A rolling update replaces old service instances with new ones incrementally, one at a time or in small batches. The service stays available through the rollout, without requiring a second full fleet. section 13.3.1 section 13.12
A typical rollout proceeds instance by instance:
- Take an old instance out of load-balancer rotation, replace it with the new version, check its health, and return it to rotation; then repeat for the remaining instances. section 13.3.1
- Expect both versions to serve traffic during the transition. Keep this dual-version window in mind when changing database schemas or other shared state. section 13.3.1 section 13.12
Four machines, one upgrade at a time
An Order Service runs on four virtual machines behind a load balancer. Upgrade the first machine to version 2.0, confirm it passes health checks, and continue through the other three. During the rollout, a user’s requests may reach either version. This approach avoids extra fleet hardware, but offers less rollback safety than a Blue-Green deployment. section 13.3.1 section 13.3.5
[Load Balancer] ├── Instance 1: v2.0 (Active) ├── Instance 2: v1.0 -> Upgrading to v2.0 ├── Instance 3: v1.0 (Active) └── Instance 4: v1.0 (Active)
The instance-by-instance diagram makes the mixed-version state during an update visible. section 13.3.1
Choose rolling updates when infrastructure cost is the main constraint. They have low resource cost, medium operational complexity, and high dual-version risk; they do not provide an instant rollback. Blue-Green favors fast rollback, while Canary limits exposure to a small share of traffic. section 13.3.5








