Changing a Deployment's Pod template, its image, environment or resources, starts a rollout. With the default RollingUpdate strategy, a new ReplicaSet grows a few Pods at a time while the old one shrinks, so the application keeps serving throughout. Two numbers set the pace: maxSurge, how many Pods may run above the desired count, and maxUnavailable, how many may be missing. Both default to 25%.
If the new Pods never become Ready, the rollout stops after the first few and the old Pods keep serving. kubectl rollout status waits for the end or reports the stall, which is why release pipelines run it.
kubectl set image deployment/NAME CONTAINER=IMAGE names the container, so in a Pod with a sidecar only that one changes. The Recreate strategy stops every old Pod first: a short outage, used when two versions must never run together.
A new ReplicaSet takes over from the old one.