A Note on Amazon Autoscale Groups

AWS provides auto scaling to allow applications to scale up and down to meet demand. Autoscaling is an excellent tactic for ensuring that an application can add and remove capacity based on demand. Autoscale policies can be infinitely tuned to use anything that can be recorded by CloudWatch. However, autoscale groups are also incredibly useful for ensuring that a set number of instances is always available. This can be accomplished by omitting scaling policies.

But for some reason, AWS users seem to shy away from using autoscale groups for anything but satisfying elastic demand. However, in my experience, autoscale groups are a critical component of any highly available deployment. Even deployments of a single instance.

The utility of a service that does nothing but ensure that the desired number of instances available is the basis for creating self healing infrastructure. However, the name Auto Scaling has created a widely held opinion that the number of instances must regularly change to make use of the service. This is simply not true.

For example, you may find a situation where you need to ensure high availability for an application only built to run on a single server. Or you may have a snowflake server, or an application where it’s not worth having more than one instance. Instead of creating a standalone instance, you can create an autoscale group with min, max and desired set to 1. As long as proper automation is in place, and the application is a proper 12 factor app, the autoscale group will ensure there’s always a single instance available.

As a design principle, any infrastructure created should have the means to heal itself. Don’t be put off by the name Auto Scaling, as the utility and value of keeping a single instance running (that doesn’t need to scale) using a well understood and easily accessible service is worth it’s weight in gold.

Aug 26th, 2015

Comments