MC Servicegurad - configuration

[Omar Alvi_1]

Hi All,

I'm going to try my hand on ServiceGuard shortly. Needed some guidance. Is there some bare basics guide I can follow to grasp the concepts AND perform a successful, running configuration.

The environment will be one for Oracle Real Application Clusters. I'm going to configure the MC ServiceGurad already installed as a prerequisiste for that setup.

Also, are there any different versins for MC ServiceGurad, OPS, RAC ... what're the differences? Does Oracle definitely require an OPS or RAC, or is my simple MC ServiceGurad enough?

Apprreciate all the help.

Thanks and Regards,

-Alvi

[Sridhar Bhaskarla]

Hi Alvi,

Following guide should give you all the details you need. There are example configurations of cluster and package etc., so you can build your cluster using them.

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

Real Application cluster -RAC is the successor for Oracle Parallel Server - OPS.

Does oracle definitely require OPC/RAC?. It depends on your high availability requirements. With simple MC ServiceGuard, if a node fails, there will be a downtime involved while the package moves from the one node to the other. With RAC, it will continue to run even after a hardware failure.

Below document gives you more details on RAC.

http://docs.hp.com/hpux/onlinedocs/T1859-90017/T1859-90017.html

-Sri

[Steven E. Protter]

ServiceGuard provides high availability not 100
% availability.

RAC is kind of Oracle's way of providing high availability its way.

AS noted SG provides failover abilities, there is a window, the amount of time it takes to start the database on the failover node during which user requests for data will simply hang waiting for a database to serve them.

There are a few important things to do concerning SG.

1) Make sure your shared storage is accessible and is set up correctly to allow both nodes to access it.
2) Make sure your heartbeat connection is not the public lan and has little or no congestion. Best to task a NIC card to this job.
3) Test failover in a non-production situation. Just because you have generated the scripts and done the pre-requisites is no guarantee anything will work. You must test your setup before a node fails.

There are other clustering solutions, though I kind of like SG. Veritas has a decent though expensive product.

SEP

[Kent Ostby]

Alvi --

There are different versions of SG to run with Oracle RAC versus running with any other application.

There are several gotchas that new SG engineers run into from time to time.

Here are a couple:

1) Since you are putting I/O on a shared bus, the SCSI card on one of the machines needs to have an address other then "7". Otherwise, you will have a SCSI address conflict.

2) You should set the default heartbeat up from 2000000 to 8000000 as the lower setting may cause some ServicGuard failovers of your package when you have a brief networking delay.

Take care and feel free to post here as you have questions.

Best regards,

Kent M. Ostby

[Ted Buis]

You need supported disk configurations for ServiceGuard, so you should check with your HP Solutions Architect to confirm you plan a supported configuration. Taking the class is the best way to learn about ServiceGuard. You can even take the class remotely now, with an instructor assisting.
http://www.hp.com/education/

I recommend three NICs in the configuration, one for data, one for heartbeat and at least one unconfigured for failover.

There is a ServiceGuard with Extensions for RAC if you plan to do RAC. ServiceGuard does not require RAC, but RAC can make the system more highly available.

[Sanjay_6]

Hi,

If you have a contract agreement linked to your id, you can take a look at this doc related to SG from hp recovery handbook.

http://www2.itrc.hp.com/service/iv/docDisplay.do?docId=prodITRC/DE_SW_UX_swrec_EN_01_E/ServiceGuard.pdf

Hope this helps.

Regds

[Omar Alvi_1]

Thanks everyone ...

really got a good start into HA.

Sanjay, don't have special previlidges, wasn't able to acess the link.

I had one query, about the peculiarities of Service Guard.

I mean, what of the normal OS is different.

Like I noticed that the shared volume group doesn't show up in /etc/lvmtab

Nor, I think the filesystems are activated as usual.

vgdisplay doesn't work either. Anyway I can know whats different?

Thanks and Regards,

Omar Alvi

[Sridhar Bhaskarla]

Hi Alvi,

Shared Volume groups must be present in /etc/lvmtab. But they won't (shouldn't be) not be activated on the secondary node unless you run RAC.

'strings /etc/lvmtab' should show all the VGs shared between the nodes. However, 'vgdisplay' on those nodes will work on only one node where the package is running or they have been manually enabled with '-a -e'. If you find them activated on both the machines and unless you are running RAC, then something is not right. ServiceGuard will try very best to avoid such situations.

This is for LVM configuration. VxVM configuration is different as /etc/lvmtab doesnt' come into picture in that case.

-Sri