Overview

Auto scaling is a service that automatically and economically adjusts service resources based on your service requirements and configured policies.

Context

More and more applications are developed based on Kubernetes. It becomes increasingly important to quickly scale out applications on Kubernetes to cope with service peaks and to scale in applications during off-peak hours to save resources and reduce costs.

In a Kubernetes cluster, auto scaling involves pods and nodes. A pod is an application instance. Each pod contains one or more containers and runs on a node (VM or bare-metal server). If a cluster does not have sufficient nodes to run new pods, add nodes to the cluster to ensure service running.

In CCE, auto scaling is used for online services, large-scale computing and training, deep learning GPU or shared GPU training and inference, periodic load changes, and many other scenarios.

Auto Scaling in CCE

CCE supports auto scaling for workloads and nodes.

Workload scaling: Auto scaling at the scheduling layer to change the scheduling capacity of workloads. For example, you can use the HPA, a scaling component at the scheduling layer, to adjust the number of replicas of an application. Adjusting the number of replicas changes the scheduling capacity occupied by the current workload, thereby enabling scaling at the scheduling layer.
Node scaling: Auto scaling at the resource layer. When the planned cluster nodes cannot allow workload scheduling, ECS resources are provided to support scheduling.

Components

Workload scaling components are described as follows:

**Table 1** Workload scaling components
Type	Component Name	Component Description	Reference
HPA	metrics-server	A built-in component of Kubernetes, which enables horizontal scaling of pods. It adds the application-level cooldown time window and scaling threshold functions based on the HPA.	HPA
CustomedHPA	cce-hpa-controller	An enhanced auto scaling feature, used for auto scaling of Deployments based on metrics (CPU usage and memory usage) or at a periodic interval (a specific time point every day, every week, every month, or every year).	CustomedHPA
CustomedHPA	prometheus	An open-source system monitoring and alarm framework, which collects public metrics (CPU usage and memory usage) of kubelet in the Kubernetes cluster.	CustomedHPA

Node scaling components are described as follows:

**Table 2** Node scaling components
Component Name	Component Description	Application Scenario	Reference
autoscaler	An open source Kubernetes component for horizontal scaling of nodes, which is optimized in terms of scheduling and auto scaling capabilities.	Online services, deep learning, and large-scale computing with limited resource budgets	Creating a Node Scaling Policy