You must deploy a new Cloud Run revision with SRE-mandated gradual rollout to reduce risk and minimize effort. What should you do?

The correct approach is to deploy the new revision without immediately serving all traffic, assign it a small traffic percentage, monitor for errors, and incrementally increase traffic. This describes a gradual rollout (often called a canary deployment), which is a core SRE practice for reducing risk. Cloud Run natively supports traffic splitting, allowing you to direct a percentage of requests to a new revision while the majority still goes to the stable one. This minimizes impact if issues arise and allows for controlled observation. Configuring a frontend load balancer to toggle revisions is less efficient and doesn't provide granular control over traffic percentages like Cloud Run's built-in traffic management. Modifying application code for traffic splitting is a complex and error-prone solution when Cloud Run offers this functionality natively. Deploying immediately and rolling back if needed is a "big bang" deployment, which carries high risk and is precisely what gradual rollouts aim to avoid.