Query Apache Parquet Files

You can use BlazingSQL to SQL query Apache Parquet files.

Apache Parquet has the following characteristics:

  • Self-describing

  • Columnar format

  • Compressed

BlazingSQL relies on cuIO io when reading Parquet files, which means ALL of the decompression happens on GPUs, dramatically reducing GPU over PCIe bandwidth concerns.

Querying Apache Parquet is straightforward.

bc.create_table('table_name', '...file_path/table.parquet')


  • table_name - string. Required. Name of the table you are creating.

  • file_path - string. Required. Location of the file you are creating a table off of.

  • file_format - string. Optional. A string describing the file format ("csv", "orc", "parquet"). This field must only be set if the files do not have an extension. Default: the engine will infer the file format by the extension.

You can also query multiple Parquet files. For now, you need a List of all the Parquet files.