Sunday, April 17, 2005

MySQL replication instructions

As of version 3.23.15 (try to use 3.23.29 or later), MySQL supports one-way replication. Since most web applications usually have more reads than writes, an architecture which distributes reads across multiple servers can be very beneficial.

In typical MySQL fashion, setting up replication is trivial. On your master server add this to your my.cnf file:

[mysqld]
log-bin
server-id=1

And add a replication user id for slaves to log in as:

GRANT FILE ON *.* TO repl@"%" IDENTIFIED BY 'foobar';

Then on your slave servers:

[mysqld]
set-variable = max_connections=200
log-bin
master-host=192.168.0.1
master-user=repl
master-password=foobar
master-port=3306
server-id=2


Make sure each slave has its own unique server-id. And since these will be read-only slaves, you can start them with these options to speed them up a bit:

--skip-bdb --low-priority-updates --delay-key-write-for-all-tables

Stop your master server. Copy the table files to each of your slave servers. Restart the master, then start all the slaves. And you are done. Combining MySQL replication with a Squid reverse cache and redirector and you might have an architecture like this:

You would then write your application to send all database writes to the master server and all reads to the local slave. It is also possible to set up two-way replication, but you would need to supply your own application-level logic to maintain atomicity of distributed writes. And you lose a lot of the advantages of this architecture if you do this as the writes would have to go to all the slaves anyway.

No comments: