Hive Atlas Bridge

Hive Model

The default hive model includes the following types:

  • Entity types:
    • hive_db
      • super-types: Referenceable
      • attributes: name, clusterName, description, locationUri, parameters, ownerName, ownerType
    • hive_storagedesc
      • super-types: Referenceable
      • attributes: cols, location, inputFormat, outputFormat, compressed, numBuckets, serdeInfo, bucketCols, sortCols, parameters, storedAsSubDirectories
    • hive_column
      • super-types: Referenceable
      • attributes: name, type, comment, table
    • hive_table
      • super-types: DataSet
      • attributes: name, db, owner, createTime, lastAccessTime, comment, retention, sd, partitionKeys, columns, aliases, parameters, viewOriginalText, viewExpandedText, tableType, temporary
    • hive_process
      • super-types: Process
      • attributes: name, startTime, endTime, userName, operationType, queryText, queryPlan, queryId
    • hive_column_lineage
      • super-types: Process
      • attributes: query, depenendencyType, expression

  • Enum types:
    • hive_principal_type
      • values: USER, ROLE, GROUP

  • Struct types:
    • hive_order
      • attributes: col, order
    • hive_serde
      • attributes: name, serializationLib, parameters

The entities are created and de-duped using unique qualified name. They provide namespace and can be used for querying/lineage as well. Note that dbName, tableName and columnName should be in lower case. clusterName is explained below.

  • hive_db.qualifiedName - <dbName>@<clusterName>
  • hive_table.qualifiedName - <dbName>.<tableName>@<clusterName>
  • hive_column.qualifiedName - <dbName>.<tableName>.<columnName>@<clusterName>
  • hive_process.queryString - trimmed query string in lower case

Importing Hive Metadata

org.apache.atlas.hive.bridge.HiveMetaStoreBridge imports the Hive metadata into Atlas using the model defined above. import-hive.sh command can be used to facilitate this.

    Usage: <atlas package>/hook-bin/import-hive.sh

The logs are in <atlas package>/logs/import-hive.log

Hive Hook

Atlas Hive hook registers with Hive to listen for create/update/delete operations and updates the metadata in Atlas, via Kafka notifications, for the changes in Hive. Follow the instructions below to setup Atlas hook in Hive:

  • Set-up Atlas hook in hive-site.xml by adding the following:
    <property>
      <name>hive.exec.post.hooks</name>
      <value>org.apache.atlas.hive.hook.HiveHook</value>
    </property>

  • Add 'export HIVE_AUX_JARS_PATH=<atlas package>/hook/hive' in hive-env.sh of your hive configuration
  • Copy <atlas-conf>/atlas-application.properties to the hive conf directory.

The following properties in <atlas-conf>/atlas-application.properties control the thread pool and notification details:

  • atlas.hook.hive.synchronous - boolean, true to run the hook synchronously. default false. Recommended to be set to false to avoid delays in hive query completion.
  • atlas.hook.hive.numRetries - number of retries for notification failure. default 3
  • atlas.hook.hive.minThreads - core number of threads. default 1
  • atlas.hook.hive.maxThreads - maximum number of threads. default 5
  • atlas.hook.hive.keepAliveTime - keep alive time in msecs. default 10
  • atlas.hook.hive.queueSize - queue size for the threadpool. default 10000

Refer Configuration for notification related configurations

Column Level Lineage

Starting from 0.8-incubating version of Atlas, Column level lineage is captured in Atlas. Below are the details

Model

  • ColumnLineageProcess type is a subtype of Process

  • This relates an output Column to a set of input Columns or the Input Table

  • The lineage also captures the kind of dependency, as listed below:
    • SIMPLE: output column has the same value as the input
    • EXPRESSION: output column is transformed by some expression at runtime (for e.g. a Hive SQL expression) on the Input Columns.
    • SCRIPT: output column is transformed by a user provided script.

  • In case of EXPRESSION dependency the expression attribute contains the expression in string form

  • Since Process links input and output DataSets, Column is a subtype of DataSet

Examples

For a simple CTAS below:

create table t2 as select id, name from T1

The lineage is captured as

Extracting Lineage from Hive commands

* The HiveHook maps the LineageInfo in the HookContext to Column lineage instances

* The LineageInfo in Hive provides column-level lineage for the final FileSinkOperator, linking them to the input columns in the Hive Query

NOTES

  • Column level lineage works with Hive version 1.2.1 after the patch for HIVE-13112 is applied to Hive source
  • Since database name, table name and column names are case insensitive in hive, the corresponding names in entities are lowercase. So, any search APIs should use lowercase while querying on the entity names
  • The following hive operations are captured by hive hook currently
    • create database
    • create table/view, create table as select
    • load, import, export
    • DMLs (insert)
    • alter database
    • alter table (skewed table information, stored as, protection is not supported)
    • alter view