Cassandra DROP COLUMN rollback and DataStax Java driver schema metadata consistency

Goal

Determine whether a Cassandra schema rollback using DROP COLUMN after ALTER TABLE ... ADD COLUMN leaves the DataStax Java driver's cached schema metadata consistent with server-side schema agreement, particularly when old and new application versions run concurrently.

Constraints and uncertainty

Schema agreement confirms that Cassandra nodes have converged on metadata, but it does not confirm that every connected driver instance has refreshed its cache or that in-flight prepared statements remain valid. Dropping a column removes the schema contract without rewriting existing SSTables, so re-adding a column with the same name may expose stale values until compaction occurs. Whether the rollback should preserve the column as a compatibility bridge or treat the drop as a destructive migration is an unresolved interoperability boundary between the database and the driver.

Does server-side schema agreement after DROP COLUMN guarantee driver-side metadata refresh for all active sessions? If not, what conditions indicate that cached metadata has converged before re-adding or reusing the column?