How cluster (server group) isolation in Wildfly 9 could be done?

[salamont]

We are developing a system consisting of a number of packages. Each of them is operated within a different server group, however they together form the same cluster (we are using TCP stack with TCPPING discovery - the only possible way for us now).

This configuration has however serious problems, long delays, timeouts, strange errors, like:

WFLYCTL0348: Timeout after [300] seconds waiting for service container stability. Operation wil

l roll back.

ISPN000329: Unable to read rebalancing status from coordinator

It seems that it helps us when the individual server groups are isolated (e.g. when we switch off the other server groups). This also makes sense - using TCP, every packet needs to be sent several times - for each virtual server in cluster separately.

Unfortunately, I don't know how to efficiently separate the particular server groups. Thanks for help.

[pferraro]

Each server group should use a distinct set of initial_hosts.

e.g.

<protocol type="TCPPING">
 <property name="initial_hosts">...</property>
</protocol>

You can parameterize this value using a ${...} expression, the value for which can be provided via system property.

[salamont]

Thank you, Paul.

This is what I thought. But it means - in the case of managed domain - to have a separate profiles for each server group, right?

[pferraro]

Not if you parameterize the initial_hosts via system property.

[salamont]

OK, I didn't know it is possible to bind it to a server group. Now I have read that it is much more flexible than I thought. Many thanks!!!