A DMS CDC task falls behind. Source capture latency stays low, but target latency grows and the target reports write throttling. Which investigation is best supported by this evidence?

1. Increase only log retention and assume it removes target write throttling.
2. Assume the source transaction log is always the bottleneck and disable CDC logging.
3. Delete the task checkpoints and restart from an arbitrary position without assessing data loss.
4. Investigate target capacity and apply performance, including batch configuration and resource bottlenecks, while preserving correctness.