You must deploy an internet-facing microservices app on GKE and run A/B tests with no downtime, sending a subset of production traffic to new releases. Which deployment approach meets these requirements?

The correct option describes a blue/green or canary deployment strategy using a service mesh, specifically Istio (part of Anthos Service Mesh). This approach allows for fine-grained traffic management, enabling a small percentage of traffic to be routed to the new version (A/B testing) without downtime. If issues arise, traffic can be instantly shifted back. The first incorrect option describes a basic rolling update, which doesn't inherently support A/B testing or controlled traffic splitting for new releases. The second incorrect option describes deploying to a separate namespace and then switching all traffic via Ingress, which is a blue/green deployment but lacks the ability to send a "subset" of traffic for A/B testing; it's an all-or-nothing switch. The fourth incorrect option describes a standard rolling update, which replaces pods gradually but doesn't provide the traffic splitting capabilities needed for A/B testing.