maria-developers team mailing list archive

[Kristian Nielsen <knielsen@xxxxxxxxxxxxxxx>]

Peter Laursen <peter_laursen@xxxxxxxxxx> writes:

> I wonder if it will be possible to use replication with GTID's on a 'mixed
> pool' or MariaDB and MySQL5.6 machines? Is this a concern or not? I see a
> problem with the CLOUD where users may not be able to choose the server as
> they want.  But if it is *plain impossible* to have both server versions in
> a setup, that's it of course.  However if so this would then be MariaDB's
> first 'major break-off' from Oracle/MySQL.

There will be no problem replicating from MySQL -> MariaDB, or from MariaDB ->
MySQL. Just as it will be possible to replicate between eg. MySQL 5.6 and
MySQL 5.5. But such replication will use the old-style file-name-and-position
mechanism, not global transaction IDs.

Rest asured that maintaining replication compatibility between different
versions remain a top priority!

About using MySQL-style global transaction IDs on a MariaDB slave, or
supplying a MySQL slave with MySQL-style global transaction IDs from a MariaDB
master, this is not something I plan on doing. Their design is just too awkward
and the implementation too lousy for me to want to support it.

But think about it: Global transaction ID is really about being able to do
automated scripting of failover to a new master, on production-critical
systems. Do we really want to have a mixed setup of MariaDB and MySQL slaves,
and automatically fail-over between them at random? I think this is an
unlikely scenario, and not worth the effort to try to make work (if even
possible).

It might be more useful to have eg. a MariaDB slave, and be able to use
MySQL-style global transaction IDs to automatically switch the MariaDB slave
among a pool of MySQL masters. This may be possible, but I have not thought
much about it yet.

 - Kristian.