Andrew Donkin wrote:
So I tried to convert bayes_token to InnoDB to take advantage of its row-level locking (this is advised by the developers but not reflected in bayes_mysql.sql). After MySQL worked on that for a few days I stopped it, dropped the database (innodb was very confused), and recreated the database and all tables using innodb and two-byte IDs.
I wonder if you could have saved yourself the effort of having to retrain in this case by wrapping the DB recreation in a sa-learn --backup / sa-learn --restore pair?
Bill
