0027. High Availability for Argo Workflows During Office Outage

• Status: accepted
• Date: 2026-06-24
• Deciders: lusoris
• Related: PR #1874

Context

An office outage exposed that the Argo Workflows controller and server did not have replicated, cross-topology placement. PR #1874 changed both components so an office-worker loss would not remove their only scheduled instance.

Decision

Run two Argo Workflows controller replicas and two server replicas, each with soft topology.kubernetes.io/zone and kubernetes.io/hostname spread constraints.

Alternatives considered

Keep one controller and one server replica; rejected because the recorded office outage removed Argo Workflows availability.

Consequences

argocd-apps/infra/argo-workflows.yaml now renders two controller pods and two server pods, with ScheduleAnyway spread constraints across zone and hostname for both components.

References

• PR #1874 — fix(ai): keep argo workflows available during office outage