Blue-green deployment
Two identical production environments. Deploy to the idle one, flip all traffic at once, keep the old one warm for instant rollback.
See it
What it is
You run two complete production environments: blue (currently serving users) and green (idle). You deploy the new release to green, smoke test it while it takes zero real traffic, then flip the router or load balancer so every user lands on green in one move. Blue sits there warm, which means rollback is another flip, not a rebuild.
Reach for it when the change is risky, the rollback needs to be instant, and you can afford two of everything for a while. Netflix popularized the same idea under the name 'red/black'. It pairs well with a health check endpoint: the flip only happens once green answers healthy.
Gotcha: the database usually is not duplicated, so both colors talk to one schema. That forces backward-compatible migrations (expand, deploy, contract) or your instant rollback flips users back to code that cannot read the new columns. In-flight sessions, warm caches, and open websockets also do not follow the flip unless you plan for them.
Ask AI for it
Set up a blue-green deployment for this service. Provision two identical environments (blue and green) behind one load balancer or router, deploy the new build to the idle color, run smoke tests plus a /health check against it with no live traffic, then switch 100 percent of traffic in a single atomic change. Keep the previous color running and untouched so rollback is one flip back. Ensure database changes are backward compatible so both colors can run against the same schema, and add a documented rollback command.