This is follow up on discussion in WFCORE-1912. There is difference in behavior if deployment is deployed like:

deploy ~/tmp/mdb1.jar --unmanaged --headers={rollback-on-runtime-failure=false}

and if it's deployed by copying to deployments directory and setting rollback-on-runtime-failure=false in model:

/subsystem=deployment-scanner/scanner=default:write-attribute(name=runtime-failure-causes-rollback,value=false)

If it's deployed by the 1st way using CLI deploy command then If deployment is missing some dependencies (like connection factories, queues/topics) and those are added later then deployment is able recover from it and start working without need to reload/restart server or redeploy.

However if it's deployed the 2nd way then deployment does not recover when missing dependencies are added. It looks like attribute runtime-failure-causes-rollback does not have the effect in this case.

This is causing problems when Artemis is configured in colocated HA topology with replicated journal where it takes some time for Artemis to activate but deployment which depends on some connection factories and queues has already tried to deploy. We need deployment to recover from this situation automatically. Restart/Reload will not help in this case as it would end up in the same situation. Only manual redeploy will help which is a workaround.