Ensuring Paging State Consistency in Cassandra Token Range Queries

Goal

Iterate over a very large table using a token range query with paging to avoid loading the entire result set into memory.

Uncertainty

The application must guarantee that each page contains a unique slice of the token range and that the paging token remains valid even if the underlying data or cluster topology changes during the iteration.

Questions

Does Cassandra capture the exact position within a token range when generating a paging token so that no rows are missed or duplicated if new rows are inserted or deleted during paging?

Which DataStax Java driver version reliably serializes the paging state across node topology changes without causing incomplete result sets?