Gradual Rollout and Rollback
But that on/off binary is not the whole risk reduction story. After all, you can have problems that don’t crop up until a hundred or a thousand users start using something. With any decently sized user base, you’ll know that there’s a huge difference between on and off.
This brings into play the idea of gradual rollout. Introduce a brand new feature using a feature flag, and then, leveraging target groups, you can expose users gradually to it. Measure its performance with just a few users, then crank up the dial to dozens or hundreds, evaluating every step of the way. If you hit a problem, you can simply turn the news feature off. Or you can roll back gradually as well, getting back to safe levels while fixing the problem.
In a world where downtime is no longer acceptable, you can’t afford big bang deployment or other risky rollout strategies. Use gradual rollout to deploy new features and functionality with confidence.