LVM, Multipath, and ServiceGuard

[David G. Douthitt]

I'm trying to multipath a LVM volume on HP-UX 11i v3 with ServiceGuard. The disks are on an EMC array, and there are two links to the disk on each of two ServiceGuard cluster nodes.

One (data) disk is currently configured with logical volumes and so forth. Attempting to use vgextend to add a second path results in:

# vgextend -g pvg1 /dev/clvol2 dev/dsk/c6t0d1
Verification of unique LVM disk id on each disk in the volume group /dev/clvol2 failed.
Following are the sets of disks having identical LVM disk id
/dev/dsk/c7t0d1 /dev/dsk/c6t0d1

What am I missing? The initial vgcreate did not have a -g option (that was before I knew about that option). I created /etc/lvmpvg
with the appropriate disk in it, but the message above is what results.

I don't know scsimgr, and I'm still learning ServiceGuard. What am I missing here?

Now in research, I've discovered that the v2 documentation claimed scsimgr is deprecated (!?).

There are some docs that appear to say that multipathing is handled automagically by the native multipathing in 11i v3. It appears that v3 has a *lot* of changes in it from v2.

What am I missing here? Thanks for the help!

[Sandman!]

Imho it might be trivial as having the correct VG name. Is the VG "/dev/clvol2"? Is that correct? Please list the contents of the /etc/lvmpvg file viz.,

# cat /etc/lvmpvg

[David G. Douthitt]

This is /etc/lvmpvg (created by hand):

VG /dev/clvol2
PVG pvg0
/dev/dsk/c7t0d1

The disks presented by the EMC SAN were put into separate volume groups: clvol1, clvol2, clvol3 (CLuster VOLume).

[Mridul Shrivastava]

What I understood is that, following set of disk refers to the same physical disk:

/dev/dsk/c7t0d1 /dev/dsk/c6t0d1

Am I correct ?

If that's the case then I don't think vgextend is required since multipathing will take care of switching.

If it refers to different disk then correct command syntax should be as follows:


vgextend /dev/clvol2 -g pvg0 /dev/dsk/c6t0d1

[David G. Douthitt]

Looking at c6t0d1 and c7t0d1 (which *should* be the same), I find that both have the same bits in the PVRA: the CPU-ID, PV-ID, CPU-ID, and VG-ID (per the CSE manual). In my case, they are:

0xec9b7a8a 0x4738c84f 0xec9b7a8a 0x4738c98f

However, their LUNs and HW paths appear to be different. c6t0d1 is labeled as disk13, and c7t0d1 is labeled as disk16:

disk13:
64000/0xfa00/0xc 0/5/1/0/4/0.0x...08.0x4001...

disk16:
64000/0xfa00/0xf 0/5/1/0/4/0.0x...07.0x4001...

(The missing hex numbers are identical)

All are "online" and labeled as "EMC SYMMETRIX". I know nothing about the SAN itself; it does use FC (on two 1000BaseT/FC combo cards...).

Thanks for the help!

[Sandman!]

Previous command tries to extend pvg1 of clvol2 by adding another PV to it. Instead the syntax to add an alternate path to clvol1 for c7t0d1 is:

# vgextend /dev/clvol2 /dev/dsk/c6t0d1

[Mridul Shrivastava]

Ok.. so both device files refers to the same disk.. this is the reason you were getting error while trying to create a new PVG...

As Sandman said you have to just add the alternate path so need to give the vgextend command as suggested by Sandman orã me in last post.

[David G. Douthitt]

That command (vgextend) was the first error that got me going:

# vgextend /dev/clvol2 /dev/dsk/c6t0d1
Verification of unique LVM disk id on each disk in the volume group
/dev/clvol2 failed.
Following are the sets of disks having identical LVM disk id
/dev/dsk/c7t0d1 /dev/dsk/c6t0d1

I double-checked; above is the result. I'm not sure, but I think that the native multipathing is already in place (as someone suggested already). Now how do I verify this?

[melvyn burnard · Accepted solution]

with HP-UX 11.31, native multipathing is included and ACTIVE when you install the OS.

ioscan has been updated to be used with this, for exampole:
ioscan -m dsf /dev/dsk/c19t0d1

Persistent DSF Legacy DSF(s)
========================================
/dev/disk/disk33 /dev/dsk/c19t0d1

man ioscan

[David G. Douthitt]

Is there a way I can test to verify that multipathing is present and working? Is there a command I can run to see where multipathing is in place?