Now let's say we want to have 200 instances in a cluster. Are you proposing we create 200 groups?
No. I would create 4 groups, each having 50 servers. You could them upgrade one group at a time (or more if you like). I was merely trying to point out that groups are not similar to blue/green cluster system. You do not lose sessions while bringing down a group and you do not need any additional hardware for that.
... your solution only works because two different versions of the app with the same "runtime deployment name" is artificially isolated in a group that contains only one app.
Not sure I can see what you mean here. You 'push' the app into domain, meaning more or less onto the master node and then you decide which slaves (hosts) will actually have this app deployed and master node takes case of that. In case of WFLY you point to a group of slaves rather than single one though the groups play no role in 'isolating' anything. When you push the app into domain you basically say "Hey, whichever node will have this app deployed, it's name is gonna be 'test.war'." And based on this same name they are actually clustered and sessions are being replicated (+ some config to say you want to have it clustered ofc).
All I want is for Weld is to support this use case.
I understand that, I am just trying to get as close to your use case with WFLY/EAP as possible. This will allow us to see if the problem is general or specific for some servers.
|