With distribution=latest, ycsb is attempting to read data that Cassandra does not know about. This produces numerous "Error reading key" messages in the logs. I suspect that it is querying the data before the data is even written as this is reproduced on a single node Cassandra cluster which should never exhibit this behaviour regardless of consistency level.

This may be more deeply rooted in the internals of ycsb than just the cql driver, but I have not been able to reproduce the same error with thrift, but it may be just be presenting itself differently.