topic Re: node is grabbing the system disk of the other node in Operating System - OpenVMS

node is grabbing the system disk of the other node

PhilHowes — Fri, 18 Jun 2010 13:31:49 GMT

Hi,
I've got a problem with my alloclasses (I think). I have 2 rx4640's in a cluster, with both nodes having 2 system disks. On each node, the system disks are $1$dka0 and $1$dka100. I started a shadow copy between the disks on one one node. But when I tried to do the same on the other node, dka100 said that is was already remotely mounted, and I couldn't do anything with it. I've seen this before with CD drives and i fxed it by setting device_naming 1 and set/class dqa0. I tried a similar thing for dka0 but it still names the systems disks $1$dka on both nodes, rather than $xx$dka. Any suggestions?
Thanks.
Phil

Re: node is grabbing the system disk of the other node

Ian Miller. — Fri, 18 Jun 2010 14:18:53 GMT

If I understand your question then

SYSBOOT> set/class pka

is one answer

Re: node is grabbing the system disk of the other node

P Muralidhar Kini — Fri, 18 Jun 2010 14:36:16 GMT

Hi Mark,

To modify the allocation class of a particular node, you need to modify the
SYSGEN parameter ALLOCLASS. This is a static sysgen parameter.
Hence, once you modify this, you need to reboot the system for the changes to
take effect.

To modify the Node A' allocation class to 5, do the following
$MC SYSGEN
USE CURRENT
SET ALLOCLASS 5
WRITE CURRENT
$

Hence in your case, you need to do the above on each of the cluster nodes,
changing the allocation class of each node to a unique value.
Say on node A you set it to 1 and on node B you set it to 2 and so on.

Another point to consider is that, once you change the allocation class of a
particular node say node A, you have to reboot all the nodes in the cluster.
This is required so that all the nodes in the cluster get updated with the fact
that Node A's allocation class has been changed.

Hope this helps.

Regards,
Murali

Re: node is grabbing the system disk of the other node

Hoff — Fri, 18 Jun 2010 14:38:46 GMT

I can go a couple of different ways with this question...

You have two hosts each with a pair of disks and each of those pair named DKA0 and DKA100, and both in the $1$ allocation class, and these hosts are clustered?

If that's the case...

Yeah, that's a problem. Potentially a big one.

Get one of the boxes offline RIGHT NOW.

You're very close to a corruption if you haven't already encountered a corruption, as you've told OpenVMS that these are two paths to the same physical disks.

If these are local (non-shared) disks, get into separate allocation classes, either via host allocation classes or disk allocation classes.

If these are not FC SAN disks, then get these disks into unique (host or disk) allocation classes that aren't $1$ and $2$ classes; out of the FC SAN disk class $1$. This as (while you're messing with your allocation classes) you might as well get out of the classes expected for FC.)

Now if you have a shared SCSI bus here, then you can have these devices in the same allocation class (usually a port allocation class), as these devices are the same disks. I'd still get out of $1$ here, but (if these are two paths to the same disks) you're not in any particular trouble.

Allocation Class Info:
http://labs.hoffmanlabs.com/node/349

Re: node is grabbing the system disk of the other node

Shriniketan Bhagwat — Fri, 18 Jun 2010 15:05:35 GMT

Hi,

Try this

$ mc sysgen
SYSGEN> SHOW ALLOCLASS
SYSGEN> USE CURRENT
SYSGEN> SET ALLOCLASS 1 ! set different values for each node.
SYSGEN> WRITE CURRENT
SYSGEN> SHOW ALLOCLASS

ALLOCLASS is not dynamic parameter. Hence you need to reboot the machine after setting the ALLOCLASS parameter.

Regards,
Ketan

Re: node is grabbing the system disk of the other node

Hoff — Fri, 18 Jun 2010 16:09:34 GMT

>ALLOCLASS is not dynamic parameter. Hence you need to reboot the machine after setting the ALLOCLASS parameter.

Or reboot the cluster to clear all knowledge of the previous $1$ stuff.

Re: node is grabbing the system disk of the other node

Robert Gezelter — Fri, 18 Jun 2010 16:14:35 GMT

Phil,

I concur with what has been said.

That said, I will add the somewhat obvious "And run ANALYZE/DISK_STRUCTURE on all of the disk volumes on the each system; be alert for files that may be internally corrupted."

- Bob Gezelter, http://www.rlgsc.com

Re: node is grabbing the system disk of the other node

PhilHowes — Mon, 21 Jun 2010 06:56:08 GMT

Firstly, Thanks everyone for your responses. All very useful.
I've got to the bottom of the problem. I had set the ALLOCLASS differently for both nodes, but when we installed our aplication files, the alloclass value had been set to 1 in the modparams file that we install on both nodes. However, I tried resetting the alloclass values on both nodes, and although it did set it to the correct values, the problem with the system disks being $1$dka on both nodes persisted.

I then tried the set/class pka xx command with different values for each node, and this was successful!
Luckily this is a new cluster that I am building, and isn't yet live, so I was able to play around with it.
Thanks again,
Phil

Re: node is grabbing the system disk of the other node

P Muralidhar Kini — Mon, 21 Jun 2010 07:14:15 GMT

Hi Mark,

Good to know that the problem is resolved.
Thanks for posting the solution as well. ("set/class pka" command as advised by Ian).

Also, Refer the following link which says how to thank the forum -
http://forums13.itrc.hp.com/service/forums/helptips.do?#28

Regards,
Murali

Re: node is grabbing the system disk of the other node

P Muralidhar Kini — Mon, 21 Jun 2010 11:45:15 GMT

Hi Mark,

>> Luckily this is a new cluster that I am building, and isn't yet live,
>> so I was able to play around with it.
To finish off, as the cluster was in a state where both the nodes saw the
disks are $1$dka0 and $1$dka1, there could have been corruption on these
disks during this time period.

Execute the following command on both the disk and ensure there is no
corruption -
$ANALYZE/DISK/LOCK

Regards,
Murali