Reverting a system to a known-good prior state, the fastest path to MTTR when a deploy caused the incident.
Rollback is the action of reverting a system, code, config, infrastructure, to a prior known-good state. Modern infrastructure (Kubernetes, Helm, Terraform, GitOps) supports rollback as a first-class operation, e.g. 'helm rollback', 'kubectl rollout undo', 'terraform plan' against the previous state. The fastest possible MTTR is when the incident's cause is a deploy and the rollback path is automated and tested.
Most production incidents follow a deploy. Teams that can roll back in under 60 seconds (one command, no human typing) compress MTTR by an order of magnitude versus teams that have to write a hotfix. Investing in tested, automated rollback is one of the highest-leverage SRE investments because the value compounds across every future deploy-related incident.
See the part of the platform that handles rollback in production.