It is important to distinguish between hybrid node groups and hybrid clusters:
-
Hybrid node groups may include both Static nodes deployed in the cloud and servers deployed in the customer’s on-premises data center (bare metal or virtual machines).
For example, the primary workload may run on bare-metal servers, while cloud instances are used as a scalable addition during peak loads. Since nodes of the same type (Static) are used both in the customer’s data center and in the cloud, the architecture of the
node-managermodule in this case corresponds to the Static node variant. The only mandatory requirement is the presence of L3 connectivity between the on-premises data center and the cloud. For example, Yandex Cloud provides a Cloud Interconnect mechanism. -
Hybrid clusters are DKP clusters that simultaneously include:
- NodeGroup with Static nodes deployed in the customer’s on-premises data center (bare metal or virtual machines).
- NodeGroup with CloudEphemeral nodes deployed in the cloud.
In this case, the components required to manage both node types are deployed. The architecture of the
node-managermodule corresponds to the Static node variant. In addition, components required for CloudEphemeral nodes are deployed:- cloud-provider: Provides interaction with the cloud infrastructure. A configured provider for the corresponding cloud is required. It also includes csi-driver and cloud-controller-manager.
- cluster-autoscaler: Provides automatic scaling of cluster nodes.
As with hybrid node groups, hybrid clusters require L3 connectivity between the on-premises data center and the cloud.