Tips learnt in Mysql HA classes

Replication

• the --read-only switch (or read_only in my.cnf) it’s the best way to block unwanted writes in a replication slave. So, ditch your specific user’s privileges :)
• log-slave-updates in my.cnf needs underscores _ because with normal dashes doesn’t work at all (at least with Mysql 5.1.22). So, it’s log_slave_updates
• Talking about log_slave_updates again, it’s fundamental if you’re mounting a circular replication system. Without it, replication will stop at first hop.
• SET GLOBAL SQL_SLAVE_SKIP_COUNTER=n could be usefull if you have to skip on a replication error. Be careful anyway because it can lead to logical inconsistences in your data if abused.

Cluster

• When starting up data nodes using ndbd, you can add the --nostart flag to have the process in memory without doing nothing. Then, you can tell the node to actually start working from the management console with ID_node start. Moreover, you can put the nostart flag in your config.ini, in the [ndbd default] section, if you want the change to apply to every node.
• in the management console, as said, you can start (or stop, or restart) the data nodes manually, but if you try with the SQL nodes, you’ll get a strange error. That’s because SQL nodes can only be restarted from the local machine, as it were a normal mysql installation (in fact, it is a normal mysql install :)

Mysql Proxy

• If you are trying to execute mysql-proxy with a LUA script, and you have the scripts tree in /usr/share/mysql-proxy, then you have to issue a command like this, otherways mysql-proxy will complain about LUA includes that cannot be found:
  LUA_PATH=/usr/share/mysql-proxy/?.lua mysql-proxy --proxy-backend-addresses=192.168.45.112:3306 --proxy-lua-script=/usr/share/mysql-proxy/rw-splitting.lua