kubeflow/testing

[GCP] Migrate machine type to e2 family to save costs

Bobgy opened this issue · 8 comments

Bobgy commented

First PR: #850

Bobgy commented

Got errors:

*kf-ci-management
  --------------------
  <root>   https:/github.com/kubeflow/testing.git/acm-repos/kf-ci-management@master   
  ERROR    472c24c0                                                                   
  Error:   KNV2010: unable to update resource: could not patch: admission webhook "deny-immutable-field-updates.cnrm.cloud.google.com" denied the request: cannot make changes on immutable fields: [initialNodeCount nodeConfig]

source: namespaces/issue-label-bot-dev/container.cnrm.cloud.google.com_v1beta1_containercluster_code-intelligence.yaml
namespace: issue-label-bot-dev
metadata.name: code-intelligence
group: container.cnrm.cloud.google.com
version: v1beta1
kind: ContainerCluster

So I cannot change nodeInitial or nodeConfig fields, I'll add them back.

ContainerNodepool change fails with error

status:
    conditions:
    - lastTransitionTime: "2021-01-19T05:33:07Z"
      message: reference IAMServiceAccount /code-intelligence-vm@issue-label-bot-dev.iam.gserviceaccount.com
        is not found
      reason: DependencyNotFound
      status: "False"
      type: Ready

Looks like the service account format is incorrect

Bobgy commented

Done, actions taken for code-intelligence cluster:

  1. update container node pool to e2-standard-8 machine type, change nodepool name, because it's an immutable field
  2. delete default nodepool
  3. disable node autoprovisioning, delete auto-provisioned nodepool
Bobgy commented

I tried to disable node autoprovisioning via config connector, but it keeps failing with GoogleCloudPlatform/k8s-config-connector#277.
I had to disable auto provisioning manually for now.

Bobgy commented

TODO: delete auto provisioned nodepool from issue-label-bot cluster.

Bobgy commented

For record, 3h10min spent on e2 migration.

Bobgy commented

Fixed the last e2 usage in 91f10f6