Who has implemented the 6.5 Broker in an HA configuration? on Solaris Sun Cluster? with multiple Broker Server instances?

I was just reading the 6.5 HA document, and it states the broker monitor process will be running on both sides of a two node cluster.

My question is, how does a broker server instance fail over?

If you have two broker server instances (A and B) running under the broker monitor on nodeA and broker server instanceA fails:

Will only Broker Server instanceA fail over to node B? And you would now have a broker server instance on each physical node of your cluster? (BrokerServerB on nodeA and BrokerServerA on nodeB)
-OR-
Will both broker server instances need to fail over? Hence the only reason for the broker monitor on both nodes is a ‘quicker’ fail over?

Thanks!
Ray

Ray,
I have ours set up this way. The awbrokermon process runs on both nodes. Each server’s awbrokermon.cfg has all the broker servers configurations for both servers. Then I use the monitor-start-servers=no flag so that the awbrokermon process doesn’t auto-start the broker servers.

I then use Vertias Server cluster to control the failover of the individual broker servers. The reason at least for me, is that I have separate instances running on both servers in an active-active configuration (not load balanced). So server A’s stuff can failover to server B and vice versa. And they can be done individually. So I can take one IS instance and its broker over to server B and leave the rest on Server A.

Mark,

OK… so I guess it works…

But is it ‘webMethods magic’ for territories and gateways? or is each broker server a logical host definition?

Which would mean you go from one cluster service controlling the brokermon process to multiple cluster services, one per broker server instance…

And are all the details for setup called out in the HA doc? I was skimming and didn’t notice … or is this the ‘magic’ of Professional Services?

Ray

Ray,
You bring up some good points about gateways and territories. I know your environment has a lot more of that. :proud: Right now I don’t have any gateways and territories in prod. I’m moving some up pretty soon. I’m planning on using the hostname switch in the awbroker.cfg to provide the virtual hostname to the brokers. Without that it picks the real physical hostname which means the above active/active wouldn’t work.

I’m in the process of testing it out. :eek: I’ll let you know how it goes. Hopefully I will not need professional services to come in and type the parameter for me. :rolleyes:

Mark,

After doing some more reading, it looks like each broker server would have to have it’s own logical hostname … which could get ugly if you have 50+ broker servers. :eek:

Has your testing discovered I am wrong? Please, tell me I am wrong…

Ray

No I think you are right. I haven’t seen any other way to do it. The gateways will pick up the physical hostname if the flag in the awbroker.cfg is not set to a virtual ip.

On the bright side, you get to know your network folks even better. :eek: