Disk replacement in SG environment

[Mihails Nikitins]

Hi,

I need to replace one shared disk in SG environment on _running_ system. I have experience with standard procedures of mirrored disks replacement, but I'm unsure about some cluster specific things.

1. The disk to be changed is still running, this is a preventive replacement because of some EMS warnings. Should I really take it off from LVM first rather than just pulling disk out?

E.g.

lvreduce -m 0 /dev/vgxx/lvolx /dev/dsk/cxtxdx
vgreduce /dev/vgxx /dev/dsk/cxtxdx

The volume group occupies 6 disks, they are 3 mirrored pairs, mirroring is restricted by pvg. Will lvreduce command break mirroring for all pairs?

2. The real problem for me is that the disk is cluster lock. Does vgcfgbackup/vgcfgrestore copy cluster lock structures? As I understand, I can use vgrestore only if extracting the disk without LVM commands. As fas I understand, when using lvreduce/vgreduce one can't retrore new disk by vgcfgrestore. So cluster lock should be made by cminitlock or during offline in the second case.

3. Should I run vgexport/vgcfgbackup procedures on cluster standby node after replacement?

Many thanks and points in advance for your help! I really need to prepare a smooth replacement procedure.

BR,
Mihails

[Sanjay_6 · Accepted solution]

Hi,

You can replace the disk in the same manner in which you replace a normal mirror disk. Do it from the node on which the vg is active. The Lock disk info is saved through vgcfgbackup if a vgcfgbackup was run after the lock disk info was created on the disk.

Try this link from docs site to help you more.

http://docs.hp.com/cgi-bin/fsearch/framedisplay?top=/hpux/onlinedocs/B3936-90065/B3936-90065_top.html&con=/hpux/onlinedocs/B3936-90065/00/00/58-con.html&toc=/hpux/onlinedocs/B3936-90065/00/00/58-toc.html&searchterms=cluster%7clock%7cdisk%7creplace&queryid=20040405-090129

Hope this helps.

Regads

[Sanjay_6]

Hi,

here is a full procedure which should take you through any issues that you might have if the simplest one doesn't work.

http://www1.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000072157535

The itrc doc id is UXSGKBAN00000022.

Hope this helps.

Regds

[A. Clay Stephenson]

As already noted, vgcfgrestore will restore the lock disk structures and since the cluster is already formed a lock disk is not needed at the moment so it is safe to remove the disk from a running system. This does assume you are mirrored which you indicate that you are.

Next do an lvdisplay on all affected LVOL's and make sure that all the extents on the remaining good mirror are marked "current".
I would then yank the flakey drive (actually pull it out gently and allow it to spin-down while still partially inserted). At this point wait a few minutes to make sure the box has recogized that one of the disks is bad (i.e. can't be communicated with). You can now proceed with a standard mirrored disk replacement (vgcfgrestore, vgchange -a e, vgsync). Nothing need to done on the other nodes.

NOTE: I am assuming that you have a valid vgcfgbackup on the active node. It is a very good idea to do a vgcfgbackup on all the nodes in a cluster so that disk replacements can be done on whichever nodes happen to be hosting a given disk at the moment.

[Sanjay_6]

Once the disk is replaced and the disk is recognized And the vgcfgrestore is done. You need to do a vgchnage -a y vg_name to reactive the newly added disk in the vg and then do a vgsync to sync any/all lvs on the disk that failed.

Regds