Goals

• Enable read from kube-apiserver list cache by default for container inputs
• Allow tuning of the collector rollout strategy
• Declare the existing annotations as deprecated
• Declare the new api (for rollout) and new default config setting (for cache) as GA
• Drop support for the annotations in release N+1 (e.g. 6.5)

Non-Goals

• To provide a tunable knob to modify the kubernetes_log api cache setting

Motivation

Several customers with large clusters experienced significant cluster performance issues upon upgrade due to log collection. Resolution to those issues were fixed as a bug that introduced a tech preview feature. This epic captures the work necessary for these features to become GA

Alternatives

• Continue to depend upon the annotations and graduate them to GA

Acceptance Criteria

• Verify collector deployments are configured to make read list requests from the api server cache
• Verify the collector rollout strategy is configurable through the CLF API

Risk and Assumptions

• Upstream issue claims it is possible for "inconsistencies" but does not elaborate. We are assuming the responses from the kube-apiserver will converge to a "consistent" state and the collectors will be informed of the changes.

Documentation Considerations

• API update
• API field MaxUnavailable takes precedence over annotation
• `observability.openshift.io/max-unavailable-rollout` annotation is deprecated and subject to being removed in a future release
• `observability.openshift.io/use-apiserver-cache` annotation is no longer honored being replaced by always enabled

Open Questions

Additional Notes

• There was mention by support of introducing a deprecation alert