I'm investigating a cluster (running 2.4) where sorted scroll queries fails in the middle of retrieving the result set.
The result sets are very large (millions of docs). My impression is that large sorted scrolls are very cpu intensive. Could the cpu activity lead to the cluster node becoming unresponsive resulting in the Java client transport connection dropping the connection?
Some of the indexes also have very large mappings (10000+ fields). Can this have an impact on scrolls with sort?
Have anyone else done large sorted scrolls with or without success?