The correct option is They are created at the beginning of each job and they are deleted once that job completes. This aligns with how GitHub Actions scopes service containers to the lifetime of the job so they are available to every step in that job and are then cleaned up when the job ends.

Services defined under a job are started before any steps in that job run and they persist for the duration of the job. GitHub then stops and removes them as part of job teardown. This ensures clean isolation between jobs and consistent environments for integration tests, including matrix jobs where each job gets its own fresh set of services.

They are provisioned when the workflow starts and they are removed only after the workflow run finishes is incorrect because services are not workflow scoped. They do not span multiple jobs and they do not live for the entire run.

They start only when a step directly references the service and they stop when that step ends is incorrect because services are not started lazily per step. They are brought up at job start and remain available to all steps until the job completes.

They are created one time for the runner session and they are reused across multiple jobs on the same runner is incorrect because GitHub hosted runners provide a fresh environment for each job and even with self hosted runners services defined in a job are still created and cleaned up per job rather than reused across jobs.

Full AWS Practitioner Certification Question