Cassandra, strange behavior after changing replication_factor

I ran into a strange situation when changing the replication_factor of an existing keyspace.

Using a cluster of three nodes, each launching the Datastax Cassandra 2.0.11 community community.

I created a "workout" in the key space:

create keyspace training with replication = {
'class': 'SimpleStrategy',
'replication_factor': 1
};

      

        
        
        
      

    

Then

use training;

      

        
        
        
      

    

Then

create table customers (...);

      

        
        
        
      

    

Insert three records. Query a table from another node and got three records back as expected.

Then the replication_factor file changed:

alter keyspace training with replication = {
'class': 'SimpleStrategy',
'replication_factor': 3
};

      

        
        
        
      

    

Table query. He showed zero entries! After a few seconds, he showed 2 entries. Then again zero records. Insert two records. Now it sometimes shows 2 (new) entries, sometimes 4 (2 new, 2 old). Same behavior on all three machines.

I am absolutely stumped. Does anyone know why it behaves this way?

Only one strange possibility occurs to me: out of three nodes in the cluster, two are running Ubuntu, 64-bit OS, and the third is running 32-bit Windows 7. Does this mean / explain this behavior?

Thank.