This option tells the server, on startup, to iterate through all MyISAM tables in all databases and open them (thus doing whatever work is specified by the other options supplied for myisam-recover) -- and *not* to accept connections from clients (other than root@localhost) until after the check is complete. The problem with the current behavior is that if a large table is corrupted, individual queries can hang for hours with no response. Sometimes that behavior is not acceptable. It can be much better for a server to seem down (not accepting connections) than to accept queries but not be able to respond to them. Ideally the initial table checking would be multi-threaded.