Manual Fail-over

[Motorola - AIEG]

I have two L2000 systems. Each system has two internal disks which are in the vg00 volume group and are complete boot-disk mirrors. In addition, each system has a third disk which is in a FC-Array attached via a FC-Hub. This third disk is also part of vg00 and contains some logical volumes/filesystems with application related data (NOTE: These data LV/FSs are also mirrored onto the internal disks).

In the case of a system failure, I would like to "import" that third disk onto the other system, mount the data related logical volumes/filesystems onto this working system, bind a IP address from the failed system to the working system, and restart that application from the failed system.

This environment would give me a poor-mans, manual fail-over. Some of the questions I have are:

(1) Both systems use vg00 as the root VG, and the data LVs are in the vg00 VG. Can I add this disk to the currect vg00 VG and only mount the necessary LVs?

(2) Do I need to run a crontab entry which periodically runs a vgexport and copies it to the other system?

Thanks in advance for any comments you can offer.

[Patrick Wallek]

This would be a whole lot easier if you data VG was in its own VG, like VG01, instead VG00.

Since it is part of VG00, I know of no easy way to move that disk from one machine to the other.

As it is, I think you are going to have a hard time.

I would recommend backing up all of your data, removing that disk from VG00, create VG01, create your LVOL's and restore your data.

Then you could run a 'vgexport -m mapfilename.map -p -v -s vg01', copy that mapfile to your second machine, and then import that vg when needed via 'vgimport -m mapfilename.map -s -v vg01'. Naturally you would need to create the /dev/vg01 directory and /dev/vg01/group file as well.

[A. Clay Stephenson]

By keeping data in vg00 you have made this more difficult. You can't add these disks to another vg00 (at least and keep the data intact) but you could vgimport the entire vg00 into another VG (e.g. vg00b) and then mount the filesystems in these LVOL's.

You really don't need a crontab. The only thing you need is an up to date mapfile of the VG's that is kept on both boxes. You might do a cron to do a vgexport -p -m mapfile and rcp the mapfile to the other box unless your lvols are simply lvol1, lvol2, ... .

What you really need to do is look at MC/ServiceGuard and do all this crazy stuff automatically.

[James R. Ferguson]

Hi:

I wouldn't want to try to make this work. The fact that you have non-operating system (application) data on logical volumes belonging to vg00 is poor in and of itself.

You are attempting to mimic an MC/ServiceGuard environment. If you need to avail yourself of those features (and guarantee that it will work), I'd purchase MCSG.

Regards!

...JRF...

[George Petrides_1]

The only safe poor man's MC/SG I have seen was having both server exactly the same HW configuration with external hard drives and physically move the HDs from one storage enclosure to another. If you get into sharing disks without MC/SG then anyone can wipe out the data by activating the group and frankly it won't be their fault for deleting the data. However, MC/SG is still the best way to go, unless there are not enough money in the budget for the implementation...
Hope this helps,
George