What is the difference between concurrency and partitions in Spring Kafka's listener container?

Concurrency, set on the container factory or overridden by the concurrency attribute of @KafkaListener, is the number of consumers the container creates: ConcurrentMessageListenerContainer delegates to that many KafkaMessageListenerContainer instances, each with its own KafkaConsumer. Partitions belong to the topic. With topics or a topic pattern, Kafka's group management distributes the partitions among those consumers, so concurrency=3 over 6 partitions gives each consumer some of the partitions (two each if the assignor spreads them evenly). With explicit TopicPartitionOffsets, a concurrency above the number of partitions is adjusted down so each container gets one partition; the same documented example of 6 partitions and concurrency=3 gives each container two. With group management, a consumer that receives no partition sits idle: three topics of five partitions each with concurrency=15 and the default RangeAssignor leave only five active consumers, and RoundRobinAssignor is one way to spread them. Concurrency is per application instance; to add consumers across JVMs, run more instances of the application, and a common practice is not to exceed the partition count because extra consumers would be idle.

Expected key concepts: concurrency, ConcurrentMessageListenerContainer, KafkaMessageListenerContainer, KafkaConsumer, partitions, idle, RangeAssignor, RoundRobinAssignor, instances