The expectation is that participants are configured to point to a load balancer cluster with each BalancerMember hosting a coordinator for starting Long Running Actions. The coordinator that is picked to start the LRA is then responsible for completing its lifecycle, ie join and cancel/close.

Note that this should work out of the box provided the participant is configured to use the balancer (via the -Dlra.coordinator.url system property, or otherwise) but it has never been tested.

Also note that there is currently no expectation that different coordinators can finish the lifecycle of an LRA started by a different one, for that we'd need JBTM-2997 (Ensure that any LRA coordinator instance can control any LRA).

Finally if we use Wildfly's Undertow subsystem for the load balancing then we can also provide a quickstart (I have managed to get the load balancing working using Apache httpd so may end up just following up with the WildFly quickstart in a separate JIRA).