The cluster has 3 nodes with Vnodes. We were adding a 4th one (in auto bootstrap mode). The 4 node has stream sessions, receiving data from the others.
While the streaming was going on (it takes quite a while because there are 200Gb of data of worth), an existing table is dropped in the cluster and a new one is created.
A few time after, we caught an NPE in the log file of the joining node (still not finished joining):
It can be a race condition where schema agreement has not reached the joining node.