A partition leader handles writes and coordinates followers. When it fails, an eligible in-sync replica can become leader. That replacement depends on available replicas and cluster policy; a replica assignment alone does not guarantee an eligible survivor. These tasks start with three synchronized copies and fail one broker. Produce through the surviving leader without bringing the failed broker back. The visible partition state shows which replica took over. The model performs the election immediately to keep the exercise deterministic. A real incident includes detection and election delays, client metadata refresh and retries. A successful recovery claim should state which failure was tested and which durability requirement stayed in force.
Inspect the topics, partition logs and group offsets before changing the cluster.