Autoscaling
Adding and removing machines automatically so that capacity follows demand. A controller watches a signal such as CPU use, compares it with a target, and changes the number of copies. New capacity takes time to boot, so autoscaling follows slow changes in load well and sudden spikes badly.
Performance
In Quorum#
The Autoscaler (Autoscaling Research) controls the App Servers and Workers that you link into it. When a Node runs above its target (40%, 60%, or 80% of CPU or of its whole load), it asks for more copies, up to a max of 4, 8, 16, or no limit. New copies serve after 180 s on VMs or 30 s in a Kubernetes Cluster, and they start cold. It removes one copy when load falls below 60% of the target, and it acts at most once every 30 s, so keep spare capacity for a sudden Traffic spike.
Read next#
Vertical and horizontal scaling, Diurnal traffic, Cold start, Container orchestration