Ok here is what i dont understand.
In a typical MASTER_AMASTER_B >> MANY_REPLICA how do you handle, if you set up all slave to replicate from MASTER_A and its failes (mean the server go offline) that all the replicaes change their master to MASTER_B without loosing data in an MMM cluster?

Thanks in advance.
drk

> With this “Master-Master” in the project name we mean Active+Passive
> master scheme which is the same as Master+Slave, but one of the slaves
> being set up to be spare master and be able to handle writes when
> active master dies.

Does that mean that there will be no latency difference in terms of a
Master-Master and Master-Slave (no additional priority for the passive
master to get replication updates when compared to a pure slave) ,
expect for the fact that it is Administratively easier to bring back
the failed master in Master-Master.

> Something like this:
> # mmm_control set_offline
> …doing schema change on the passive master…
> # mmm_control set_online
> # mmm_control move_role writer
> # mmm_control set_offline
> …doing schema change on former active master…
> # mmm_control set_online

> So, major idea of this process with mmm is that you could say “some
> server is offline” and it would be offloaded and be available for any
> changes/backups/etc.

In the above scenario how to restore the changes that occurred to the
active master when the passive master is offline. How the changes
during that period will be applied to the passive master when both are
online.

Moreover on the destructive/non destructive schema change, I assume
data type change, new column additions,modify column to be non
destructive and drop a column, table name changes to be destructive.
Would you recommend the above approach to both these cases.

I’ve answered your question here: http://groups.google.com/group/mmm-devel/browse_thread/thread/2738b71da466a3eb.

Alexey

Apart from the above I have other questions which will help me in deciding on a solution for our project,

1. Whether a Backup server is required in a model of Master-Master replication and when I have slaves of each master in a different machine, I totally will have 4 machines in each cluster.
2. How is the Alter table should be done in a Master-Master replication model? How do we distinguish between a destructive and non destructive schema change?

Srini

It depends on what are you looking for. You can take any software and use it with MMM Cluster, using only Master for both reads and writes and passive node simply for fallbacks.

If you want to send some queries to second node you will need to change software appropriately.

This is very good point. There are bunch of documented things which are not handled by MySQL 5.0 replication for example UUID and I guess there are more less known things like UPDATE/DELETE statements with LIMIT which can have different result depending on data layout and of course bugs.

This is why I recommend taking people taking backups from the Master whenever possible.

Regarding your script, I’m curious if it works with active replication going as this is when you run into the problems as you need to checksum tables at the state, which would require table lock for checksum duration for MyISAM tables.