EMC related issue

[Ragni Singh]

we have 2 systems that access one set of disks. Sometime failover works, other times it don't.

I want to know if anyone here is aware of any EMC internal command that I can run which will tell when 2 systems are trying to access the same file-system.

[generic_1]

I dont think one exists. But I think you have a Vgchange issue in that other other system is not being deactivated on the volume group or whatever you are using for a locking mechinism is screwing up. Worse case only one vg should be activated in write mode.

[Nobody's Hero]

We have 2 EMC systems.
Not aware of any such tool. I imagine you could write something for exclusive rights to a vg. As you know, EMC like to keep things a great secret. I know of some commands internally to the emc, but not what you are looking for.

[Pedro Cirne]

Hi,

I agree that probably you have a problem with vg activation. Is this a SG cluster?

Check if you have AUTO_VG_ACTIVATE=0 on /etc/lvmrc on both node.

Enjoy :)

Pedro

[Matti_Kurkela]

(Pedro, this is a Linux forum: there might not be AUTO_VG_ACTIVATE setting or even /etc/lvmrc on a Linux distribution.)

On HP-UX, ServiceGuard is more tightly integrated to the OS: when you are creating a cluster, you can set the volume group to a cluster mode, like "vgchange -c y vgsomething".

However, this locking does not go to the EMC level: it works only among members of the cluster. Two unrelated hosts will happily use the same disk at the same time, if mistakenly configured to do so. We unfortunately had this fact experimentally confirmed :-(

A big part of ServiceGuard setup is ensuring that the volume group activation/deactivation scripts are absolutely bullet-proof. One of the nodes may fail totally at any time, with no time to unlock the locks it has set up.

Because of this, the remaining node(s) *must* be able to grab the disks held by the failed node, if (and only if) convinced that the node that held the disks has failed.

This also means that if a node detects it's in a situation where the other nodes assume it has failed, it must completely stop using any cluster disks *immediately*. This is why ServiceGuard has the TOC functionality for HP-UX and the deadman driver for Linux: the chosen method is to intentionally crash the server when necessary.

[Ragni Singh]

thanks all for the time. Like you all said, EMC don't have any commaonds of such.