Mathias <[EMAIL PROTECTED]> wrote on 06/28/2005 01:11:59 PM: > Selon [EMAIL PROTECTED]: > <snip> > > Thanks Shawn, but i'm not speaking about data consistency during > transaction and > isolation levels. > I spoke about what is seen in the data dictionary as num_rows an why > it can not > be used even it's quite faster. > > > > Hope that helps > :o) > Mathias
And I was trying to explain why there is not a number IN the data dictionary that represents "row count". Unless a separate dictionary is maintained FOR EACH TRANSACTION, the record counts will be wrong. The record counts determined by SELECT COUNT(*) are *per transaction* so the only way to do a record count is by checking each row (pending or not) against cross-transaction isolation. This has everything to do with the row-level locking built into InnoDB and unless they enhance the engine to maintain a list of table statistics (I think this is part of what you are calling the dictionary) for each transaction, there can't be a rapid lookup of the row count. Right now I don't see that as high on their priorities. Shawn Green Database Administrator Unimin Corporation - Spruce Pine
