failover disks to alt path when upgrading firmware

[Kevin Jacobson]

Hi all,

I have a SAN environment with an XP256 and 4 brocade switches in fabric mode. There are 12 HP and 9 SUN server attached.
My question is...
Is it nessesary/common practice to fail the disks over to the alternate path during a firmware upgrade??

The servers that are running Sysbase will be shutdown since Sybase doesn't hndle scsi resets well/or at all. The concerns are with the Oracle servers.

Thanks in advance.

Kevin

[Ashwani Kashyap]

Well, believe it or not , you will be surprised by oracle , it doesn't respond too well to scsi resets either in most cases .
Ypu are better off , doing the upgrade in single user mode or at least shutdown the database and apps .

[Kevin Jacobson]

So.. any maintenence such as adding a lun and doing a scsi reset
after Secure manager... I should fail over to the alternate paths?
I am not able to go to single user mode or shutdown the databases on most of these servers at any time. (reason for high availability). I have had conflicting answers from both HP techs and other operating groups here. Some say fail the disks over and some say let them fail over automatically. I have done scsi resets several times on the oracle boxes without any problems but I suppose there is always a first time for something to happen..ouch.

Thanks

[Ashwani Kashyap]

Kevin ,

If you have to do this , then I strongly suggest to failover to alternate links manually . I have seen cases where all the links did not failover to alternate paths automatically and had to be failed over manually.

And yes , backup your boxes before that

[Stanislav Markovic · Accepted solution]

Our environment is quite similar (HP/Sun mix, Brocades, XP arrays, Informix/Oracle DBs).
We recently completed a firmware upgrade on all our Brocade switches. We ran a script on HP-UX hosts to failover to alternate paths. You can get the script from HP support, it is called switchxp.sh. On Suns it was much easier, since we are using VxVM, so it was only necessary to do vxdmpadm disable/enable on each host.

HTH

[Kevin Jacobson]

Thanks for the information Stanislav,

I didn't know about the script which would make things a bit easier.
Unfortunetly, from what I'm understandong, the fact remains that it is common practice (manditory if no risks can be taken) to failover the disks to the altenate paths everytime there's a scsi reset no matter if it's Oracle or Sysbase????
On a SUN with Vertas, I believe failing over it is as simple as disabling one controler. Does anyone have experience with that?

Thanks again!

[Stanislav Markovic]

In my experience it really matters how heavily your DB system is loaded. We had issues here with Informix even while simply removing LUNs, on a different system connected to the same XP port.

Another reason to do a manual failover is to avoid a storm of messages if you have any kind of system/network management like ITO implemented.

Regarding your Solaris/VxVM question, it is really quite trivial:

Identify controllers:
vxdmpadm listctlr all

Make sure both controllers are ENABLEd, and then disable/enable controllers:
vxdmpadm disable/enable ctlr=

Take care,
Stanislav