Autoscaler
¶ About Autoscaler
Autoscaler monitors your system’s demand and adapts to optimize the use of your resources. With reservations, it ensures you have sufficient capacity to scale up within your limits. It is purposed to automatically scale a worker pool, increasing the number of workers to handle the workload spikes, and decreasing the number of workers when the workload subsides, which allows you to save on cluster maintenance costs.
We are utilizing our custom modification of the Kubernetes native cluster autoscaler component. You can find the native Kubernetes Cluster autoscaler detailed description here.
¶ Autoscaler work algorithm
.png)
¶ Upscaling
is a process of worker number incrementation within a worker pool.
¶ Downscaling
is a process of worker pool decremention via updated K8s API request.
¶ Notes
- Maximum and minimum autoscaling limits are set per worker pool.
- Workers are considered unused if not utilized for 10+ minutes.
- Pods in pending state are pods waiting to be settled.