Skip to content

Move atelet's GKE ComputeClass toleration into the GKE install overlay #2760

Open

Description

Description

#2679 adds a toleration for GKE's ComputeClass taint to the atelet DaemonSet in manifests/ate-install/atelet.yaml:

- key: cloud.google.com/compute-class
  operator: Exists
  effect: NoSchedule

GKE taints the nodes it creates for a ComputeClass with cloud.google.com/compute-class=<class>:NoSchedule, so atelet needs this toleration to run on them. The toleration is GKE-specific, though, and as discussed in #2679 it belongs with the rest of the GKE install configuration rather than in the core manifests. It stays in atelet.yaml until there is a provider overlay to put it in.

Once the new installer lands (see the extensible install discussion in #2304) and substrate-gke ships its own kustomization with a GKE component, move the toleration there.

Proposed action items

  • Add the toleration to the GKE component in ai-on-gke/substrate-gke.
  • Remove it, and its TODO(roycaihw), from manifests/ate-install/atelet.yaml, and drop the ComputeClass sentence from docs/api-guide.md.

Related: #2679, #2304

Metadata

Metadata

Assignees

No one assigned

    Labels

    Type

    No type

    Projects

    No projects

      Milestone

      No milestone

      Relationships

      None yet

      Development

      No branches or pull requests

      Issue actions