I have been thinking about MySQL replication myself but it haven't
tried anything out yet. Your patches solves one of the problems
involved.
Another one is to assure that auto_increment id:s from the two
(or more) servers are unique. One method I came up with is to
let server-A assign odd numbers and server-B assign even ones.
This could trivially be expanded to more than two servers.
Will this concept work?
How do one solve all the problems arising from the missing atomicity
of operations and defunct locking. I reckon one will have to anaylyze
all inserts and updates in the code _very_ carefully to be sure nothing
can go wrong.
How does commercial dbservers handle replication and syncronisation?
It is possible to keep the servers in sync as long as they are
connected together (it may affect performance but it should be
possible),
but what happens if the connection goes down and then up.
How can they safely syncronize if they both accepted updates during
the time they were disconnected?
/Martin
> The following two patches create a new command line argument
> for mysqld called --ignore-log-user. Database updates
> that are caused by the user specified in this command line
> argument will not be logged to the update-log. This feature
> is useful for implementing a two-way-replication system
> between MySQL servers.
---
Martin Nilsson
Web/db developer
Lund, Sweden
