I've done the background reading, spent some time with GemFire so I know some of the quirks and what it can and can't do.
We are using it mainly for caching - both our own data and we intend to extend it to HTTP session and Hibernate L2.
We just hit the regions all have to live in all the servers in a server-group and cache proliferation problem, which has inspired this question.
How do people manage this stuff in production environments?
We are running a client server setup with multiple cache-servers in multiple server-groups with a couple of locators to keep track or them.
Multiply that by a dozen disparate environments and you can see that the number of things one needs to look after starts to get unpleasant without automation.
We have a couple of choices:
1. Is to use spring-gemfire and run everything in app servers. This fits in with our current deploy practises, but seems a bit wasteful of resources (and tcServer licenses). Configuration is better than it used to be (although we still end up using factory methods for the locators) but it is still awkward.
2. Launch everything from the command line using gfsh.
3. Launch everything with shell scripts (or batch files) and properties and cache.xml files - using gfsh like 2 but automated.
2 was intended as a joke btw...
Seriously though how have other shops handled this?
I can't find a recommended methodology in any of the documents I looked at. We are leaning towards 3 - it isn't that hard to get cache-servers and locators up, we can automate deployment readily and I can see them all in Pulse, but I'm wondering if there are gotchas to this technique and a better solution. I'd desperately like for this to be the last go-around for us on this particular topic so am looking for the voice of experience...
Also does anyone know why Pulse refuses to show the record count in replicated regions, but has no problem with partitioned? Is it a real bug or something silly we overlooked?