Friday, 19 October 2012

ejabberd clustering: a clever idea for Mnesia replication setup

The process of configuring a cluster of ejabberd entities serving the same domain is apparently simple (considering the ejabberd Installation and Operation Guide).

Assuming you have a node running, you create a second node with the same Erlang cookie, then setup Mnesia replication between the two. Assuming both ejabberd instances have a similar configuration setup and can connect to each other over the network, you're pretty much done.
This can be generalized and made N times to build your cluster.

ejabberd uses Mnesia as its internal DB. Although you can easily move to MySQL as storage backend, it's important to note that  Mnesia is still necessary to successfully build the cluster.

ejabberdctl is the control script that can be used to start, stop, restart ejabberd. It can be also used to attach a debug console to a running ejabberd instance, or execute any command exposed by an ejabberd module.

The idea presented in Easy ejabberd clustering procedure is to extend ejabberdctl with an additional command ("attach") which hides the complexity of connecting to an Erlang node and setup Mnesia replication. I've found it an interesting idea, firstly because it becomes easier to automate the clustering configuration, secondly because goes in the right direction of hiding the setup complexities behind a single control tool.

Check the table size of a MySQL DB

I wanted to create a MySQL dump of a small DB, so run mysqldump and realized that what I was expecting to take a few MB was instead half a GB.
Where did things go wrong? Which table is storing such an unexpected amount of data?

Googling around I've found a few interesting posts on how to check the size of a table, and then selected this one.

My suggested approach is:
SELECT TABLE_NAME, table_rows, data_length, index_length, 
round(((data_length + index_length) / 1024 / 1024),2) "Size in MB"
FROM information_schema.TABLES WHERE table_schema = "schema_name";
where schema_name is the DB you're interested in.

I've then immediately spotted the "offending" table :-)