When hibernate search is used (with capedwarf for example), adding a fourth server to the cluster crashes infinispan.

I've narrowed the problem down to org.hibernate.search.backend.impl.jgroups.AutoNodeSelector:78

Here the following formula is used to calculate an array index

int selectionRange = members.size() - 1;
int selected = ( indexName.hashCode() % selectionRange) + 1;

if indexname is long (like "default_taskworker-java__com.google.appengine.api.datastore.Entity")
the hashcode becomes negative due to integer overflow, resulting in a negative index