Red Hat Cluster Part 1: Introduction

Introduction

Cluster is a group of two or more nodes put together. There are four types of clusters:

1. Storage cluster
2. High availability cluster
3. Load balancing cluster
4. High performance cluster

The most important point to remember is "eliminate single points of failure".

We have three main components to a HA cluster:

1. Cluster infrastructure: Management of configuration, membership, lock and also fencing.

2. High availability service management: Responsible for service fail over to another node.

3. Cluster administration tools: Tools for managing and configuring HA.

The concept of quorum

There are some fundamental concepts to be understood, the first of which is quorum. Quorum is based on the concept of voting. Each node has one vote by default. There must be a majority vote among the nodes for the cluster to function properly. In a five node cluster, if at least three nodes are active then the cluster has quorum. If three nodes fail then the cluster will not function. 

Quorum is also important so that there are no "split-brain" situations to provide data integrity. When nodes are writing to the same storage at the same time, there is high chance the file system and data might be corrupted. Quorum decides the dominant nodes and allows them to be active.  

We can configure quorum by means of Ethernet and quorum disk. Quorum by means of Ethernet has a majority voting  (50 percent voting + 1 extra). However, user specified conditions are used when quorum disk is used. For this blog post we will be configuring a quorum disk. 

In 2 node clusters, if a node has lost access to the quorum disk or the network, then it will lose a vote and will be made the passive node as it will be fenced.

Fencing

Fencing a node means to disconnect the node from the storage when the node fails, to prevent it from writing to the storage. There are different types of fencing:

1. Power fencing: Uses power controller to power off the failed node.

2. Storage fencing: Disconnects the failed node from the storage by disabling the FC ports to the        storage from the failed node.

3. Other types of fencing include the use of iLO, DRAC, IPMI, etc. In a virtualized environment such as a VMware vCenter, a power user is used to power off or reboot the nodes which are virtual machines.

Shared Storage

All the nodes will have a shared storage.  However, only the active node will be able to write to the shared storage. When a previously active node fails and is fenced, the shared storage will be accessible to the node with majority quorum votes in the cluster. Normally the shared storage is provided to the nodes through storage devices using fiber channel.

Virtual or Floating IP

The services running on the HA cluster nodes will have a common virtual IP. If a website with the IP address 202.65.31.125 running on Apache server is configured as a service in the cluster, then this IP will be the virtual IP for all nodes. When the active node changes, the website will still be available through this floating IP as the services are moved to another node, which is the objective of the highly available active/passive cluster.

In the next part we will configure a 2 node HA cluster on RHEL 6.5.