v0.18.0

(Optional) Add livenessProbe check for Ansible-based operators

Existing operators will have a healthz endpoint without intervention, but to take advantage of it, a liveness probe should be manually added to the operator manifest. For example, the deploy/operator.yaml file would need to include:

  1. livenessProbe:
  2. httpGet:
  3. path: /healthz
  4. port: 6789
  5. initialDelaySeconds: 5
  6. periodSeconds: 3

See #2761 for more details.

Remove the file /bin/ao-logs for Ansible based-operators

The inotify-tools as a dependency of Ansible based-operator images which was deprecated in the version v0.16.0 was removed. This dependency was used to output the Ansible logs in a side-car container called ansible, which has no longer been scaffolded by default. In this way, if you have any customization using this dependency be aware that it is no longer in the Ansible based-operator image and then, feel free to remove this file /bin/ao-logs for your projects since it has no longer usage. For further information also see the section Remove Ansible container sidecar on the migration guide of the version v0.16.0.

See #2852 for more details.

Modify permissions on your bundle manifests directory

Run chmod 0755 deploy/olm-catalog/<operator-name>/manifests to update your operator’s bundle directory permissions.

See #3129 for more details.

Default CRD version

The default CRD version was changed from apiextensions.k8s.io/v1beta1 to apiextensions.k8s.io/v1 for commands that create or generate CRDs. To migrate CRDs from v1beta1 to v1, simply re-run operator-sdk generate crds. To continue generating CRDs with apiextensions.k8s.io/v1beta1, use --crd-version=v1beta1.

The following CRD versions are (or will be) compatible with the following Kubernetes versions:

  • Kubernetes <= v1.15 - Only apiextensions.k8s.io/v1beta1 supported
  • Kubernetes v1.16 through v1.18 - Both apiextensions.k8s.io/v1beta1 and apiextensions.k8s.io/v1 supported
  • Kubernetes >= v1.19 - Only apiextensions.k8s.io/v1 supported

See #2874 for more details.

Helm-based operators should not use k8s.io API version suffix

Newly created Helm-based operators no longer use the k8s.io domain suffix as a default API version because it is reserved. APIs and CRDs using this domain require API review and approval. Existing projects that do not have this approval should migrate to use a new CRD API version suffix.

Make the following changes to migrate to a new group name:

  • watches.yaml - Update the group field.
  • deploy/role.yaml - Update the apiGroups field in the policy rule giving access to the CRD.
  • deploy/crds/
    • Update the metadata.name and spec.group fields in the CRD files.
    • Update the apiVersion field in the CR files.
    • Rename the CRD and CR files to reflect the new group name.

See #2859 for more details.

Kubernetes client-go contexts

Operators using client-go directly and operators that have generated clientsets will be required to update many of their client-go function calls, which have signature changes related to passing contexts and options.

Users can use http://sigs.k8s.io/clientgofix to rewrite method invocations to the new signatures.

See #2918 for more details.

Controller-runtime changes

There were various breaking changes in controller-runtime v0.6.0.

  • Informers methods now require a context parameter
  • Owns, For, and Watches funtions now take variadic functional option arguments.

See #2918 for more details.

Helm operator no longer migrates v2 releases to v3

The deprecated support for Helm v2 releases has been removed. If you require automigration of Helm v2 releases, continue using v0.17.x.

This removal does NOT affect support for v2 Helm charts, which are still supported by Helm v3.

See #2918 for more details.

Helm operator no longer supports deprecated UID-based release names

In v0.11.0, the Helm operator began using the CR name as the release name while also maintaining backwards compatibility for legacy UID-based release names. Continued support for the UID-based release names has been removed.

Before upgrading to this release, delete and recreate custom resources that are still using the deprecated UID-based release name. To determine if you have any UID-based release names still in use, inspect the status.deployedRelease field of your custom resources and look for release names that end with a 25-character random string.

See #2918 for more details.

Last modified May 28, 2020: *: release v0.18.0 (#3134) (921c3398)