I have restarted the three node cluster with new directories for data and commitlog and made the test again.
This time the resultset size is 62 rows for the select. If I execute the select often it jumps between 62 and 65 rows. After inserting a second row I get 129 rows back I select using the id2 of the new row. I have no idea what the cause is, espacially if it works for you. One thing: cassandra startup log says I should use the newest oracle release instead of openjdk. Is this possibly causing trouble? cheers, Michael