When I last did this, we had a manual cut over to the slave, which usually only happened when something really bad happened. The master was taken out of service entirely until we could diagnose what was wrong with it. We'd then manually sync things up and restore the the master into service. Its probably possible to automate much of this now, but I haven't need to do use replication recently.

We did have a "read only" mode for the webapp, so that if the master went down the website would still work reading content from the slave but no updates to the content could be made until we did our manual cutover, or the temporary problem with the master was resolved.

Mubs