Uploaded image for project: 'Apache Drill'
  1. Apache Drill
  2. DRILL-7233

Format Plugin for HDF5



    • New Feature
    • Status: Resolved
    • Major
    • Resolution: Fixed
    • 1.17.0
    • 1.18.0
    • None


      Drill HDF5 Format Plugin

      Per wikipedia, Hierarchical Data Format (HDF) is a set of file formats designed to store and organize large amounts of data. Originally developed at the National Center for Supercomputing Applications, it is supported by The HDF Group, a non-profit corporation whose mission is to ensure continued development of HDF5 technologies and the continued accessibility of data stored in HDF.

      This plugin enables Apache Drill to query HDF5 files.


      There are three configuration variables in this plugin:

      type: This should be set to hdf5.
      extensions: This is a list of the file extensions used to identify HDF5 files. Typically HDF5 uses .h5 or .hdf5 as file extensions. This defaults to .h5.

      Example Configuration

      For most uses, the configuration below will suffice to enable Drill to query HDF5 files.


      { "type": "hdf5", "extensions": [ "h5" ], "defaultPath": null }



      Since HDF5 can be viewed as a file system within a file, a single file can contain many datasets. For instance, if you have a simple HDF5 file, a star query will produce the following result:

      {{apache drill> select * from dfs.test.`dset.h5`;

      path data_type file_name int_data


      /dset DATASET dset.h5 [[1,2,3,4,5,6],[7,8,9,10,11,12],[13,14,15,16,17,18],[19,20,21,22,23,24]]

      The actual data in this file is mapped to a column called int_data. In order to effectively access the data, you should use Drill's FLATTEN() function on the int_data column, which produces the following result.

      {{apache drill> select flatten(int_data) as int_data from dfs.test.`dset.h5`;




      Once you have the data in this form, you can access it similarly to how you might access nested data in JSON or other files.

      {{apache drill> SELECT int_data[0] as col_0,
      . .semicolon> int_data[1] as col_1,
      . .semicolon> int_data[2] as col_2
      . .semicolon> FROM ( SELECT flatten(int_data) AS int_data
      . . . . . .)> FROM dfs.test.`dset.h5`
      . . . . . .)> );

      col_0 col_1 col_2


      1 2 3
      7 8 9
      13 14 15
      19 20 21

      Alternatively, a better way to query the actual data in an HDF5 file is to use the defaultPath field in your query. If the defaultPath field is defined in the query, or via the plugin configuration, Drill will only return the data, rather than the file metadata.

        • Note: Once you have determined which data set you are querying, it is advisable to use this method to query HDF5 data. **

      You can set the defaultPath variable in either the plugin configuration, or at query time using the table() function as shown in the example below:

      {{SELECT *
      FROM table(dfs.test.`dset.h5` (type => 'hdf5', defaultPath => '/dset'))}}
      This query will return the result below:

      {{apache drill> SELECT * FROM table(dfs.test.`dset.h5` (type => 'hdf5', defaultPath => '/dset'));

      int_col_0 int_col_1 int_col_2 int_col_3 int_col_4 int_col_5


      1 2 3 4 5 6
      7 8 9 10 11 12
      13 14 15 16 17 18
      19 20 21 22 23 24

      4 rows selected (0.223 seconds)}}

      If the data in defaultPath is a column, the column name will be the last part of the path. If the data is multidimensional, the columns will get a name of <data_type>_col_n . Therefore a column of integers will be called int_col_1.


      Occasionally, HDF5 paths will contain attributes. Drill will map these to a map data structure called attributes, as shown in the query below.

      {{apache drill> SELECT attributes FROM dfs.test.`browsing.h5`;




      8 rows selected (0.292 seconds)}}
      You can access the individual fields within the attributes map by using the structure table.map.key. Note that you will have to give the table an alias for this to work properly.

      {{apache drill> SELECT path, data_type, file_name
      FROM dfs.test.`browsing.h5` AS t1 WHERE t1.attributes.important = false;

      path data_type file_name


      /groupB GROUP browsing.h5


      Known Limitations

      There are several limitations with the HDF5 format plugin in Drill.

      • Drill cannot read unsigned 64 bit integers. When the plugin encounters this data type, it will write an INFO message to the log.
      • Drill cannot read compressed fields in HDF5 files.
      • HDF5 files can contain nested data sets of up to n dimensions. Since Drill works best with two dimensional data, datasets with more than two dimensions are flattened.


        Issue Links



              cgivre Charles Givre
              cgivre Charles Givre
              Paul Rogers Paul Rogers
              0 Vote for this issue
              3 Start watching this issue