After patch 16278, DEVICE_ADDED events are no longer replicated to all instances in the cluster once they are seen by one of the instances which causes a large increase in switch-up latency. Specifically, without the replication method, instances need to wait for gossip to synchronize on DEVICE events which takes seconds to finish.
And it seems the reason that the replication method is not working is because mastership assignment is not finished yet when the device store created a DEVICE event. So the following code is not executed
|17998,2||Experimental cheap fix for device events issue ONOS-7623||master||onos||Status: MERGED||+2||+1|
|18007,2||Experimental cheap fix for device events issue ONOS-7623||onos-1.13||onos||Status: MERGED||+2||+1|
|18899,1||Experimental cheap fix for device events issue ONOS-7623||onos-1.12||onos||Status: MERGED||+2||+1|