What is the difference between Kafka's @KafkaListener and manual polling?

@KafkaListener (Spring for Apache Kafka) marks a bean method as a listener for a listener container: the bean is wrapped in a MessagingMessageListenerAdapter that converts the data to match the method parameters. It needs @EnableKafka and a listener container factory (by default a bean named kafkaListenerContainerFactory) that configures the underlying ConcurrentMessageListenerContainer. The container runs the consumer poll() loop for you: each KafkaMessageListenerContainer consumes on a single thread, and the concurrency setting creates several of them. The container also handles offset commits (the default AckMode is BATCH, committing once all records returned by a poll() are processed, and enable.auto.commit is set to false unless you set it) and error handling (the DefaultErrorHandler retries a failed record, FixedBackOff(0L, 9) by default, and then logs it). With manual polling you create a KafkaConsumer, call poll() in your own loop, and handle exceptions, offset commits and thread safety yourself, since KafkaConsumer is not thread-safe. The listener container is the common choice in Spring applications; manual polling gives full control.

Expected key concepts: @KafkaListener, @EnableKafka, listener container, ConcurrentMessageListenerContainer, AckMode, BATCH, DefaultErrorHandler, KafkaConsumer, poll, offset commit