Scaling cluster nodes

A Cloud Containers cluster can be scaled by changing the number of its master nodes and worker nodes. This allows you to adapt the cluster to the changing needs of workloads (workloads).

Two types of scaling are supported:

• Vertical scaling: The virtual machine templates that are used by the master and worker nodes of the cluster are being changed. In this way, you can manage the computing resources of the cluster without affecting the number of nodes.

  Scaling of worker nodes is performed within the framework of the worker node group.

• Horizontal scaling: The number of worker nodes within a separate node group is changed. In this way, you can manage the computing resources of the cluster without affecting the virtual machine templates that are used by the nodes.

Both types of scaling are performed manually. Also can be customized automatic horizontal scaling for a group of nodes (autoscaling): the number of worker nodes in the group will be adjusted depending on the needs of the workload. This mechanism allows you to save up to 60% on computing power.

• Vertical scaling capabilities are limited by current quotas and available virtual machine templates.
• Horizontal scaling capabilities are limited by the current quotas and the limit on the number of worker nodes in a separate node group: from 1 to 100 nodes.
• It is not possible to perform horizontal scaling manually if automatic scaling is configured. To perform manual scaling, turn off automatic scaling.

When vertically scaling any cluster nodes, the virtual machines that are used by the nodes are sequentially restarted. This is necessary to apply the new virtual machine template. Therefore, the scaling process affects the availability of both the cluster API and the workloads hosted in the cluster:

• If the cluster is not fault-tolerant and contains a single master node, then the Kubernetes API will be unavailable until scaling is completed.

• If a node group contains only one worker node, then the workload hosted on it will be unavailable until scaling is completed.

• If a node group contains multiple worker nodes, the workload hosted on them will not be available until scaling is complete if replication for the load is not configured or configured incorrectly.

  For example, if you place all replicas on one worker node, then when it is restarted, the workload will become unavailable, even if there are other worker nodes in the node group.

  Configure replication so that some of the workload replicas are available when worker nodes are restarted.

• If there are not enough computing resources for the workload after scaling the node group, then this load may not work correctly or become unavailable.

  Make sure that when scaling down computing resources, the total amount of resources in the node group is sufficient.