Details
-
Bug
-
Resolution: Done
-
Major
-
7.1.3.Final (EAP)
-
None
Description
HotRod connector startup often fails in JDG 6.1.0.ER2. This is caused by early initialization of the JGroups stack, where the node can join cluster sooner than Infinispan is properly initialized, specifically sooner than the JGroupsTransport class initializes the JGroups channel and related RPC dispatcher.
In some early phase of startup each node with HotRod connector inserts data with key nodeName/clusterName into __hotRodTopologyCache (this is a replicated one). The PutKeyValueCommand is broadcasted to all nodes in the cluster, including the node which joined the cluster but where the Infinispan has not started yet. As there is no RPC dispatcher on the destination node, the message gets discarded, but source node requesting GET_ALL reponses waits for the response for 4 minutes (default timeout) and then gets an exception failing the HotRod connector.