You lose aggregates and joins, arguably the most powerful parts of an ad-hoc query system (business intelligence kind of stuff). Without those features, you mus

You lose aggregates and joins, arguably the most powerful parts of an ad-hoc query system (business intelligence kind of stuff). Without those features, you must move a great deal of these concerns into the application layer (or whatever manage your shards), even if it's just rolling up aggregates across shards. This works, but again, at the loss of the real advantages of the database here as you've got to do it yourself. I agree with your second assertion. Secondary indexes should be avoided, but are a useful tool if absolutely needed. I've found that Cassandra's richer data model really allows them to be avoided in most scenarios.