4. Sqoop-HCatalog Interaction

HCatalog is a table and storage management service for Hadoop that enables users with different data processing tools – Pig, MapReduce, and Hive – to more easily read and write data on the grid. HCatalog’s table abstraction presents users with a relational view of data in the Hadoop distributed file system (HDFS) and ensures that users need not worry about where or in what format their data is stored: RCFile format, text files, or SequenceFiles.

HCatalog supports reading and writing files in any format for which a Hive SerDe (serializer-deserializer) has been written. By default, HCatalog supports RCFile, CSV, JSON, and SequenceFile formats. To use a custom format, you must provide the InputFormat and OutputFormat as well as the SerDe.

The ability of HCatalog to abstract various storage formats is used in providing RCFile (and future file types) support to Sqoop.

HCatalog interaction with Sqoop is patterned on an existing feature set that supports Avro and Hive tables. This section introduces five command line options, and some command line options defined for Hive are reused.

 4.2.3. Direct Mode Support

HCatalog integration in Sqoop has been enhanced to support direct mode connectors. Direct mode connectors are high performance connectors specific to a database. The Netezza direct mode connector is enhanced to use this feature for HCatalog jobs.

	Important
	Only the Netezza direct mode connector is currently enabled to work with HCatalog.

Sqoop uses read-committed transaction isolation in its mappers to import data. However, this may not be ideal for all ETL workflows, and you might want to reduce the isolation guarantees. Use the --relaxed-isolation option to instruct Sqoop to use read-uncommitted isolation level.

The read-uncommitted transaction isolation level is not supported on all databases, such as Oracle. Specifying the --relaxed-isolation may also not be supported on all databases.

	Note
	There is no guarantee that two identical and subsequent uncommitted reads will return the same data.

One of the key features of Sqoop is to manage and create the table metadata when importing into Hadoop. HCatalog import jobs also provide for this feature with the option --create-hcatalog-table. Furthermore, one of the important benefits of the HCatalog integration is to provide storage agnosticism to Sqoop data movement jobs. To provide for that feature, HCatalog import jobs provide an option that lets a user specifiy the storage format for the created table.

The option --create-hcatalog-table is used as an indicator that a table has to be created as part of the HCatalog import job.

The option --hcatalog-storage-stanza can be used to specify the storage format of the newly created table. The default value for this option is "stored as rcfile". The value specified for this option is assumed to be a valid Hive storage format expression. It will be appended to the CREATE TABLE command generated by the HCatalog import job as part of automatic table creation. Any error in the storage stanza will cause the table creation to fail and the import job will be aborted.

Any additional resources needed to support the storage format referenced in the option --hcatalog-storage-stanza should be provided to the job either by placing them in $HIVE_HOME/lib or by providing them in HADOOP_CLASSPATH and LIBJAR files.

If the option --hive-partition-key is specified, then the value of this option is used as the partitioning key for the newly created table. Only one partitioning key can be specified with this option.

Object names are mapped to the lowercase equivalents as specified below when mapped to an HCatalog table. This includes the table name (which is the same as the external store table name converted to lower case) and field names.

HCatalog supports delimited text format as one of the table storage formats. But when delimited text is used and the imported data has fields that contain those delimiters, then the data may be parsed into a different number of fields and records by Hive, thereby losing data fidelity.

For this case, one of these existing Sqoop import options can be used:

If either of these options is provided on input, then any column of type STRING will be formatted with the Hive delimiter processing and then written to the HCatalog table.

The HCatalog table should be created before using it as part of a Sqoop job if the default table creation options (with optional storage stanza) are not sufficient. All storage formats supported by HCatalog can be used with the creation of the HCatalog tables. This makes this feature readily adopt new storage formats that come into the Hive project, such as ORC files.

The Sqoop HCatalog feature supports the following table types:

Sqoop currently does not support column name mapping. However, the user is allowed to override the type mapping. Type mapping loosely follows the Hive type mapping already present in Sqoop except that SQL types “FLOAT” and “REAL” are mapped to HCatalog type “float”. In the Sqoop type mapping for Hive, these two are mapped to “double”. Type mapping is primarily used for checking the column definition correctness only and can be overridden with the --map-column-hive option.

All types except binary are assignable to a String type.

Any field of number type (int, shortint, tinyint, bigint and bigdecimal, float and double) is assignable to another field of any number type during exports and imports. Depending on the precision and scale of the target type of assignment, truncations can occur.

Furthermore, date/time/timestamps are mapped to string (the full date/time/timestamp representation) or bigint (the number of milliseconds since epoch) during imports and exports.

BLOBs and CLOBs are only supported for imports. The BLOB/CLOB objects when imported are stored in a Sqoop-specific format and knowledge of this format is needed for processing these objects in a Pig/Hive job or another Map Reduce job.

Database column names are mapped to their lowercase equivalents when mapped to the HCatalog fields. Currently, case-sensitive database object names are not supported.

Projection of a set of columns from a table to an HCatalog table or loading to a column projection is allowed (subject to table constraints). The dynamic partitioning columns, if any, must be part of the projection when importing data into HCatalog tables.

Dynamic partitioning fields should be mapped to database columns that are defined with the NOT NULL attribute (although this is not validated). A null value during import for a dynamic partitioning column will abort the Sqoop job.

All the primitive HCatalog types are supported. Currently all the complex HCatalog types are unsupported.

BLOB/CLOB database types are only supported for imports.

With the support for HCatalog added to Sqoop, any HCatalog job depends on a set of jar files being available both on the Sqoop client host and where the Map/Reduce tasks run. To run HCatalog jobs, the environment variable HADOOP_CLASSPATH must be set up as shown below before launching the Sqoop HCatalog jobs.

HADOOP_CLASSPATH=$(hcat -classpath)
export HADOOP_CLASSPATH
            

The necessary HCatalog dependencies will be copied to the distributed cache automatically by the Sqoop job.