The Marionette Collective (MCollective) is a server orchestration/parallel job execution framework available from Puppet Labs (http://docs.puppetlabs.com/mcollective/). It can be used to programmatically execute administrative tasks on clusters of servers. Rather than directly connecting to each host (think SSH in a for loop), it uses publish/subscribe middleware to communicate with many hosts at once. Instead of relying on a static list of hosts to command, it uses metadata-based discovery and filtering and can do real-time discovery across the network.
Getting MCollective up and running is not a trivial task. In this article I’ll walk through the steps required to setup a simple MCollective deployment. The middleware of choice as recommended by the Puppet Labs documentation is ActiveMQ. We’ll use a single ActiveMQ node for the purposes of this article. For a Production deployment, you should definitely consider the use of a clustered ActiveMQ configuration. Again for the sake of simplicity we will only configure a single MCollective client (i.e. our “admin” workstation). For real-world applications you’ll need to manage clients as per the standard deployment guide.
There are four hosts in the lab – centos01 which is our Puppet Master and MCollective client, centos02 which will be the ActiveMQ server and an MCollective server, centos03 and centos04 which are both MCollective servers. All hosts run Puppet clients already, which I’ll use to distribute the appropriate configuration across the deployment. All hosts are running Centos 6.5 x86_64.