Apache Parquet is a free, open-source file format that is popular for analytic workloads. Parquet stores columns together, rather than rows. It's a fixed-schema format with support for complex data structures like arrays and nested documents. These features have the following advantages:
Performant queries. Parquet is column-oriented, and therefore queries on Parquet data can be extremely performant. For example, a query that selects just one of thousands of columns can immediately extract that data from the Parquet file, rather than trying to find the desired value in each row.
Efficient storage. Parquet stores columns contiguously, which enables very efficient compression. Parquet requires that the values in a given column must have the same type, and the values in a column are generally more similar than values in other columns. This enables a wider variety of encoding and compression schemes.
Compatibility with analytics tools. Parquet files have a fixed schema, and therefore Parquet data is compatible with many analytics tools that require data in a tabular, fixed-schema format.
Atlas Data Federation can read from and write to Parquet data files.
Reading Parquet. You can query Parquet data from S3 with Atlas Data Federation. These queries might be more performant than queries on other data formats. To learn more about why queries against Parquet data might be more performant than other data formats, see About Parquet Format.
Writing Parquet. Atlas Data Federation allows you to also write data to Parquet using the $out to S3 stage. Atlas Data Federation automatically infers what Parquet schema to use based on the MongoDB data that you are writing to Parquet. You can transform your data to Parquet data format if you want to query that data with another analytics tool such as a data warehouse.