Every format in this chapter ends up compressed somewhere: Avro 129 blocks, Parquet 129 and ORC pages, Arrow 129 buffers, Kafka 129 record batches (Apache Kafka and Managed Cloud Kafka) and gzipped JSON Lines in object storage. The codec is a trade between three numbers, size ratio, compression speed and decompression speed, and the right balance depends on whether data is written once and read often, or streamed through as fast as possible.