topic Serviceguard+SecurePath in Operating System - HP-UX

Serviceguard+SecurePath

Karthik S S — Wed, 04 Feb 2004 23:13:10 GMT

Hi,

I had posted this quiry in storage section,

http://forums1.itrc.hp.com/service/forums/questionanswer.do?threadId=402467

but did not get any response. So trying my luck with this new thread ....

One of my customer has problems with Secure path. He has a RP8400 systems with 2 hard partitions and has confgured a cluster. Each node has 2 FC HBAs system and both are configured under Securepath thus it is visible to the cluster node as a single controller. He runs informix database on this machine. The storage box connected to this machine is an EVA array. For simulaing a Failover scenario if he pulls out both the FC cables that is connected to cluster node1, the informics application just hangs also the LVM doesn't seem to be recognizing the failure. The expected behaviour for the customer is to failover the package to the other node when the first node's FC cables are disconnected. How do we go about this problem?

Can any one explain me what happens if the FC cables are removed from the HBA end?. How securepath responds to this in a clustered configuration? How do we configure EMS for securepath?. Also I am looking for some documents on Securepath+HP-UX+VA7xxx envirionment.

Pl. help

Thanks,
Karthik S S

Re: Serviceguard+SecurePath

Stephen Doud — Thu, 05 Feb 2004 08:28:28 GMT

I will not address the Securepath topic - I don't know that product.

I will however address a resolution to losing paths to disks by way of a document I wrote:

UMCSGKBRC00012483
TITLE: Creating a disk monitor for a ServiceGuard package

Edited to comply wit ITRC Guidelines and policies

Re: Serviceguard+SecurePath

Kenneth Platz — Thu, 05 Feb 2004 09:39:58 GMT

Stephen,

Excellent document - I think I'm going to add that to my library in case we decide to go that way.

Karthik,

You do realise that running a MC/SG cluster within the same chassis (ie, 2 nodes in your rp8400) is a potential risk? There *are* certain components within the rp8400 that, if they fail, they will require you to shutdown the entire chassis to replace. I believe the first PCI bridge is one of these components (I once found this out the hard way).

Re: Serviceguard+SecurePath

Sridhar Bhaskarla — Thu, 05 Feb 2004 10:34:54 GMT

Hi Karthik,

I don't know how SecurePath works but I worked with Powerpath with EMC.

By default serviceguard will not do anything if the connection to the disk system is lost. One way is to get the monitor setup as indicated by Stephan. Configure a service to monitor your application processes and exit the service if one of them are not running. Your application should be (can be made) able to identify if it is not able to access the data and quit itself. For ex., oracle will spill out numorous errors and may crash in such an event. If the SERVICE_FAIL_FAST_ENABLED is set to YES in your package configuration file, the system will toc and the package will failover to the other box.

-Sri

Re: Serviceguard+SecurePath

Duncan Edmonstone — Tue, 29 Jun 2021 05:25:42 GMT

I think the problem with Stephens solution is that it assumes you will be using PV links and physical volume groups (required for the EMS resource pv_summary to function correctly), which of course you won't be doing with Secure Path.

I suspect your only option will be to script your own monitor which runs the secure path commands that give you details on the links, and run a failover if both links are down for more than N seconds.

Of course you'll still have to setup node fail-fast as the failover can't unmount a filesystem if the paths to the disk are hung.

HTH

Duncan

Re: Serviceguard+SecurePath

Duncan Edmonstone — Thu, 05 Feb 2004 10:44:52 GMT

..and of course I meant service fail-fast (as Sridhar said)

HTH

Duncan

Re: Serviceguard+SecurePath

Karthik S S — Fri, 06 Feb 2004 23:57:26 GMT

Hi All,

Thanks for all those valuable inputs. I am sorry I could not reply you back immediately because for whole of yesterday I could not access the ITRC site :-( ...

I will go through this today and update you.

Thanks,
Karthik S S

Re: Serviceguard+SecurePath

Karthik S S — Sat, 07 Feb 2004 01:58:27 GMT

Hi,

I am using secure path 3.0 for the VA arrays. The installation creates a folder called autopath under /opt (autopath commands) and doesn't contain the spmgr binary. Is there any difference between secure path manager product for VA/XP arrays and EVA arrays?? When you run the installation script it gives an option for that. I think the spmgr binary is installed only when you select the EVA array. What I want to understand is that, does autopath create a single device file for multiple paths. In my case it didn't happen and all the the alternate link device files are also shown in the ioscan output. Someone who tried securepath with EVA array was telling me that it creates a virtual device for each physical device. file clubbing all the alternate links under it.

I somehow managed to check the failover scenarios by pulling the FC cables ( using PV sumary yesterday). But since at the customer end they have the EVA array I am really clueless how to go about configuring failover of package without usin PVGs.

Find the the document attached explaining the step by step procedure of what I tried.

Please help

Thanks,
Karthik S S

Re: Serviceguard+SecurePath

Karthik S S — Sun, 08 Feb 2004 23:06:33 GMT

Bringing the thread to the top .. :-)

-Karthik S S

Re: Serviceguard+SecurePath

Orrin — Sun, 15 Feb 2004 21:05:22 GMT

Hi Karthik,

Securepath will handle the alternate paths to the same host and thats it. From the discription of your test this seems to work as when you disconnect one FC cable, connection to the PV is still up.

As for failover to the secondary server this is a function of the clustering system,
and essentially the cycle that checks if the application is running on the primary node, when both FC are removed and primary node loses access, The secondary node will see the PV, the clustering software should notice that the application is down and restart it on the secondary node.

Have cnfigured this and works fine with Oracle and Tru64unix. On Tru^4 the clustering side is handled by the CAAD,
which monitors all the cluster applications.

Hope this helps,

Regards,
Orrin.