reducing multiple paths on write disabled disk

[IT_2007]

Is it possible to reduce alternate paths on write disabled disk? Isn't going to update LVM header on the disks it self? Just want to double check.

Thanks.

[Sandman!]

What do you mean by "write disabled disk"? Is it still readable? And yes the alternate paths can be reduced without a problem as they exist purely for redundancy to maximize uptime. And no the LVM header won't be updated. Just that the LVM software won't be aware of the existence of the alternate path to the same storage.

[IT_2007]

This is service-guard cluster nodes. LUNS are write enabled on one host and write-disabled on another host. Want to reduce two alternate paths out of 6 on second host which got LUN in write-disabled mode.

Don't have any test box to try it out. Want to confirm the above question.

[Torsten.]

I'm not really sure about your question, but perhaps "pvchange -a n|N ..." can help you.

[Sandman!]

Are these LUNs part of a MC/ServiceGuard pkg? If so then reducing two paths on one node without making that change to all the other cluster nodes would cause problems in a failover situation. Would be better to make the changes on the configuration node and then vgexport/vgimport the desired VG on all the rest of the cluster nodes.

[IT_2007]

Primary node has 4 paths and failover node has 6 paths for some reason (don't know why) and now Storage wants to reclaim two paths from failover node and LUNS are write-disabled mode. Thought it can be non-intrusive method to take out two unwanted paths from failover node even though they are read only.

[Sandman!]

You can reduce the undesired paths from the VG on the secondary node. But it is better to make sure that the paths to be reduced are not in use on primary node. Just so that the in use paths are not clobbered. So vgexport the VG on the primary node; remote copy the mapfile from the primary node and onto the secondary node; vgexport the VG on the secondary node; and vgimport the VG using the mapfile created on the primary node:

# vgexport -p -v -s -m /tmp/vg??.map
# [rs]cp /tmp/vg??.map node2:/tmp
# vgexport vg??
# vgimport -m /tmp/vgname.map -s -v vg??

[Carsten Krege · Accepted solution]

This is not a 100% answer, I did not test this. In my eyes there is no problem to reduce redundant LVM pathes from write disabled disks as it does not require writing into the LVM header. The LVM header does not reference alternate pathes and also does not assign them separate ids or so. The main changes are done to the lvmtab file and to the kernel memory of LVM. I don't see a problem. Nevertheless it could still require that the VGDA is updated (for time stamp or so).


Nevertheless if the disks are bootable, ie. carry a Boot Data Reserved Area (BDRA) in the LVM header, then it *would have* to be written to the disks. It doesn't sound this is the case here though.

I don't see a problem in simply trying this out by running vgreduce. Either it succeeds or fails with an error.


Carsten

[Juan M Leon]

IT_2007,
Like sandman said, you can reduce the unwanted paths without affecting the cluster availability.
Just make sure the paths that are being disable are not the ones used on the other node. For example:

Node A:
/dev/dsk/c45t0d0
/dev/dsk/c46t0d0
/dev/dsk/c47t0d0
/dev/dsk/c48t0d0

Node B:
/dev/dsk/c45t0d0
/dev/dsk/c46t0d0


Using the above chart, you know that you will need to vgreduce the alternate paths for c47 and c48.

Then vgexport a map file using -p (previous) and -s (no disk names required when vgimport) option and do a vgimport on the altername node.

Just remember to have the running package on the node that is vgexport -p the map file.

Good luck

[Sandman!]

Update to my last posting as I forgot a few steps in the procedure i.e.

# vgexport -p -v -s -m /tmp/vg??.map vg??
# [rs]cp /tmp/vg??.map node2:/tmp
# vgexport /dev/vg??
# mkdir /dev/vg??
# mknod /dev/vg??/group c 64 0xNN0000
# vgimport -m /tmp/vg??.map -s -v vg??

[IT_2007]

ok. Sounds good to me.

thanks.

[IT_2007]

yes. got the solution.