A checkout service on GKE needs a promotional banner turned on for 10 percent of users and turned off later. Changes must reach running pods within about a minute, and no image rebuild may be required. Which approach fits best?
Show answer and explanation
Answer: C
A remote configuration service, or a ConfigMap that the application reloads, lets the team change a value without touching the container image. A flag service can also evaluate percentage rollouts, so only a share of users sees the banner, and the decision is consistent for each user. Application code reads the current value on each check or on a refresh interval, so a change takes effect across all replicas within about a minute.
Why the other options are wrong
- A. An environment variable is fixed when the container starts. Every flag change would require a rebuild and a rollout, which is exactly the cost the team wants to avoid.
- B. Node pools are infrastructure units. Creating one per variant multiplies cost and operational work, and it cannot split individual users into a stable 10 percent cohort.
- D. DNS answers are cached and propagate slowly, and TXT records are not a feature-flag mechanism. Using them to split users per request is impractical and unreliable.