Node configurations in a Managed Kubernetes cluster
In a Managed Kubernetes cluster, you can use cloud server configurations and dedicated server configurations to create node groups.
You can check configuration availability in the regions in the Managed Kubernetes availability matrix.
When selecting a configuration, keep in mind that some resources are reserved for system components, so the number of resources available for placing pods differs from the amount specified in the configuration.
Node configurations on the cloud server
There are three types of configurations available for cloud server nodes:
- custom configurations — you can specify any resource ratio;
- fixed configurations with GPU — prebuilt node configurations with graphics processors where the resource ratio is fixed;
- fixed cloud server configurations — cloud server configurations (flavors) with a fixed resource ratio. They can only be added via the Managed Kubernetes API or Terraform.
You can compare available configurations in the Comparing Node Configurations on Cloud Server table.
If the available configurations do not meet your needs, for example, if you need more vCPU or RAM, create a ticket — we will select configurations with a different amount of resources.
Custom configurations
In custom node configurations, you can specify the required amount of resources. Limits depend on the pool, you can see them in the Comparing Node Configurations on Cloud Server table.
You can select a local disk or one of the network volume types as the boot disk for nodes.
When selecting a configuration with a network volume, keep in mind that the ability to change the IOPS limit is not supported for Universal SSD v2 and Fast SSD v2 network volumes.
A fault-tolerant multizonal cluster uses cloud servers with dedicated cores and Hyper-Threading technology. In this cluster type, only an even number of vCPUs can be selected.
Fixed configurations with GPU
You can select a fixed configuration with dedicated GPUs and a set resource ratio. The configurations correspond to the GPU line of cloud servers.
Detailed characteristics of graphics processors can be viewed in the Available GPUs subsection of the Creating a Managed Kubernetes cluster with GPU guide.
You can select a local disk or one of the network volume types as the boot disk for nodes. For a network volume, you can select the disk size; for a local disk, the size is fixed.
When selecting a configuration, keep in mind that:
- the ability to change the IOPS limit is not supported for Universal SSD v2 and Fast SSD v2 network volumes;
- some fixed configurations with GPU only have a local disk available.
Fixed cloud server configurations
Fixed cloud server configurations are only available via the Managed Kubernetes API or Terraform.
If the Managed Kubernetes configurations in the control panel do not meet your needs, you can use fixed cloud server configurations (flavors). You can view available flavors in the Cloud server configuration line flavors list guide.
In a fault-tolerant multizonal cluster, only the Standard and HighFreq lines with dedicated cores are available.
Comparison of node configurations on the cloud server
Managed Kubernetes configurations
Fixed cloud server configurations
* If a configuration has more than 8 vCPUs, you can select RAM with a ratio of at least 1:2. For example, if you chose 10 vCPUs, the RAM must be at least 20 GB.
** If a configuration has more than 8 vCPUs, you can select the local disk size with a ratio of at least 1:32. For example, for 10 vCPUs, the minimum disk size is 320 GB.
You can check configuration availability in the Managed Kubernetes and GPU for cloud servers and Managed Kubernetes availability matrix.
Dedicated server node configurations
In Managed Kubernetes clusters, you can create nodes on dedicated servers of prebuilt configurations that meet the following requirements:
- has a local port;
- x86 processor.
You can view detailed information about the configurations in the configurator.
Dedicated servers are pre-assembled and mounted in racks, and are ready for operation within one hour after creating a cluster or node group (excluding OS installation time).