[hendzen]

I understand the issues you've brought up in your post as well as the traditional methods of pruning the search space of query plans through using left-deep join trees and histograms for calculating selectivity.

My top-level point was that there is no way humans will come up with the optimal query plan by hand, and like you said even if they do - the data will change.

As an aside, the reason Mongo can use such a simple but effective strategy is because it doesn't support joins, which makes calculating the optimal plan much, much easier. RethinkDB does, however, and as such, a CBO is that much more important.