0026. Stateless Cluster Helper Failover and Scheduling Adjustments

• Status: accepted
• Date: 2026-06-24
• Deciders: lusoris
• Related: PR #1877

Context

During an office-worker outage, stateless infrastructure helpers pinned to the office zone could not run elsewhere. PR #1877 changed the placement of Argo Rollouts, Reflector, and Reloader so those helpers could schedule on Linux workers outside that site.

Decision

Replace office-zone selectors with kubernetes.io/os: linux for Argo Rollouts, Reflector, and Reloader. Run two Argo Rollouts controller and dashboard replicas and two HA Reloader replicas, with soft zone and hostname topology spread for the replicated components.

Alternatives considered

Keep the helpers pinned to the office zone; rejected because that placement left them unschedulable during the recorded office-worker outage.

Consequences

argocd-apps/infra/argo-rollouts.yaml now configures two controllers and two dashboard replicas with soft zone/hostname spread; reloader.yaml enables HA with two similarly spread replicas; and reflector.yaml can run on any Linux worker.

References

• PR #1877 — fix(infra): fail over stateless cluster helpers