Automating patch installation.

[siostra]

I would like to minimize downtime for my Solaris9 systems when installing patch cluster.
I am thinking about a running a script with following steps:
- detaching mirrored root disk (metadetach)
- installing patch cluster on the detached disk.
- putting back the mirror (attaching back the patched disk to root disk)
- rebooting system (resyncing disks) : Problem- which disk will be the master? Should I boot from the second disk?
What do you think?
Does anyone have experience doing this?

 

[bfitzmai]

When you set up your mirror, one of the disks become the primary mirror. You can find out which one by running command "metastat" or running the Solaris Management Console (smc).

Don't understand why you want to detach your mirrors since the cluster will be downloaded as a single compressed file and you have to uncompress the file and do the install. I agree the write time is slower with mirrors but, I am not sure just how much time you will be saving.

 

[spamly]

It may work but I wouldn't count on it.

What you're trying to do is give yourself a good backout plan. You've got the general idea right, but there are a couple of gotcha's.

First, in the event of a problem, you'll ask your server to begin to boot up off of your backup drive. Unfortunately, it will reference the /etc/system file and see that you have Solaris Volume Manager configured. It'll then use your metadb's (also defined in your /etc/lvm directory) to figure out which drives are managed. Referencing the /etc/vfstab it will try to mount up your root partition metadevice as defined by your metadbs (probably /dev/md/dsk/d0). As you know, you've detached your backup drive so it will try to mount the only drive that it knows about, which unfortunately, is your patched problem drive.

Talk about a pain, eh?

 

[coffeysm]

I agree with bfitz and spamly you are setting yourself up to create more work for yourself. I would either setup an identical test environment if possible. You might also have to take a shot and back out of the patch cluster if something gets messed up.