Suppose we have 1,000+ instances for deployment, and we are deploying in 10 phases at 10% per phase.

Currently, we are in Phase 9, which means 900 instances have already been deployed to the new ASG across the previous 9 phases.

The problem arises during a rollback — when the pipeline triggers a rollback, the old ASG spins up 900+ instances all at once. Only after those 900+ instances are healthy does the new ASG begin terminating its instances. This sudden surge of 900+ servers coming up simultaneously is not ideal and puts significant strain on the system.

What we need instead is a phased rollback, meaning the rollback should mirror the deployment strategy: rolling back 10% at a time (phase by phase), rather than restoring all 900 instances in one go.