[ https://issues.apache.org/jira/browse/DRILL-7233?page=com.atlassian.jira.plugin.system.issuetabpanels:all-tabpanel ]
Paul Rogers updated DRILL-7233: ------------------------------- Reviewer: Paul Rogers Labels: doc-impacting ready-to-commit (was: doc-impacting) > Format Plugin for HDF5 > ---------------------- > > Key: DRILL-7233 > URL: https://issues.apache.org/jira/browse/DRILL-7233 > Project: Apache Drill > Issue Type: New Feature > Affects Versions: 1.17.0 > Reporter: Charles Givre > Assignee: Charles Givre > Priority: Major > Labels: doc-impacting, ready-to-commit > Fix For: 1.18.0 > > > h2. Drill HDF5 Format Plugin > h2. > 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. > h3. Configuration > 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. > defaultPath: > h3. Example Configuration > h3. > For most uses, the configuration below will suffice to enable Drill to query > HDF5 files. > {{"hdf5": { > "type": "hdf5", > "extensions": [ > "h5" > ], > "defaultPath": null > }}} > h3. Usage > 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`; > +---------------------+ > | int_data | > +---------------------+ > | [1,2,3,4,5,6] | > | [7,8,9,10,11,12] | > | [13,14,15,16,17,18] | > | [19,20,21,22,23,24] | > +---------------------+}} > 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. > h3. Attributes > 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`; > +----------------------------------------------------------------------------------+ > | attributes > | > +----------------------------------------------------------------------------------+ > | {} > | > | {"__TYPE_VARIANT__":"TIMESTAMP_MILLISECONDS_SINCE_START_OF_THE_EPOCH"} > | > | {} > | > | {} > | > | > {"important":false,"__TYPE_VARIANT__timestamp__":"TIMESTAMP_MILLISECONDS_SINCE_START_OF_THE_EPOCH","timestamp":1550033296762} > | > | {} > | > | {} > | > | {} > | > +----------------------------------------------------------------------------------+ > 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 | > +---------+-----------+-------------+}} > h3. Known Limitations > h3. > 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. -- This message was sent by Atlassian Jira (v8.3.4#803005)