[ https://issues.jboss.org/browse/WFLY-11645?page=com.atlassian.jira.plugin... ] Richard Achmatowicz edited comment on WFLY-11645 at 1/25/19 4:01 AM: --------------------------------------------------------------------- What is a clustered bean? On the one hand, this is a bean deployed on a node in a cluster. In the case of SFSBs, if this bean has passivationEnabled=true, arrangements are made for making that beans state available across the cluster to support fail-over (selecting a different target for the invocation when a previous one failed). So, SFSBs are special in this respect. On the other hand, from the point of view of the EJB client, a clustered bean is a proxy which has affinity to a cluster. This can apply to proxies for SFSB as well as SLSBs. To tie the proxy to the cluster, you need only (in theory) assign the strong affinity or the weak affinity of the proxy as type ClusterAffinity, but in practice, we usually tie the proxy to a cluster using the strong affinity. What this does is place a restriction on the targets chosen for the invocation; in this case, the target must point to some node within the cluster named in the affinity. For a clustered SFSB proxy, where the strong affinity is used to specify affinity to the cluster, if you wanted to send a invocations to a particular node in the cluster, you can do it by setting the weak affinity of the proxy to a NodeAffinity for the node within the cluster. For a clustered SLSB proxy, you should be able to do the same. Whether or not you still can is another matter; i'm looking at writing a test to verify that these "edge case" affinity combinations work properly. It is true that on invocations for SFSB and SLSBs, strong affinity was being forced to the "cluster affinity", which would result in the behavior that you observed, but this was incorrect and a fix is in place to change it (https://issues.jboss.org/browse/WFLY-11489). However, the bean will still be tied to the cluster as before, even after the fix. One of the reasons why the affinity to the cluster is retained after invocation is that if you remove affinity to the cluster (as in your example), from the EJB client point of view, this bean is no longer a clustered bean and fail-over will not happen. was (Author: rachmato): What is a clustered bean? On the one hand, this is a bean deployed on a node in a cluster. In the case of SFSBs, if this bean has passivationEnabled=true, arrangements are made for making that beans state available across the cluster to support fail-over (selecting a different target for the invocation when a previous one failed). So, SFSBs are special in this respect. On the other hand, from the point of view of the EJB client, a clustered bean is a proxy which has affinity to a cluster. This can apply to proxies for SFSB as well as SLSBs. To tie the proxy to the cluster, you need only (in theory) assign the strong affinity or the weak affinity of the proxy as type ClusterAffinity, but in practice, we usually tie the proxy to a cluster using the strong affinity. What this does is place a restriction on the targets chosen for the invocation; in this case, the target must point to some node within the cluster named in the affinity. For a clustered SFSB proxy, where the strong affinity is used to specify affinity to the cluster, if you wanted to send a invocations to a particular node in the cluster, you can do it by setting the weak affinity of the proxy to a NodeAffinity for the node within the cluster. For a clustered SLSB proxy, you should be able to do the same. Whether or not you still can is another matter; i'm looking at writing a test to verify that these "egde case" affinity combinations work properly. It is true that on invocations for SFSB and SLSBs, strong affinity was being forced to the "cluster affinity", which would result in the behavior that you observed, but this was incorrect and a fix is in place to change it (https://issues.jboss.org/browse/WFLY-11489) -- This message was sent by Atlassian Jira (v7.12.1#712002)