Well. The benefits (monitoring, telemetry/observability) are nice enough that I might think about running it on k8s in prod. But only after sufficient number of dev/staging clusters have been sacrificed.