Avro in Beam Before, we always talked about PCollections that were not aware about the structure of their records. For example, in this post we have had to parse text files manually, extract the fields and perform GroupByKey operation based on this. This is inconvenient and error-prone way of working with structured data. It can be done sufficiently easier and in more clear and elegant way if we had schema-based records as an input data collection. For the moment, there are many different data structure formats “on the market” - JSON, Avro, Parquet, Protocol Buffers, etc. - that could be used for structuring our data in more fancy way than just plain text. Apache Beam provides a good support of Avro format , so we will see in this post how to work efficiently with it. First of all, there is an AvroIO , which allows to work with files, written in Avro format, in Beam pipelines. Under the hood, it uses FileIO , so it means that we can read/write files from/to different File Systems