I developed the second iteration of our Orchestration tool to enable different deployment models for sets of customers. The software component pods are deployed from a manifest of images that constitute a production release version.
This added a PostgreSQL configuration management database. Hundreds of GCP projects have an agent running which reports back cluster state. The orchestrator holds target and current state and when a new version has passed testing, Argo workflows triggers a new deployment phase by messaging a Go receiver microservice which sets new target states for the relevant projects. First to canary deployments and once that is successful to sets of customer projects.