Poor Mans HA

[Stephen Zelenak_1]

I want to setup two SuperDome partitions to boot from the same fibre channel JBOD device as a "poor mans HA" configuration.

I would boot only one partition at a time and if the partitions hardware pukes I would use the other partition to continue operation.

My questions are: Is there any problem with using this configuration and does HP support such a setup?

[Joshua Scott]

I'm slightly confused, are you going to have the backup partition up or down?

Josh

[Stephen Zelenak_1]

Down until needed to replace the origional if needed because of a hardware failure

[Stefan Stechemesser]

Hi,

HP does not support such a solution, but this does not mean you are not allowed to do it this way. It should work, but you have to make sure that the data on the FC disk is only be access from one partition at the same time, otherwise you risk data corruption.
Another thing you have to keep in mind is that the hardware paths to the disks differ because the cell number is the first digit in the hardware path.
When you boot from the other partition, the operating system does not see the old hardware paths but sees new hardware and will create new device files (unfortunately it will crash before doing that, because it does not find the root filesystem).
The Solution is more a workaround and as I said not supported or recomended by HP.
In case of a hardware crash, you would have do do the following:

1.) boot from the 2nd partition in maintenance mode (hpux -lm)

2.) export all volumegroups

3.) import all volumegroups with the new device files

4.) vgchange -a y /dev/vg00 then init 3 or make a reboot

You can write a script to make it easier, but there a several things that might go wrong in case of problems.

Although the best solution would be to use Service Guard (this is the only solution we "support"), I would at least recomend to buy one additional disk to have separate root disks for the two partitions. Then you can have different volume group configurations for the data volume groups and simply activate them (no import/export needed then) if the original system crashes.

I hope this helps

Stefan

[Ted Buis]

This is a major problem with your approach. If one partition fails, and the other one takes over, what will stop the first partition from rebooting and mounting the disks, thus having both mounted at once and corrupting the file systems. This is always the problem with home grown HA solutions, the failover is the easy part. Preventing split-brain is hard. You need a cluster lock manager as ServiceGuard has.

[Ted Buis]

If you study the ServiceGuard manuals at www.docs.hp.com I think you will better understand the issues you face.