Description
Describe the feature
currently every time knative does rollout for a new revision it requires 2x resources when minReplica is set for a short time of period and then release the resource required for running the old revision after it is scaled down, this is a problem for platform that are enabled with resource quota and user has to budget 2x resources to run the service, it becomes a bigger problem when service is running on precious GPU hardware.
With progressive rollout, do we still need to wait for all the minReplicas to be ready to start migrating the traffic from old revision to new revision? Can we do something like rolling update, when 20% is moved to the new revision we can then start scaling down the old revision accordingly, so that we do not need to require 2x resources during the rollout.