[MySQL-replication] master-master replication broken (conflicts) - what now?

Here is the documentation I wrote regarding the process for my company. In your case, I would STOP SLAVE on both servers to insure that you don't affect the primary Master while you are restoring the secondary.

Do you have the filesystems which host the datadir running on Logical Volumes? If not, you'll have to run mysqldump, and the outage duration will be increased. Also remember that when you install all the databases from the primary master to the secondary master, you won't have the replication user the primary master uses as a user on the secondary db.... if that makes any sense? I would also insure that you do not have the log slave updates flag enabled, that you have different server id's configured, and the that auto-increment-offset values differ between the masters.

Here is a brief overview of the process:

-Lock tables on Master 1
-Record log & log pos from master
-Create an LVM snapshot of the datadir partition
-Transfer the database files to Master 2
-Bring up Master 2 as a slave
-Lock tables on Master 2
-Record log & log pos on Master 2
-Unlock tables on Master 2
-Start Slave replication on Master 1

Connect to the Master 1 MySQL database with an admin account. Run
Record this data!

Exit the MySQL shell. Run
#/usr/sbin/lvmcreate -L16G -s -n dbbackup <path/to/LVM/datadir/device>

Reconnect to the database and run:
Return to the shell and run

#mkdir /mnt/dbbak
#mount <path/to/snapshot/device> /mnt/dbbak

If you are using an XFS filesystem, you will need to run mount with the -onouuid,ro options!

Now cd to the dbbak directory.

run
#tar cvzf - ./* | ssh <username>@<Master 2 URL> "cat > /<target datadir>/dbbak.tgz"

After this completes, turn off the mysql process on master 2. umount the the /mnt/dbbak directory and run lvremove against the dbbak device. Unpack the tarball, and bring the db on Master 2 back up. Connect to the db and run
Look for the output noted above.

On Master 2
You should see both slaves 0 seconds behind master, with no errors. This can be tested by creating and dropping a test database on each master, verifying that the changes are reflected on the paired machine.

Finally, we need to add the machines to V.I.P., and bring up the pulse daemon to enable failover.
Run

#/sbin/modprobe bonding

Edit the /etc/sysconfig/network-scripts/ifcfg-eth0 and /etc/sysconfig/network-scripts/ifcfg-bond0 so that they are similar to the following

ETH0:
DEVICE=eth0
USERCTL=no
ONBOOT=yes
MASTER=bond0
SLAVE=yes
BOOTPROTO=static

BOND0:
DEVICE=bond0
ONBOOT=yes
BOOTPROTO=none
IPADDR=10.11.8.22
NETMASK=255.255.255.0
NETWORK=10.11.8.0

Restart the network daemon, restart the pulse daemon, and verify that the V.I.P. is functional. A sample lvs.cf file is attached.

You are done!