HPUX 11.23 Serviceguard Oracle 9i and MSA1000

[cpdla]

Hi 

We currenlty have an rx2620 connected to a single controller MSA1000 through a SAN switch.

The rx2620 is running 11.23 and one instance of Oracle 9i.

We would like to make this solution HA using ServiceGuard in a two (2) nodes cluster.

we purchased a second rx2620 and another MSA1000 and installed them in a different location but in the same building.

1-

Clients connect to the oracle instance using a specific hostname.

Now with serviceguard how are they going to connect if the package switches to the second node with a different hostname?

Is there a way to keep the clients connecting using the same hostname?

2-

What is the best way to configure the new MSA similar to the first one?

What is the best tool used to manage configure the MSA?

3-

Any recommendations gotchas we need to look into in such scenario?

Thank you in advance.

[Matti_Kurkela]

1.) In a Serviceguard set-up, you can (and normally should) define a "package IP address" which moves along with the service in the failover package. In effect, the service is reachable using that IP, no matter which node it's currently running on.

You would assign the hostname used by the clients to the package IP address, not to the static IP of any particular cluster node.

When transitioning from single-host to clustered operation, you can make this change in various ways:

• you could assign a new IP address and hostname for the service package and change the clients' configurations to point to it
• or you could assign a new IP address, but change the DNS to make the old hostname point to the package IP address instead of the old host IP, then assign a new name for the node itself
• or you could switch the IP addresses around, so that the old host IP address becomes the package address, and assign new IP addresses and names to both cluster nodes.

The choice is yours, depending on e.g. how difficult it is to change the clients' configuration and how restrictive your site's hostname/IP address management policies are. I would recommend picking the method that produces the cleanest result, not necessarily the one that requires the least amount of work!

2.) Thinking about this may be premature: see below.

3.) If you are planning to use two separate MSA1000s, the important question is: how is the data going to be kept in sync between the two MSAs? Serviceguard does not help there: it is designed mainly for situations when all the nodes connect to a single shared storage system. You could add cross-connections so that both storage systems are reachable from both nodes, and use MirrorDisk to replicate the data. There is also MetroCluster, an extension of Serviceguard for managing replicated storage systems... but I'm not sure if it works with MSA1000s: it's intended more for HP EVAs, EMC Symmetrixes and other large enterprise arrays.

You should download the "Managing Serviceguard" manual (available as PDF from http://www.hp.com/go/hpux-serviceguard-docs ) and read the chapters on Serviceguard cluster planning.

[cpdla]

MK,

Thank you for the answers.  They are very helpful.

We will be using MirrordiskUX to sync data between both MSAs.

[Matti_Kurkela]

OK, then the next question is: what are you going to use as a cluster lock?

A cluster lock acts as a tie-breaker: in two-node clusters this is very important, since network failures can make both nodes in a two-node cluster assume "I'm fine, the other node has failed" - that is known as "split-brain" situation and it causes data corruption, so it must be avoided at any cost. Serviceguard uses a cluster lock determine which node is allowed to run the service in this situation.

A cluster lock can be implemented as a disk-based lock, or as a separate Quorum Server. A Quorum Server is a small piece of software that can run on a HP-UX or Linux system: it takes a minimal amount of resources. If you happen to have a VMware cluster or similar high-availability virtualization platform, one minimal Linux VM with the QS software installed on it would be a very nice way of guaranteeing that the QS itself is highly available. Technically the QS does not necessarily have to be highly available: it just needs a reasonable guarantee that it won't fail or be inaccessible at the same time as your cluster node(s). Some amount of physical separation would be useful, and redundant network connections would be important.

Disk-based locks would be more problematic in your situation: it would tend to either make one of your two MSAs a critical component (= if that particular MSA failed, the cluster could not run), or add a requirement that both MSAs must be accessible for the cluster to work. Either of those would be less than desirable in your configuration.

You should also take a look in your network topology: it would be less than ideal if your cluster ends up dependent on the functionality of a single switch or other network device.

All this is covered in the planning chapters of the "Managing Serviceguard" document I mentioned.

Another issue in your configuration is the version levels: both Oracle 9i and HP-UX 11.23 are soon going out of vendor support... but I guess this is secondary if you have a critical application that strictly requires Oracle 9i. I hope you have some kind of plan for the future of your systems.