How to configure a Spark/filesystem Datasource

This guide shows how to connect to a Spark Datasource such that the data is accessible in the form of files on a local or NFS type of a filesystem.


Show Docs for Stable API (up to 0.12.x)

Prerequisites: This how-to guide assumes you have already:

To add a filesystem-backed Spark datasource do this:

  1. Run datasource new

    From the command line, run:

    great_expectations datasource new
  2. Choose “Files on a filesystem (for processing with Pandas or Spark)”

    What data would you like Great Expectations to connect to?
        1. Files on a filesystem (for processing with Pandas or Spark)
        2. Relational database (SQL)
    : 1
  3. Choose PySpark

    What are you processing your files with?
        1. Pandas
        2. PySpark
    : 2
  4. Specify the directory path for data files

    Enter the path (relative or absolute) of the root directory where the data files are stored.
    : /path/to/directory/containing/your/data/files
  5. Give your Datasource a name

    When prompted, provide a custom name for your filesystem-backed Spark data source, or hit Enter to accept the default.

    Give your new Datasource a short name.

    Great Expectations will now add a new Datasource ‘my_data_files_dir’ to your deployment, by adding this entry to your great_expectations.yml:

        class_name: SparkDFDataset
        module_name: great_expectations.dataset
      spark_config: {}
          class_name: SubdirReaderBatchKwargsGenerator
          base_directory: /path/to/directory/containing/your/data/files
      class_name: SparkDFDatasource
      Would you like to proceed? [Y/n]:
  6. Wait for confirmation

    If all goes well, it will be followed by the message:

    A new datasource 'my_data_files_dir' was added to your project.

    If you run into an error, you will see something like:

    Error: Directory '/nonexistent/path/to/directory/containing/your/data/files' does not exist.
    Enter the path (relative or absolute) of the root directory where the data files are stored.

    In this case, please check your data directory path, permissions, etc. and try again.

  7. Finally, if all goes well and you receive a confirmation on your Terminal screen, you can proceed with exploring the data sets in your new filesystem-backed Spark data source.

Show Docs for Experimental API (0.13)

Prerequisites: This how-to guide assumes you have already:

To add a Pandas filesystem datasource, do the following:

  1. Instantiate a DataContext

    import great_expectations as ge
    context = ge.get_context()
  2. List files in your directory

    Use a utility like tree on the command line or glob to list files, so that you can see how paths and filenames are formatted. Our example will use the following 3 files in the test_directory/ folder.

  3. Create or copy a yaml config

    Parameters can be set as strings, or passed in as environment variables. In the following example, a yaml config is configured for a DataSource, with a ConfiguredAssetFilesystemDataConnector and SparkDFExecutionEngine.

The config also defines TestAsset, which has name, timestamp and size as group_names, which are informative fields of the filename that are extracted by the regex pattern.

Note: The ConfiguredAssetFilesystemDataConnector used in this example is closely related to the InferredAssetFilesystemDataConnector with some key differences. More information can be found in the Core Great Expectations Concepts document.

base_directory = "test_directory/"

config = f"""
class_name: DataSource
    class_name: SparkDFExecutionEngine
        class_name: ConfiguredAssetFilesystemDataConnector
        base_directory: {base_directory}
        glob_directive: "*.csv"
                pattern: (.+)_(\\d+)_(\\d+)\\.csv
                    - name
                    - timestamp
                    - size

Additional examples of yaml configurations for various filesystems and databases can be found in the following document: How to configure DataContext components using test_yaml_config

  1. Run context.test_yaml_config.


    When executed, test_yaml_config will instantiate the component and run through a self_check procedure to verify that the component works as expected.

    The resulting output will look something like this:

    Attempting to instantiate class from config...
    Instantiating as a Datasource, since class_name is Datasource
    Instantiating class from config without an explicit class_name is dangerous. Consider adding an explicit class_name for None
        Successfully instantiated Datasource
    Execution engine: SparkDFExecutionEngine
    Data connectors:
        my_data_connector : ConfiguredAssetFilesystemDataConnector
        Available data_asset_names (1 of 1):
            Titanic (3 of 3): ['abe_20201119_200.csv', 'alex_20201212_300.csv', 'will_20201008_100.csv']
        Unmatched data_references (0 of 0): []
        Choosing an example data reference...
            Reference chosen: abe_20201119_200.csv
            Fetching batch data...
            Showing 5 rows
    0  None                                           Name  PClass  Age     Sex  Survived  SexCode
    1     1                   Allen, Miss Elisabeth Walton     1st   29  female         1        1
    2     2                    Allison, Miss Helen Loraine     1st    2  female         0        1
    3     3            Allison, Mr Hudson Joshua Creighton     1st   30    male         0        0
    4     4  Allison, Mrs Hudson JC (Bessie Waldo Daniels)     1st   25  female         0        1

    This means all has gone well and you can proceed with exploring the data sets in your new filesystem-backed Pandas data source.

    Note : In the current example, the yaml config will only create a connection to the datasource for the current session. After you exit python, the datasource and configuration will be gone. To make the datasource and configuration persistent, please add information to great_expectations.yml in your great_expectations/ directory.

Additional Notes

  1. Relative path locations should be specified from the perspective of the directory, in which the

    great_expectations datasource new

    command is executed.

  2. For the Experimental API (0.13), relative path locations should be specified from the perspective of the great_expectations/ directory.