While testing a configuration with replicated HA plus mirroring, we noticed that the backup on the source cluster uses a large amount of memory, even though it is not live. Digging into the memory usage, the bulk of the memory is in a ConcurrentHashMap associated with the ReplicationEndpoint. Looking a little further, it appears that almost all of this memory is occupied by org.apache.activemq.artemis.core.paging.impl.Page objects.

Is there some way to speed evacuation of these objects from the live heap after they are copied over? In our case, we have observed several GB used by the ReplicationEndpoint in this scenario.