- Rolling Update: You replace instances incrementally, batch by batch, until all are running the new version. No extra infrastructure is needed and there’s no downtime. The downside is that if something goes wrong mid-way, rollback is slow and messy.
- Canary Deployment: You route a small percentage of real traffic to the new version, watch your metrics, and gradually increase that percentage only if things look healthy. It gives you a safety gate, you can abort before the bad version reaches most users. The tradeoff is that it requires traffic splitting infrastructure and good observability to be useful.
- Blue-Green Deployment: You run two identical environments,one live, one with the new version, and switch all traffic at once when you’re ready. Rollback is instant: just flip back to the old environment. The cost is that you need to maintain double the infrastructure, at least temporarily during the release.
Rolling Update / Canary Deployment / Blue-Green Deployment
·1 min