How baseline instead of stable in "rollout-baseline-vs-canary" example

[dsbos]

In the rollout-baseline-vs-canary example currently at https://github.com/argoproj/argo-rollouts/blob/master/examples/rollout-baseline-vs-canary.yaml, the comment says:

This example demonstrates how [to] start a new baseline and canary stack as part of the Rollout update. Comparing a canary against a baseline (as opposed to existing production instances) ...

How is that example running a new baseline version that is separate from and in addition to a stable/production version?

Won't that Rollout maintain the stable/production version like a normal Rollout (transitioning from the previous version to a new version when the pod template spec. changes)?

And don't the specRef members simply refer to the pods/ReplicaSets running those previous and new versions? (The template names (baseline and canary) don't have any semantic meaning to Argo Rollouts, do they?)

---

From the comment at the top, I thought that the example would cause running of three things (during a triggered re-deployment):

• a ReplicaSet/Pods still running the previous version of the pod. spec, running for production (setting one set of pod template hashes or Istio subsets)
• a new, temporary, separate ReplicaSet/Pods also running the previous version, but doing something slightly different with hashes/subsets so production traffic doesn't hit it (so it's clean baseline?), and
• a new, temporary ReplicaSet/Pods running the new version of the pod template spec., running as a canary version (setting some pod template hashes or Istio subsets)

However, I don't yet see how it would do those three things instead of just the normal two that canarying (or blue-green deployment) does.