken a copy of a data disk this way.

If you have room to put a new disk in, you could copy from disk to disk, this way is less time consuming in that you don't have to 'bounce' the data from the tape. It goes without saying please backup you system first before playing with disks.

regards

Mike.

[Howard Marshall]

Unfortunatly I cant stick another disk in the system and there is one thing about the dd process that scares me. Yes, I tired it on the test server and it does work however, wouldn't it also copy over, and over write existing bad block relocations and bad cyl information on the disk its self?

Any way, I think I may have solved it. The current root vg is three disks but if I reduce the size of two of the lvs that have extra space I can get them onto two disks so what I am planning is to create a make_tape_recovery on both the prod system, and the standby, put the new disk in the standby and "borrow" another free disk in the backup machine and restore it there then swap both disk, then I can re include the third disk and expand my lvs and file systems back out.

I will let you know what happens.

[Howard Marshall]

Well, this process worked with only a few small itty bitty "issues" that I am sure could be corrected with time.

I created the make tape recovery on the prod system, then installed the new disk and another spare disk in my backup server and recovered the tape there. After the recovery process restored all the files but before it could reboot I shut the system down and removed the two disks.

Then, during my small window of opportunity, I pulled two rootvg disks from them prod machine and installed the two I had recovered to and booted it up. It finished the recovery process on the prod server and with a few minor shifts (permissions on some files in the dev dir) it came back up and the software started running fine.

Now I have a new boot disk and simply rotated a spare into service and rotated an in service disk out to spare status.

One little "issue" I have noticed sense is that apparently the tape drives were on different scsi ids between the two machines because now my only tape drive is 1 instead of 0 and I feel certain there are other little minor things that will come bite me as time goes on.

Thanks for the help guys

H