In some business scenarios you may want to direct the Run:AI scheduler to schedule a Workload to a specific node or a node group. For example, in some academic institutions, a hardware is bought using a specific grant and thus "belongs" to a specific research group.
Run:AI allows this "taint" by labeling a node, or a set of nodes and then during scheduling, using the flag --node-type <label> to force this allocation
Configuring Node Types
To configure a node type:
- Get the names of the nodes where you want to limit Run:AI. To get a list of nodes, run:
kubectl get nodes
- For each node run the following:
kubectl label node <node-name> run.ai/type=my-nodes
The same value (e.g. "my-nodes") can be set to a single node, or for multiple nodes.
Using Node Types
Use the node type label with the --node-type flag, such as:
runai submit job1 ... --node-type "my-nodes"
See the runai submit documentation for further information