When our instances are stateful, we need to consider whether the tasks being processed can be distributed across the cluster and any data being received and stored is shareable across instances.

In the worst case, the stateful instances will also have stickiness in their application processing, meaning that one client request will need to be fully processed by one particular instance. When clients and instances need to have a strict relationship, the failure of an instance will cause the client or clients using this instance to lose their connection and possibly lose their data that has not been saved to a persistent disk capable of withstanding an instance failure. In this case, a traditional backup scenario needs to be implemented. Such an instance within AWS should be configured with an ELB volume for stateful data and both the instance and the ELB volume will need to have frequent snapshots created against them. The instance should also have an ENI attached to it if a strict MAC or IP relationship is required. You can also use the EC2 Auto Recovery service that can monitor your instance and automatically recover it with the same metadata of the instance that failed (bring it up with the same IP, MAC, ID, and so on) in the case of a hardware failure within AWS.