Skip to content

node-selection.md

Latest commit

 

History

History
86 lines (64 loc) · 3.57 KB

node-selection.md

File metadata and controls

86 lines (64 loc) · 3.57 KB

Physical node selection

When starting a new VM/container instance, CrownLabs selects automatically the physical node where the instance has to be executed.

However, this behaviour can be changed, hence allowing users to select exactly the node (or the set of nodes) where the instance has to be executed. This can be achieved by defining (1) the proper label selectors on the nodes, and (2) a specific spec on the VM/container template, as follows.

Add the proper label crownlabs.polito.it/mylabel to the worker node

This is a possible example:

admin@k8s-master:~$ kubectl get node worker-2 -o yaml |more
apiVersion: v1
kind: Node
metadata:
  annotations:
    projectcalico.org/IPv4Address: 192.168.24.25/26
creationTimestamp: "2021-04-29T16:33:48Z"
labels:
  beta.kubernetes.io/arch: amd64
  cpumanager: "false"
  crownlabs.polito.it/gpu-available: "true"
  crownlabs.polito.it/node-size: big
  crownlabs.polito.it/node-name: worker-2
...

In the above example, the node has three labels, one boolean (if a GPU is available ot not), the other which tells the size of the node (their values, in this case 'big', are arbitrary), and the third one that keeps the name of the worker.

To add a new label to the node, you can simply use the following command:

kubectl label nodes worker-3 "crownlabs.polito.it/node-name=worker-3"

This adds label crownlabs.polito.it/node-name: worker-3 to the worker-3 node.

To remove the above label, you can use the following example:

kubectl label nodes worker-3 "crownlabs.polito.it/node-name"-

Add the proper specification in the VM/container template

This can be done as follows (this refers to a template called vscode-rust in namespace workspace-sid):

admin@k8s-master:~$ kubectl get templates -n workspace-sid vscode-rust -o yaml
apiVersion: crownlabs.polito.it/v1alpha2
kind: Template
metadata:
  annotations:
    kubectl.kubernetes.io/last-applied-configuration: ...
  creationTimestamp: "2025-04-07T14:51:49Z"
  generation: 13
  name: vscode-rust
  namespace: workspace-sid
  resourceVersion: "1846893079"
  uid: ba307d8c-fb40-4d3c-b6cd-7721dba8c388
spec:
  allowPublicExposure: false
  deleteAfter: 30d
  environmentList:
  - containerStartupOptions:
      contentPath: /vscode/workspace
      enforceWorkdir: false
    disableControls: false
    environmentType: Standalone
    guiEnabled: true
    image: crownlabs/vscode-rust:v0.2.0
    mode: Standard
    mountMyDriveVolume: false
    name: vscode-environment
    nodeSelector: {}
    persistent: true
    ...

The key information here is the line nodeSelector: {}, which says that this template would support the node selector feature. However, since the nodeSelector is empty, it means that the user can choose among any label available within the cluster that allows the selection of the node.

In this example, when the user will start this instance, it can select whether the VM/container can be started on nodes where the gpu-available is true, where node-size = big, where node-name = worker-2, or any other label existing in the cluster.

In case a template must be started on nodes that have a given characteristic (e.g., include a GPU), you can customize the nodeSelector in the template and list only the labels that are allowed for the associated instances.

Algorithm for node selection

The algorithm for node selection is depicted in the following image.

Node selection algorithm