Description
The docs at http://smalldatum.blogspot.com/2014/04/mongodb-tokumx-and-innodb-for-disk-io_15.html were not clear to me on three issues:
1) the per-db RW-lock is not yielded from page faults while searching an index.
2) this is only done for "predicted page faults", not real page faults. And I am not suggesting that it should be done on real page faults.
3) a pending write lock request prevents new read requests when the per-db RW-lock is currently locked in read-mode. The FAQ states that W has precedence over read but I think that you should also mention this behavior.
See http://smalldatum.blogspot.com/2014/04/mongodb-tokumx-and-innodb-for-disk-io_15.html for performance results and a brief description of the code.