Split site ServiceGuard cluster using IBM storage

[Simon Millard]

I am trying to set up a 2 node serviceguard cluster using 2 x IBM 4800 storage enclosures. My problem is that I cannot see how I can get the intersite mirroring to flip from within ServiceGuard.

I know that with our XP512 RAID Manager handles this, but cannot see how ServiceGuard calls it.

I think I need to get ServiceGuard to run a script (which would contain the appropriate commands) before it does anything with volume groups but cannot see how to do this.

If anyone can help, or provide a better solution I would be most grateful.

[melvyn burnard]

You need to have the storage arrays connected to both nodes and use miroring to connect all of this up.

One question is do you just hav eone XP512, or do you hav eone at each site?
If so, you could be using Metrocluster, which uses RaidManager.

By the way, IBM storage is not officially supported for use with Serviceguard

[Simon Millard]

Many thanks for your response, however I dont think I made myself clear.

I have set up mirroring between the two DS4800s and know it works. My problem is that in order to perform a site failover I need ServiceGuard to get the DS4800s to switch their mirroring roles.

Im not expecting ServiceGuard to actually issue the commands, but I need it to run a script before it starts bringing the volume groups online at the remote end.

If I was using the two XP512s then ServiceGuard would get RAID Manager to do this, however, I cannot see in the ServiceGuard configuration file for the package how it does this.

[melvyn burnard]

so you are using Continuous Access on the XP arrays?
And they are set up as Pvol/Svol pairs?
This is data replication done by the hardware.
If so, then this is via a toolkit that is purchased from HP to do this.
There is no support or toolkit for the IBM storage.
Otherwise if you are just using Mirrordisk/UX to mirror between arrays there is nothing additional to do, as this is done as per normal.
This is data replication don ein software.

[Stephen Doud]

It sounds like you are referring to a p-vol/s-vol arrangement where if the package is to operate at the other site, you want the other site's storage array to flip it's role from secondary/backup, to primary/production... is this correct?
With HP's XP series, this would be done by Metrocluster software. Use swlist to investigate whether you have that product.
If so, as Melvyn said, there is no equivalent for IBM storage. If you don't use Metrocluster, then you will have to write a script to cause the transformation to take place.
I recommend if you want to do it - modify the package control script to call the transformation script, and if the script succeeds, proceed with the rest of the package control script... but if not successful, exit 1.
Your custom transform script should account for the valid and invalid operational states of the array.

Similar scripting should be at the Primary production site as well, when you want to move the package back.

This site discussed the viability of third party storage arrays with Serviceguard:
http://www.hp.com/products1/serverconnectivity/mass_storage_devices

I don't see the 4800 mentioned.

[Simon Millard]

I agree that I need to run a script to get the IBM mirroring to swap ends, however I can't see how you get ServiceGuard to run the script. I need to run it before it does anything else (like starting up volume groups) but the only place I can see to enter scripts is in the customer defined functions area of the ctrl file. However, it looks like these scripts are run to start things like databases, which is too late in the startup chain.

Apologies if I am being thick.

[melvyn burnard]

correct, you need to do stuff BEFORE the script starts to activate VG's etc.
This is what the HP Metrocluster toolkit does for XP and EVA arrays.
There is nothing for IBM arrays

[Stephen Doud]

As Melvyn said, call the array test and modification script at the top of the package control script, before parameter declarations.

[Simon Millard]

Thanks guys, we think we have gotten it working and it passes all our tests.

[Simon Millard]

We seem to have gotten it working. It passes all our test criteria and so I am closing this thread.