For a single service per host deployment, scaling a microservice requires adding or removing additional machines that host the microservice. If your application is running on a cloud architecture (public or private), many providers offer a concept known as autoscaling groups.
Autoscaling groups define a base virtual machine image that will run on all grouped instances. Whenever a critical threshold is reached (for example, 80% CPU use), a new instance is created and added to the group. Since autoscaling groups run behind a load balancer, the increasing traffic then gets split between both the existing and the new instances, thus reducing the mean load on each one. When the spike in traffic subsides, the scaling controller shuts down the excess machines to keep the costs low.
Different metrics can act as triggers for the scaling event. The CPU load is one of the easiest to use, but it may not be the most accurate one. Other metrics, such as...
