removing LV from SG cluster

[himacs]

Hi Admins,

B.11.23

I want to remove a LV from cluster.

1.Can i achieve the same by without halting the pkg/cluster

2.If yes then, After any changes/modifications ,how to update the custer config files.I heard the cmapplyconf require cluster halting.

Plz help on this

According to many threads we can extend the VG,LV without halting the package

http://forums11.itrc.hp.com/service/forums/questionanswer.do?threadId=1314756

Regards
himacs


3.

[Jayakrishnan G Naik]

Hi himacs

This is a tricky thing as many people might not have faced this in real, I would say halt of package is needed do the changes.


1) Halt the package
2) modify the necessary package file(pkg.cntl) details to remove the filesystem & lv details to be removed
3) do the lv removal steps. (lvremove)
4) export vg in preview mode & copy file to the standby node.
5) do the export the existing vfconfig on the standby node, mkdir and mknod and vgimport(use same minor number in mknod)
6) once this is done copy the changed package cntl/config files to other node, at the correct location
7) once all these done , go for cmcheckconf -P & cmapplyconf -P
8)Start the package.

Thanks & Regards
Jayakrishnan G Naik

[himacs]

Hi J,

Thanks for the response..

How does it affects cluster if i removed the VG/LV online..

plz let me know what are te activities can be done online.. so that we can avoid downtime..

regards
himacs

[Jayakrishnan G Naik]

Hi himacs

Are you going to remove vg also?

If you are removing only one logical volume, there is no effect on cluster. You can consider this as a situation - when you create a new package.

We cmcheck and cmapply the package config when the cluster is running and that has nothing to do with running cluster & cluster configuration.

-->.I heard the cmapplyconf require cluster halting

The cmapplyconf don't need halting of packages in all the cases. I have changed some of the parameters like timeout value etc and I applied it online, but tasks where devices are involved cannot be done online.
Example addition of heartbeat NIC, volume group etc.

Adding a vg or removing a vg may need down time and can be planned in another way.

Removal of volume group can be done online with cluster up, & package (package which the vg is used) down but you need to cmcheck and cmapply package cntl immediately. (if package having multiple vgs and one is removed)
And update the cluster config file and check and apply it during the very next available cluster down time without fail.


Thanks & Regards
Jayakrishnan G Naik

[Matti_Kurkela · Accepted solution]

> How does it affects cluster if i removed the VG/LV online..

If you did not change the package configuration, then next time the package needs to failover (either manually on command, or automatically in case of a hardware fault), the cluster will try to unmount/deactivate a LV/VG that is no longer there. This will result in an error.

Whenever an error is detected in halting the package, Serviceguard thinks it might mean it's unsafe to move the package to another host, and stops any automatic failovers. The package will remain where it was, possibly in a partially or completely downed state.

When starting a package on another node, if the package configuration calls for activating a VG or mounting a LV that no longer exists, this will also result in an error. If an error is detected in package start-up, the package will automatically be halted.

In other words, after your change, your package will most likely be unable to perform a successful automatic failover.

This will seriously slow down the recovery of your packaged service in case of problems: instead of Serviceguard automatically moving the package to a working node if e.g. the network connection of the current node fails, a sysadmin needs to be alerted to fix the problem.

Logs must be read to find out why the package is not failing over, a possible incomplete shutdown must be cleared (unmount package LVs if any are still mounted, deactivate package VGs if any are still active), then the package configuration needs to be fixed and re-applied, before the package can be re-started. (Don't forget to "cmmodpkg -e" the package to re-enable failover!)

After all this is done, someone will probably have to explain to his/her boss why the automatic failover didn't work as advertised, and why there was significant unplanned package downtime and possibly a Service Level Agreement violation.


> plz let me know what are te activities can be done online..

This depends on your Serviceguard version. In general, newer versions allow more things to be done online than older ones.

For each version of Serviceguard, the appropriate edition of the "Managing Serviceguard" book has a very clear list of both cluster and package management actions and whether they can be done online or offline. A hardcopy version of the book comes with each Serviceguard license; a PDF version is downloadable at:

http://www.hp.com/go/hpux-SG-docs

See chapter "Cluster and Package Maintenance", sub-chapters "Reconfiguring a Cluster" and "Reconfiguring a Package". You'll find a table that lists maintenance actions and the required cluster/package states for each action.

MK

[rariasn]

Hi Himacs:

1.- Yes

2.- Update SGuard configuration.

- umount /vgname/lvtoremove

# lvremove /vgname/lvtoremove

- remove filesystem activation from pakage control script.

- Copy control pakage script to another cluster node

- Prymari node:

vgexport -s -p -m vgname.map vgname

copy vgname.map to secondary node

- Secondary node:

vgexport, mdkir, mknod and vgimport

rgs,



rgs,

[himacs]

Hi All,

Thanks for the responses.. points assigned..

After the activity need to modify pkg/cluster configuration file also.And no need to run cmapplyconf.

Regards
himacs