os/oracle asm issue ??

[whiz_kid]

Node node1 and node2 are in MC/SG cluster (active/passive)

package pkg1 runs in node1 with a failover enabled to qrhqsd51

pkg2 runs in node2 with a failover enabled to node1

Each node runs seperate oracle instance using ASM, ie, each node runs seperate ASM instance as well !

Each package holds - vgASM1 and vgASM2 respectively. Oracle ASM uses the raw logical volumes from these vgs.

Test Scenario:

CU tried failing over pkg2 package to node1 .

The package failed over successfully. DBA was able to bring up the database.

DBA shutted down the database and ASM instance for this database post their test (in DBENV6_PKG2) - (customer_run and halt scripts are not yet defined in the control file).

cmhaltpkg failed with "failed to halt the package" error.

The control log file reported below error

Jan 15 17:12:54 root@node1 .qr.qrgrp.local volume_group.sh[2406]: Deactivating volume group vgASM2
vgchange: Couldn't deactivate volume group "vgASM2":
Device busy
Jan 15 17:12:54 root@node1 .qr.qrgrp.local volume_group.sh[2406]: ERROR: vgchange -a n vgASM2 failed, trying again.

We tried manually de-activating vgASM2 vg, which agained failed with "device busy " error.

DBA brought down the second database which was already running, and then the ASM instance.

After that we were able to de-activate the vg and failback the package to the original node.

Apparently, it looked more of an issue related to the ASM instance. A call was logged with oracle and oracle communicated, that this is an issue related to OS file handler.
here is the reply from oracle
====

This problem only occurs on certain operating systems:

Solaris does not have this problem LINUX solves the problem by use or ASMLIB

There are only two methods that can provide a consistent release / close of file handle / descriptors:

* via the implementation of ASMLIB
* via the implementation of ODMLIB ( as this is a Veritas/Oracle product ...
ASM and Veritas must both be installed )

ASMLIB's are written by operating system vendors and as such cannot be provided by Oracle except in the case of Linux as we provide Linux support and have access to the source code



hs anyone faced this issueee?

[Steven E. Protter]

Shalom,

I suspect one of two major issues.

1) Missing software which you identify as asmlib. So why are you not talking to Oracle about a solution? Unless your configuration is off the reserveration as in the Oracle support matrix.
2) You have misconfigured ASM in such a way as to cause a resource conflict at close. You have not provided enough information to determine this. The logs may provide a way out of thsi situation.

SEP

[whiz_kid]

yes, it cud be a missing patch or a misconfigured ASM to cause that resource conflict. do u know the patch number that does "asmlib" in 11.31? Oracle doesnt!

[Duncan Edmonstone]

I'm not sure what Oracle support are talking to you about, but you *cannot* have more than one ASM instance per node - thats not supported.

Oracle document that themselves in metalink note 370915.1

So what you are trying to do is never going to work on any platform...

AFAIK ASM is only supported with Serviceguard when used with Oracle RAC and Serviceguard extension for RAC (where you effectively have one ASM instance across cluster nodes controlled by Oracle clusterware.

You need to reconsider how you are going to acheive failover between these nodes

Consider filesystems or just taw disk without ASM.

HTH

Duncan

[whiz_kid]

I have ONLY ONE ASM INSTANCE running in one node at any time which is handling two databases in case of a failover, which is supported.

[Duncan Edmonstone]

Ypu are correct - I assumed you were trying to failover the asm instance with the database, whereas I guess you're just moving the disk group across...

Nevertheless, my second point stands - ASM isn't supported in a Serviceguard failover cluster - only when used with Serviceguard Extension for RAC in an Oracle RAC Cluster.

So you still need to rethink this.

HTH

Duncan

[whiz_kid]

I dun hv any issue in failing over the package, the issue is with the failback...

[Duncan Edmonstone · Accepted solution]

I don't know how many ways I can say this... ASM is NOT SUPPORTED with Serviceguard when used with failover databases. Whether you an get it to work or not is irrelevant if your customer wants to have support from HP on this, you're going to have to either go to an active/passive RAC cluster or use filesystems or raw LVs without ASM if you stay with just a failover database druven by Serviceguard.

HTH

Duncan

[Duncan Edmonstone]

See the table in the support matrix here:

http://docs.hp.com/en/4446/ECMToolkitCompatibilityMatrix.pdf

Note the comment about 'Without ASM'

Also look in the notes here:

http://docs.hp.com/en/5971/SG-SGeRAC-SMS-EMSSupport.pdf

Note that ASM is *only* mentioned in the context of RAC.

HTH

Duncan

[whiz_kid]

duncan

are you sure that its an issue related to service guard?

assume i hv two databases in the same server using rlvols from two different vg. will i be able to deactivate the vg after i shut down the database running ontop of that vg?

its not an issue related to SG! its a bug in the operating system/oracle !

Oracle is clearly stating that it works fine in Linux and SUN!

[Duncan Edmonstone]

>> are you sure that its an issue related to service guard?

- No - only that its not supported with Servicgeuard

>> assume i hv two databases in the same server using rlvols from two different vg. will i be able to deactivate the vg after i shut down the database running ontop of that vg?

- I don't know...

>> its not an issue related to SG! its a bug in the operating system/oracle !

- One person's bug is another's feature. It may be down to differences in design of the IO stack and volume manager rather than 'a bug'. That's just conjecture though...

>> Oracle is clearly stating that it works fine in Linux and SUN!

- where are they stating this? Can you point me at a metalink document?

HTH

Duncan