3.0.0 introduced a regression, "spark.cassandra.table.size.in.bytes" parameter is ignored. There is no way to override the estimated size of the table. This results in limited control over the number of partitions generated by PartitionGenerator.

A workaround is to override spark.default.parallelism