Controlled access to shared external disks

[furtz]

Hi,

Hoping to setup a cluster with 2 nodes and 2 external disks (rx2600 and ds2100). Each node has a SCSI cable connected to each of the external disks. SCSI ID switch set to 'A' on one ds2100 and to 'B' on the second

I can boot each machine on its own and each node can see both the disks, can create a volume group/logical volume across the two disks and mount a file system on this new logical volume. But what I want to do is to control the access to these devices (Hot/Standby using ServiceGuard). How do I prevent the OS (and “Autoconfiguration”) on each node taking control of the disks on each node. Do I have to disable ioinitrc from the inittab?


Thanks in advanc

[Duncan Edmonstone]

If by 'taking control' you mean activating the volume group, then you need to edit the file /etc/lvmrc

Change the line:

AUTO_VG_ACTIVATE=1

To

AUTO_VG_ACTIVATE=0


This will stop all but the root volume group from being activated at boot time. Serviceguard will then control which host activates and mounts the volume.

Incidentally I hope you are not using the same SCSI bus that the rx2600s boot off for the shared disk, as I'm pretty sure thats not a supported configuration.

HTH

Duncan

[Patrick Wallek]

This is standard MC/SG configuration.

You disable AUTO_VG_ACTIVATION as Duncan has said (don't forget to add VG00 to the manual section in /etc/lvmrc so it will get activated). Then you control the VG activation and mounting of LV's through the MC/SG package control scripts for each MC/SG package that you configure.

[Patrick Wallek]

I should have added --- Since a MC/SG package is only running on one node at a time, the VG's for that package are active on only one node at a time.

[furtz]

Hi again,

Thanks for the quick reply. I am talking about "control" at a lower level. At the ioscan level.

Here is what I get when I boot both nodes.
On node 1:
disk 3 0/1/1/1.8.0 sdisk CLAIMED DEVICE HP 73.4GST373307LC
/dev/dsk/c3t8d0 /dev/rdsk/c3t8d0
On node 2:

disk 3 0/1/1/1.0.0 sdisk CLAIMED DEVICE HP 73.4GST373307LC
/dev/dsk/c3t0d0 /dev/rdsk/c3t0d0

What I want is to be able to see both disks on node one, create the volume group etc, do my vgchange -a y vg01. Then switch control to the other node using vgchange -a n vg01 on node one and vgchange -a y vg01 on the second node.

Thanks again,

[Duncan Edmonstone]

The 'CLAIMED' message is a little misleading - it doesn't actually mean the host has claimed the device in any way whatsoever, it simply means its an assigned device in the kernel and this host can talk to it - so can the other host, but it won't if the VG isn't activated - seeing disks from both nodes in ioscan is perfectly normal in a SG cluster. SG doesn't used 'vgchange -a y' - when a cluster is built it changes a volume group to a cluster volume group using 'vgchange -c y', and then either node can activate the volume group in exclusive mode with 'vgchange -a e' - I would have a read of the SG manual here to learn more:

http://docs.hp.com/hpux/onlinedocs/B3936-90079/B3936-90079.html

HTH

Duncan

[Uday_S_Ankolekar]

ioscan shold show "CLAIMED"for all the shared disk under SG and it's normal.

For service guard vgchange use "c" option instead of "a" man vgchange for detailed info.

-USA..

[furtz]

Hi,

Then I should see both 73GB disk from each host when both are up and running but I am not seeing this.
The one disk (/dev/rdsk/c3t8d0 on node one is /dev/rdsk/c3t0d0 on node 2?).
The second disk should be seen as something like c4t0d0 and c4t8d0.(no access to the machine right now)
As I said earlier I can see both disks if only one of the nodes is running. This would suggest that the SCSI cables are OK and it is some conflict/addressing issue when both nodes are running?

Thanks again.

[Duncan Edmonstone]

Have you not changed the SCSI ID of one of the initiators on the hosts? By default both will be set to 7 and you will have a conflict.

IIRC this can be changed from the BCH service prompt during boot. Can't remember the exact details but I'm sure someone will have them to hand.

HTH

Duncan

[furtz]

Hi I think we are on the correct path. I do not think there is a BCH prompt on the rx2600. I think it is an EFI shell but I could not find the correct commands. Anyone done this?

Thanks,

[Duncan Edmonstone · Accepted solution]

Hmmm, as with many things in integrity-land, what was once straightforward now seems fairly complex... the following document describes how to do this on a rx4640 - I'd imagine its the same on a rx2600...

http://www5.itrc.hp.com/service/cki/docDisplay.do?docLocale=en_US&docId=200000075159044

HTH

Duncan

[furtz]

Hi again,

Yes a bit of a struggle but we got there.
drvcfg -s drvr_handle cntrl_handle
is the command you need from the EFI shell for anyone doing this in the future.
It is all menu driven and is relatively easy with the aid of the document and a knowledge of the hardware paths.

Thanks Duncan