Serviceguard Active-active..

[Joe Kanakaraj]

Hi Guys,

I would like to know if any of you have done an active-active configuration on service guard, we seem to be getting an application that might require it.

Also is it possible to route outbound traffic from a service guard VIP to a client instead of using the physical IP's of the of the nodes for outbound traffic.

Documents insight and experience have points.

Thanks,

Joe

[Calandrello]

Friend config of IP of the package is configured in the archive .assci of the proprio package. to configure the HEARTBEAT IP you must configure in the archive .ascii of cluster that normally this as clustername.ascii

[Keith Clark]

Hey Joe,

We have been running an Active-Active config since '99 for SAP. We run the DB on node A and the CI (Central Instance) on node B. Node A is the backup for Node B and vice versa. If Node A fails, the oracle DB fails over to Node B. If Node B fails, the CI fails over to Node A.

I know that we used to have quite a few issues with routing years ago, but it seems to have improved in the newer versions of SG. I'd be lying if I said that you can have total control now, but I do know that the issues we faced three - four years ago have not reappeared.

Keep in mind that when you are sizing your environment, you need to have enough free memory for the second application to start. If you do not regulate the memory resouces, chances are your fail-over will fail. The rest is just disk, network and cpu, none of which should stop the applicaion from starting, but will affect performance. We set a target of 75% of normal performance levels for both applications (during fail-over) when sizing the servers.

As far as docs, check out Chapter 1, Mutual Fail-over scenarios using the two package concept in the "Managing Serviceguard Extension for
SAP": http://docs.hp.com/en/T2803-90002/T2803-90002.pdf

It is obviously geared to SAP, but it lays out the concepts of mutual failover quite nicely...

HTH,

Keith

[Joe Kanakaraj]

Thanks keith for the insight. What you have described is a standby failover for individual packages where each node acts as a standby for the other while each node is also running package.

My question relates to more of actually running a package which is available for both nodes at the same time. This is not an Oracle RAC setup. THis is a Tibco EAI setup and it is tibco's recommendation that we have individual packages for thier application modules and a shared read only area for thier scripts which are mostly for startup/shutdown/monitor for processes. The shared package in question is read only and is static.

This gives rise to another question, Is is it possible to give the same shared storage presented to both machines and which are running packages in a 2 node cluster environment, so that in case of a failover the shared part is still available.

The IP part is, Once I configure the package I get the VIP and there is a partcular server which accepts connections only from the VIP since there are strict acccess list policies on it and cannot accept from the Physical IP's of the nodes. From My understanding it is always the case that inbound traffic can be via VIP but outbound is via physical IP, correct me if I am wrong. My question...is it possible to route back via VIP. As mentioned before we are still in planning stage so I have not tried it yet to see if it might work with the newer version of MCSG.

I hope I have been clear and have not confused you guys.

[Andrew C Fieldsend]

I don't know Tibco very well, but I do have experience with active-active SG. I would imagine that you would need to configure SG similarly to an Oracle RAC config, with shared storage managed by multiple application packages and load balancing and failover handled by the client application.

Obviously there are a few variations possible with this setup, depending on the details of the application.

[nanan]

Hi
It seems you to want to configure sharable filesystem be able to be read by all of the cluster nodes.

As you stated, Tibco is one of MES application isn't it?
You can configure the sharable filesystem with CVM(cluster volume manager) or VxVM
MC/SG 11.17 allow to setup that for you

Regards
nanan

[Keith Clark]

Hi Joe,

Sorry for the misunderstanding. I think what you are referring to is known as a multi-node package. We do have something similar, but just for web services and no SG required. We use a NetApp backend and mount the doc-root on the two servers and use a Foundry ServerIron for load balancing. The Foundry monitors the http service and will redirect client traffic if one node fails.

I realize that this is a little different from what you need to do with Tibco. I do not have any experience with SG providing multi-node services as we have always used appliances (either Foundry or F5).

For the shared file system you could use NFS (mount readonly), assuming that you have an HA NFS service available (SG or NetApp), you may be able to use straight LVM, since it's read-only, but more then likely you will have to use Veritas Volume manager and clustered file system.

As far as the networking goes, for multi-node packages there is no VIP, since there is no fail-over. Perhaps you could verify that you will end up with a multi-node package for the Tibco app servers and a fail-over package for the shared scripts?

Check out multi-node configs in "Managing Serviceguard Twelfth Edition": http://docs.hp.com/en/B3936-90100/B3936-90100.pdf

Thanks,

Keith

[avik]

i guess you want the outbound traffic to be directed to a particular machine/ip. i would do a "route add" to achieve that task.

[Keith Clark]

Hi Avik,

I believe that the issue is that the outbound traffic must come from a specific IP. The application is using some kind of IPFilter like tool to limit what IP's can connect to it. I know in our case we had to add the physical IP's and the VIP's to IPFilter rules in order for it to work, but there may be a work around for this. In our case, we weren't really concerned...

Thanks,

Keith

[avik]

*outbound traffic must come from a specific IP* : i guess adding the route by giving the specific ip as the default gateway should help him to achieve that.