The Confluent Parallel Consumer is an experimental JVM-based client library for Apache Kafka that provides an alternate approach to parallelism, subdividing the unit of work from a partition down to a key or even a message. It offers three new features: increased parallelism beyond the number of partitions, key-level ordering guarantees, and non-blocking I/O to efficiently call out to other services. The library addresses issues such as slow consumer scenarios, where messages are held up by slow-to-process messages, and provides high-performance handling for these situations. By processing at three different levels - partition, key, or unordered - the Confluent Parallel Consumer increases performance while addressing common problems in Kafka-based applications.