Very good, gmail does not handle mailing lists properly...... Sorry for sending this off-list to you, Alec.
On Fri, 9 Jul 2004 17:09:45 -0300, Joćo Paulo Vasconcellos <[EMAIL PROTECTED]> wrote: > > > On Fri, 9 Jul 2004 10:44:42 +0100, [EMAIL PROTECTED] > <[EMAIL PROTECTED]> wrote: > > "L. Yeung" <[EMAIL PROTECTED]> wrote on 09/07/2004 08:38:38: > > > > > Hi! I wanted to set-up a master-slave replication on > > > my win2k box. When my master server fails, my slave > > > server will automatically becomes my new "master > > > server". And when my master server is back online, any > > > changes on my slave server is replicated back to my > > > master server. > > > > > > Normal: A -> B > > > ^ > > > L-------- inserts/updates from clients > > > Failure: X B > > > New Master: X B <- inserts/updates from clients > > > Master is back: A <- B x- inserts/updates are blocked > > > manually. > > > Normal: A -> B > > > ^ > > > L-------- inserts/updates from clients > > > > > > Any idea? Thanks. > > > > You cannot do this automatically: you need a supervisor progream. > > > > The way you need to do it is to have both machines come up with their > > slave thread *not* running. The supervisor then determines which is master > > and which slave, and starts the appropriate slave thread running. If it > > determines that the original master has failed, it stops replicatio on the > > slave and directs updates to it: the slave has now become master. When the > > original master reappears, it determines that updates have been made to > > the original slave later than those to the original master, it instructs > > the originl master to reload its databse from the slave. Master and slave > > have now exchanged roles. > > > > The determination of which is the most-recently updated is done by a > > single row, single column table which is incremented whenever the slave > > takes over from the master. If, at startup, two active machines are found > > with differeent values in this entry, the higher value becomes master and > > the lower must be re-synchronized. If the values are the same, the slave > > status can be inspected to see which is slaving to which. > > > > We have implemented such a system in our own middleware. We have a target > > changeover time of 10 seconds, which we are meeting easily. It only works > > for MyISAM tables, since LOAD DATA FROM MASTER is only available for > > these. > > > > Note to MySQL development team: this request comes up often enough that I > > hope the idee of embedding this supervisor in the MySQL daemon is at least > > on the wish list. > > > > Alec > > > > I personally think that, if my master has gone away, there must be a > reason and I do not want the former master to take over when (and if) > it came back. So, in my production environment, I set this with > heartbeat (linux-ha.org) and a perl script that makes my slave works > as master when needed. Of course, my slave is nothing but a hot backup > spare server. > > > -- > Joćo Paulo Vasconcellos > ICQ: 123-953-864 > -- Joćo Paulo Vasconcellos ICQ: 123-953-864 -- MySQL General Mailing List For list archives: http://lists.mysql.com/mysql To unsubscribe: http://lists.mysql.com/[EMAIL PROTECTED]
