Description

This cookbook exists to make a deployment of a Datastax Opscenter cluster on a Cassandra cluster in Amazon EC2.

Datastax Opscenter is a Monitoring and Management platform for Apache Cassandra - vital for visualising Cassandra's Status.

Datastax Opscenter is deployed in a traditional many-clients just-one-server model, which conflicts slightly with Cassandra's masterless deployment model.

To get around needing to know this/manage in an automated manner a leadership election takes place between nodes in the cluster, a leader is automatically assigned based on having the 'lowest' alphanumeric hostname, and that leader generates an agent package which is then distributed by http and verified by checksum.

The good news is that Opscenter stores its data within Cassandra itself in its own Keyspace (2 replicas) so if a leader is terminated and a new leader arises from the ashes, no data should be lost.

This cookbook supports two modes of deployment - multiregion OR non-multiregion, set through attributes. We default to non-multiregion aka singleregion. The way multiregion is expressed is in a difference in the agent's address configuration. Check the attributes for how to set this.

This cookbook deploys the tarball version of opscenter because the packages provided tend to install as root, whereas this cookbook installs/runs opscenter as a unique system user.

This cookbook holds certain assumptions to be true in order to easily manage its deployment:

A) You are deploying this on EC2

Reason: This cookbook was developed on/designed for EC2 deployment. If you want to support non-EC2 deployment please submit patches.

B) All members of the cluster share the same UNIQUE chef role and this is the first role in the list of roles. i.e. cassandra-cluster-one or product-production-casdb.

Reason: The unique role is used to search for other cluster members for shared information. If you want to extend/improve this please submit patches.

C) Connectivity between cluster members is suffiently open to allow for agent distribution and agent connectivity. Typically you should have a security group that allows relatively open access from that security group on port 80 for agent distrubtion.

Reason: Nothing will work without connectivity anyway. No node is an island.

Requirements

• Chef 10.16.4+
• Cassandra on each node.
• Python cookbook to run the server.
• Java cookbook to run the agent (on each node including the elected server).
• Nginx for agent distribution on the elected server.

Platform

• Ubuntu 12.04+ [tested heavily]
• Very Probably Debian
• Probably RPM-based distros. [we do attempt to differentiate where necessary]

Attributes

See the contents of attributes/default.rb where there are accurate comments and self-explanatory attribute names.

Recipes

• default.rb : A dummy recipe pointing to install.rb
• install.rb : Installs everything by calling the rest of the recipes in the right order. Includes a leadership election section for nominating the server.
• opscenter-server.rb : Installs the server
• opscenter-agent.rb : Installs the agent
• opscenter-server-shutdown.rb : Shuts down the server when a new leader takes over - we wouldn't want to waste ram, right ?

Usage

Include cassandra-opscenter in your runlist.

Ensure you have enough connectivity between cluster members so that the agent can be distributed. (over the http port)

Once this cookbook is deployed and the nodes converged, login to the cluster member with the lowest alphanumeric node.name on port 8888

Click 'Use Existing Cluster' in the dialogue box.

In the next dialogue box add the hostname you have connected to the cluster list and click 'Save Cluster'.

multiregion requires the following attribute be set - this is from the cassandra-priam cookbook, but would work for non-priam deployments of opscenter and cassandra:

{
  "cassandra": {
    "priam_multiregion_enable": "true"
  }
}

Development

See the Github page

Authors

• Author: Alex Trull atrull@mdsol.com
• Author: Benton Roberts broberts@mdsol.com

Copyright: 2013–2013 Medidata Solutions, Inc.