19. Envers

19.1. Basics

To audit changes that are performed on an entity, you only need two things:

• the hibernate-envers jar on the classpath,

• an @Audited annotation on the entity.

And that’s all. You can create, modify and delete the entities as always.

If you look at the generated schema for your entities, or at the data persisted by Hibernate, you will notice that there are no changes. However, for each audited entity, a new table is introduced - entity_table_AUD, which stores the historical data, whenever you commit a transaction.

Instead of annotating the whole class and auditing all properties, you can annotate only some persistent properties with @Audited. This will cause only these properties to be audited.

The audit (history) of an entity can be accessed using the AuditReader interface, which can be obtained having an open EntityManager or Session via the AuditReaderFactory. See the Javadocs) for these classes for details on the functionality offered.

19.2. Configuration

It is possible to configure various aspects of Hibernate Envers behavior, such as table names, etc.

Table 8. Envers Configuration Properties

19.3. Additional mapping annotations

The name of the audit table can be set on a per-entity basis, using the @AuditTable annotation. It may be tedious to add this annotation to every audited entity, so if possible, it’s better to use a prefix/suffix.

If you have a mapping with secondary tables, audit tables for them will be generated in the same way (by adding the prefix and suffix). If you wish to overwrite this behaviour, you can use the @SecondaryAuditTable and @SecondaryAuditTables annotations.

If you’d like to override auditing behaviour of some fields/properties inherited from @MappedSuperclass or in an embedded component, you can apply the @AuditOverride( s ) annotation on the subtype or usage site of the component.

If you want to audit a relation mapped with @OneToMany and @JoinColumn, please see Mapping exceptions for a description of the additional @AuditJoinTable annotation that you’ll probably want to use.

If you want to audit a relation, where the target entity is not audited (that is the case for example with dictionary-like entities, which don’t change and don’t have to be audited), just annotate it with @Audited( targetAuditMode = RelationTargetAuditMode.NOT_AUDITED ). Then, while reading historic versions of your entity, the relation will always point to the "current" related entity. By default Envers throws javax.persistence.EntityNotFoundException when "current" entity does not exist in the database. Apply @NotFound( action = NotFoundAction.IGNORE ) annotation to silence the exception and assign null value instead. Hereby solution causes implicit eager loading of to-one relations.

If you’d like to audit properties of a superclass of an entity, which are not explicitly audited (they don’t have the @Audited annotation on any properties or on the class), you can set the @AuditOverride( forClass = SomeEntity.class, isAudited = true/false ) annotation.

19.4. Choosing an audit strategy

After the basic configuration, it is important to choose the audit strategy that will be used to persist and retrieve audit information. There is a trade-off between the performance of persisting and the performance of querying the audit information. Currently there are two audit strategies.

1. The default audit strategy persists the audit data together with a start revision. For each row inserted, updated or deleted in an audited table, one or more rows are inserted in the audit tables, together with the start revision of its validity. Rows in the audit tables are never updated after insertion. Queries of audit information use subqueries to select the applicable rows in the audit tables.

   | | These subqueries are notoriously slow and difficult to index. | | --- | --- |

2. The alternative is a validity audit strategy. This strategy stores the start-revision and the end-revision of audit information. For each row inserted, updated or deleted in an audited table, one or more rows are inserted in the audit tables, together with the start revision of its validity. But at the same time the end-revision field of the previous audit rows (if available) are set to this revision. Queries on the audit information can then use 'between start and end revision' instead of subqueries as used by the default audit strategy.

   The consequence of this strategy is that persisting audit information will be a bit slower because of the extra updates involved,
   but retrieving audit information will be a lot faster.
   This can be improved even further by adding extra indexes.

19.5. Revision Log

When Envers starts a new revision, it creates a new revision entity which stores information about the revision. By default, that includes just:

• revision number

• An integral value (int/Integer or long/Long). Essentially the primary key of the revision

• revision timestamp

• either a long/Long or java.util.Date value representing the instant at which the revision was made. When using a java.util.Date, instead of a long/Long for the revision timestamp, take care not to store it to a column data type which will loose precision.

Envers handles this information as an entity. By default it uses its own internal class to act as the entity, mapped to the REVINFO table. You can, however, supply your own approach to collecting this information which might be useful to capture additional details such as who made a change or the ip address from which the request came. There are two things you need to make this work:

1. First, you will need to tell Envers about the entity you wish to use. Your entity must use the @org.hibernate.envers.RevisionEntity annotation. It must define the two attributes described above annotated with @org.hibernate.envers.RevisionNumber and @org.hibernate.envers.RevisionTimestamp, respectively. You can extend from org.hibernate.envers.DefaultRevisionEntity, if you wish, to inherit all these required behaviors.

   Simply add the custom revision entity as you do your normal entities and Envers will _find it_.

   | | It is an error for there to be multiple entities marked as @org.hibernate.envers.RevisionEntity | | --- | --- |

2. Second, you need to tell Envers how to create instances of your revision entity which is handled by the newRevision( Object revisionEntity ) method of the org.hibernate.envers.RevisionListener interface.

   You tell Envers your custom `org.hibernate.envers.RevisionListener` implementation to use by specifying it on the `@org.hibernate.envers.RevisionEntity` annotation, using the value attribute.
   If your `RevisionListener` class is inaccessible from `@RevisionEntity` (e.g. it exists in a different module), set `org.hibernate.envers.revision_listener` property to its fully qualified class name.
   Class name defined by the configuration parameter overrides revision entity's value attribute.

@RevisionEntity( MyCustomRevisionListener.class )
public class MyCustomRevisionEntity {
    ...
}

public class MyCustomRevisionListener implements RevisionListener {
    public void newRevision( Object revisionEntity ) {
        MyCustomRevisionEntity customRevisionEntity = ( MyCustomRevisionEntity ) revisionEntity;
    }
}

Example 384. ExampleRevEntity.java

package `org.hibernate.envers.example;`

import `org.hibernate.envers.RevisionEntity;`
import `org.hibernate.envers.DefaultRevisionEntity;`

import javax.persistence.Entity;

@Entity
@RevisionEntity( ExampleListener.class )
public class ExampleRevEntity extends DefaultRevisionEntity {
    private String username;

    public String getUsername() { return username; }
    public void setUsername( String username ) { this.username = username; }
}

Example 385. ExampleListener.java

package `org.hibernate.envers.example;`

import `org.hibernate.envers.RevisionListener;`
import org.jboss.seam.security.Identity;
import org.jboss.seam.Component;

public class ExampleListener implements RevisionListener {

    public void newRevision( Object revisionEntity ) {
        ExampleRevEntity exampleRevEntity = ( ExampleRevEntity ) revisionEntity;
        Identity identity =
            (Identity) Component.getInstance( "org.jboss.seam.security.identity" );

        exampleRevEntity.setUsername( identity.getUsername() );
    }
}

19.6. Tracking entity names modified during revisions

By default entity types that have been changed in each revision are not being tracked. This implies the necessity to query all tables storing audited data in order to retrieve changes made during specified revision. Envers provides a simple mechanism that creates REVCHANGES table which stores entity names of modified persistent objects. Single record encapsulates the revision identifier (foreign key to REVINFO table) and a string value.

Tracking of modified entity names can be enabled in three different ways:

1. Set org.hibernate.envers.track_entities_changed_in_revision parameter to true. In this case org.hibernate.envers.DefaultTrackingModifiedEntitiesRevisionEntity will be implicitly used as the revision log entity.

2. Create a custom revision entity that extends org.hibernate.envers.DefaultTrackingModifiedEntitiesRevisionEntity class.

   @RevisionEntity
   public class ExtendedRevisionEntity extends DefaultTrackingModifiedEntitiesRevisionEntity {
       ...
   }
   

3. Mark an appropriate field of a custom revision entity with @org.hibernate.envers.ModifiedEntityNames annotation. The property is required to be of Set<String> type.

   @RevisionEntity
   public class AnnotatedTrackingRevisionEntity {
       ...
   
       @ElementCollection
       @JoinTable( name = "REVCHANGES", joinColumns = @JoinColumn( name = "REV" ) )
       @Column( name = "ENTITYNAME" )
       @ModifiedEntityNames
       private Set<String> modifiedEntityNames;
   
       ...
   }
   

   Users, that have chosen one of the approaches listed above, can retrieve all entities modified in a specified revision by utilizing API described in Querying for entities modified in a given revision.

Users are also allowed to implement custom mechanism of tracking modified entity types. In this case, they shall pass their own implementation of org.hibernate.envers.EntityTrackingRevisionListener interface as the value of @org.hibernate.envers.RevisionEntity annotation. EntityTrackingRevisionListener interface exposes one method that notifies whenever audited entity instance has been added, modified or removed within current revision boundaries.

Example 386. CustomEntityTrackingRevisionListener.java

public class CustomEntityTrackingRevisionListener implements EntityTrackingRevisionListener {

    @Override
    public void entityChanged( Class entityClass, String entityName,
                               Serializable entityId, RevisionType revisionType,
                               Object revisionEntity ) {
        String type = entityClass.getName();
        ( ( CustomTrackingRevisionEntity ) revisionEntity ).addModifiedEntityType( type );
    }

    @Override
    public void newRevision( Object revisionEntity ) {
    }
}

Example 387. CustomTrackingRevisionEntity.java

@Entity
@RevisionEntity( CustomEntityTrackingRevisionListener.class )
public class CustomTrackingRevisionEntity {

    @Id
    @GeneratedValue
    @RevisionNumber
    private int customId;

    @RevisionTimestamp
    private long customTimestamp;

    @OneToMany( mappedBy="revision", cascade={ CascadeType.PERSIST, CascadeType.REMOVE } )
    private Set<ModifiedEntityTypeEntity> modifiedEntityTypes = new HashSet<ModifiedEntityTypeEntity>();

    public void addModifiedEntityType( String entityClassName ) {
        modifiedEntityTypes.add( new ModifiedEntityTypeEntity( this, entityClassName ) );
    }

    ...
}

Example 388. ModifiedEntityTypeEntity.java

@Entity
public class ModifiedEntityTypeEntity {

    @Id
    @GeneratedValue
    private Integer id;

    @ManyToOne
    private CustomTrackingRevisionEntity revision;

    private String entityClassName;

    ...
}

CustomTrackingRevisionEntity revEntity =
    getAuditReader().findRevision( CustomTrackingRevisionEntity.class, revisionNumber );

Set<ModifiedEntityTypeEntity> modifiedEntityTypes = revEntity.getModifiedEntityTypes();

19.7. Tracking entity changes at property level

By default, the only information stored by Envers are revisions of modified entities. This approach lets user create audit queries based on historical values of entity properties. Sometimes it is useful to store additional metadata for each revision, when you are interested also in the type of changes, not only about the resulting values.

The feature described in Tracking entity names modified during revisions makes it possible to tell which entities were modified in a given revision.

The feature described here takes it one step further. "Modification Flags" enable Envers to track which properties of audited entities were modified in a given revision.

Tracking entity changes at property level can be enabled by:

1. setting org.hibernate.envers.global_with_modified_flag configuration property to true. This global switch will cause adding modification flags to be stored for all audited properties of all audited entities.

2. using @Audited( withModifiedFlag=true ) on a property or on an entity.

The trade-off coming with this functionality is an increased size of audit tables and a very little, almost negligible, performance drop during audit writes. This is due to the fact that every tracked property has to have an accompanying boolean column in the schema that stores information about the property modifications. Of course it is Envers job to fill these columns accordingly - no additional work by the developer is required. Because of costs mentioned, it is recommended to enable the feature selectively, when needed with use of the granular configuration means described above.

To see how "Modified Flags" can be utilized, check out the very simple query API that uses them: Querying for revisions of entity that modified given property.

19.8. Queries

You can think of historic data as having two dimensions:

• horizontal

• is the state of the database at a given revision. Thus, you can query for entities as they were at revision N.

• vertical

• are the revisions, at which entities changed. Hence, you can query for revisions, in which a given entity changed.

The queries in Envers are similar to Hibernate Criteria queries, so if you are common with them, using Envers queries will be much easier.

The main limitation of the current queries implementation is that you cannot traverse relations. You can only specify constraints on the ids of the related entities, and only on the "owning" side of the relation. This however will be changed in future releases.

Please note, that queries on the audited data will be in many cases much slower than corresponding queries on "live" data, as they involve correlated subselects.

Queries are improved both in terms of speed and possibilities, when using the valid-time audit strategy, that is when storing both start and end revisions for entities. See Configuration.

19.9. Querying for entities of a class at a given revision

The entry point for this type of queries is:

AuditQuery query = getAuditReader()
    .createQuery()
    .forEntitiesAtRevision( MyEntity.class, revisionNumber );

You can then specify constraints, which should be met by the entities returned, by adding restrictions, which can be obtained using the AuditEntity factory class. For example, to select only entities where the "name" property is equal to "John":

query.add( AuditEntity.property( "name" ).eq(  "John" ) );

And to select only entities that are related to a given entity:

query.add( AuditEntity.property( "address" ).eq( relatedEntityInstance ) );
// or
query.add( AuditEntity.relatedId( "address" ).eq( relatedEntityId ) );

You can limit the number of results, order them, and set aggregations and projections (except grouping) in the usual way. When your query is complete, you can obtain the results by calling the getSingleResult() or getResultList() methods.

A full query, can look for example like this:

List personsAtAddress = getAuditReader().createQuery()
    .forEntitiesAtRevision( Person.class, 12 )
    .addOrder( AuditEntity.property( "surname" ).desc() )
    .add( AuditEntity.relatedId( "address" ).eq( addressId ) )
    .setFirstResult( 4 )
    .setMaxResults( 2 )
    .getResultList();

19.10. Querying for revisions, at which entities of a given class changed

The entry point for this type of queries is:

AuditQuery query = getAuditReader().createQuery()
    .forRevisionsOfEntity( MyEntity.class, false, true );

You can add constraints to this query in the same way as to the previous one. There are some additional possibilities:

1. using AuditEntity.revisionNumber() you can specify constraints, projections and order on the revision number, in which the audited entity was modified

2. similarly, using AuditEntity.revisionProperty( propertyName ) you can specify constraints, projections and order on a property of the revision entity, corresponding to the revision in which the audited entity was modified

3. AuditEntity.revisionType() gives you access as above to the type of the revision (ADD, MOD, DEL).

Using these methods, you can order the query results by revision number, set projection or constraint the revision number to be greater or less than a specified value, etc. For example, the following query will select the smallest revision number, at which entity of class MyEntity with id entityId has changed, after revision number 42:

Number revision = (Number) getAuditReader().createQuery()
    .forRevisionsOfEntity( MyEntity.class, false, true )
    .setProjection( AuditEntity.revisionNumber().min() )
    .add( AuditEntity.id().eq( entityId ) )
    .add( AuditEntity.revisionNumber().gt( 42 ) )
    .getSingleResult();

The second additional feature you can use in queries for revisions is the ability to maximize/minimize a property. For example, if you want to select the smallest possibler revision at which the value of the actualDate for a given entity was larger then a given value:

Number revision = (Number) getAuditReader().createQuery()
    .forRevisionsOfEntity( MyEntity.class, false, true) // We are only interested in the first revision
    .setProjection( AuditEntity.revisionNumber().min() )
    .add( AuditEntity.property( "actualDate" ).minimize()
    .add( AuditEntity.property( "actualDate" ).ge( givenDate ) )
    .add( AuditEntity.id().eq( givenEntityId ) )) .getSingleResult();

The minimize() and maximize() methods return a criteria, to which you can add constraints, which must be met by the entities with the maximized/minimized properties.

You probably also noticed that there are two boolean parameters, passed when creating the query.

• selectEntitiesOnly

• the first parameter is only valid when you don’t set an explicit projection. If true, the result of the query will be a list of entities (which changed at revisions satisfying the specified constraints). If false, the result will be a list of three element arrays:

  • the first element will be the changed entity instance.

  • the second will be an entity containing revision data (if no custom entity is used, this will be an instance of DefaultRevisionEntity).

  • the third will be the type of the revision (one of the values of the RevisionType enumeration: ADD, MOD, DEL).

• selectDeletedEntities

• the second parameter specifies if revisions, in which the entity was deleted should be included in the results. If yes, such entities will have the revision type DEL and all fields, except the id, null.

19.11. Querying for revisions of entity that modified given property

For the two types of queries described above it’s possible to use special Audit criteria called hasChanged() and hasNotChanged() that makes use of the functionality described in Tracking entity changes at property level. They’re best suited for vertical queries, however existing API doesn’t restrict their usage for horizontal ones.

Let’s have a look at following examples:

AuditQuery query = getAuditReader().createQuery()
    .forRevisionsOfEntity( MyEntity.class, false, true )
    .add( AuditEntity.id().eq( id ) );
    .add( AuditEntity.property( "actualDate" ).hasChanged() );

This query will return all revisions of MyEntity with given id, where the actualDate property has been changed. Using this query we won’t get all other revisions in which actualDate wasn’t touched. Of course, nothing prevents user from combining hasChanged condition with some additional criteria - add method can be used here in a normal way.

AuditQuery query = getAuditReader().createQuery()
    .forEntitiesAtRevision( MyEntity.class, revisionNumber )
    .add( AuditEntity.property( "prop1" ).hasChanged() )
    .add( AuditEntity.property( "prop2" ).hasNotChanged() );

This query will return horizontal slice for MyEntity at the time revisionNumber was generated. It will be limited to revisions that modified prop1 but not prop2.

Note that the result set will usually also contain revisions with numbers lower than the revisionNumber, so wem cannot read this query as "Give me all MyEntities changed in revisionNumber with prop1 modified and prop2 untouched". To get such result we have to use the forEntitiesModifiedAtRevision query:

AuditQuery query = getAuditReader().createQuery()
    .forEntitiesModifiedAtRevision( MyEntity.class, revisionNumber )
    .add( AuditEntity.property( "prop1" ).hasChanged() )
    .add( AuditEntity.property( "prop2" ).hasNotChanged() );

19.12. Querying for entities modified in a given revision

The basic query allows retrieving entity names and corresponding Java classes changed in a specified revision:

modifiedEntityTypes = getAuditReader()
    .getCrossTypeRevisionChangesReader()
    .findEntityTypes( revisionNumber );

Other queries (also accessible from org.hibernate.envers.CrossTypeRevisionChangesReader):

• List<Object> findEntities( Number )

• Returns snapshots of all audited entities changed (added, updated and removed) in a given revision. Executes N+1 SQL queries, where N is a number of different entity classes modified within specified revision.

• List<Object> findEntities( Number, RevisionType )

• Returns snapshots of all audited entities changed (added, updated or removed) in a given revision filtered by modification type. Executes N+1 SQL queries, where N is a number of different entity classes modified within specified revision.

• Map<RevisionType, List&lt;Object&gt;> findEntitiesGroupByRevisionType( Number )

• Returns a map containing lists of entity snapshots grouped by modification operation (e.g. addition, update and removal). Executes 3N+1 SQL queries, where N is a number of different entity classes modified within specified revision.

Note that methods described above can be legally used only when the default mechanism of tracking changed entity names is enabled (see Tracking entity names modified during revisions).

19.13. Conditional auditing

Envers persists audit data in reaction to various Hibernate events (e.g. post update, post insert, and so on), using a series of event listeners from the org.hibernate.envers.event.spi package. By default, if the Envers jar is in the classpath, the event listeners are auto-registered with Hibernate.

Conditional auditing can be implemented by overriding some of the Envers event listeners. To use customized Envers event listeners, the following steps are needed:

1. Turn off automatic Envers event listeners registration by setting the hibernate.listeners.envers.autoRegister Hibernate property to false.

2. Create subclasses for appropriate event listeners. For example, if you want to conditionally audit entity insertions, extend the org.hibernate.envers.event.spi.EnversPostInsertEventListenerImpl class. Place the conditional-auditing logic in the subclasses, call the super method if auditing should be performed.

3. Create your own implementation of org.hibernate.integrator.spi.Integrator, similar to org.hibernate.envers.boot.internal.EnversIntegrator. Use your event listener classes instead of the default ones.

4. For the integrator to be automatically used when Hibernate starts up, you will need to add a META-INF/services/org.hibernate.integrator.spi.Integrator file to your jar. The file should contain the fully qualified name of the class implementing the interface.

19.14. Understanding the Envers Schema

For each audited entity (that is, for each entity containing at least one audited field), an audit table is created. By default, the audit table’s name is created by adding an "_AUD" suffix to the original table name, but this can be overridden by specifying a different suffix/prefix in the configuration properties or per-entity using the @org.hibernate.envers.AuditTable annotation.

The audit table contains the following columns:

• id

• id of the original entity (this can be more then one column in the case of composite primary keys)

• revision number

• an integer, which matches to the revision number in the revision entity table.

• revision type

• a small integer

• audited fields

• propertied from the original entity being audited

The primary key of the audit table is the combination of the original id of the entity and the revision number, so there can be at most one historic entry for a given entity instance at a given revision.

The current entity data is stored in the original table and in the audit table. This is a duplication of data, however as this solution makes the query system much more powerful, and as memory is cheap, hopefully this won’t be a major drawback for the users. A row in the audit table with entity id ID, revision N and data D means: entity with id ID has data D from revision N upwards. Hence, if we want to find an entity at revision M, we have to search for a row in the audit table, which has the revision number smaller or equal to M, but as large as possible. If no such row is found, or a row with a "deleted" marker is found, it means that the entity didn’t exist at that revision.

The "revision type" field can currently have three values: 0, 1 and 2, which means ADD, MOD and DEL, respectively. A row with a revision of type DEL will only contain the id of the entity and no data (all fields NULL), as it only serves as a marker saying "this entity was deleted at that revision".

Additionally, there is a revision entity table which contains the information about the global revision. By default the generated table is named REVINFO and contains just two columns: ID and TIMESTAMP. A row is inserted into this table on each new revision, that is, on each commit of a transaction, which changes audited data. The name of this table can be configured, the name of its columns as well as adding additional columns can be achieved as discussed in Revision Log.

19.15. Generating schema with Ant

If you like to generate the database schema file with the Hibernate Tools Ant task, you’ll probably notice that the generated file doesn’t contain definitions of audit tables. To generate the audit tables, you simply need to use org.hibernate.tool.ant.EnversHibernateToolTask, instead of the usual org.hibernate.tool.ant.HibernateToolTask. The former class extends the latter, and only adds generation of the version entities, meaning you can use the task just as you are used to.

For example:

<target name="schemaexport" depends="build-demo"
  description="Exports a generated schema to DB and file">
  <taskdef name="hibernatetool"
    classname="org.hibernate.tool.ant.EnversHibernateToolTask"
    classpathref="build.demo.classpath"/>

  <hibernatetool destdir=".">
    <classpath>
      <fileset refid="lib.hibernate" />
      <path location="${build.demo.dir}" />
      <path location="${build.main.dir}" />
    </classpath>
    <jpaconfiguration persistenceunit="ConsolePU" />
    <hbm2ddl
      drop="false"
      create="true"
      export="false"
      outputfilename="versioning-ddl.sql"
      delimiter=";"
      format="true"/>
  </hibernatetool>
</target>

Will generate the following schema:

create table Address (
    id integer generated by default as identity (start with 1),
    flatNumber integer,
    houseNumber integer,
    streetName varchar(255),
    primary key (id)
);

create table Address_AUD (
    id integer not null,
    REV integer not null,
    flatNumber integer,
    houseNumber integer,
    streetName varchar(255),
    REVTYPE tinyint,
    primary key (id, REV)
);

create table Person (
    id integer generated by default as identity (start with 1),
    name varchar(255),
    surname varchar(255),
    address_id integer,
    primary key (id)
);

create table Person_AUD (
    id integer not null,
    REV integer not null,
    name varchar(255),
    surname varchar(255),
    REVTYPE tinyint,
    address_id integer,
    primary key (id, REV)
);

create table REVINFO (
    REV integer generated by default as identity (start with 1),
    REVTSTMP bigint,
    primary key (REV)
);

alter table Person
    add constraint FK8E488775E4C3EA63
    foreign key (address_id)
    references Address;

19.16. Mapping exceptions

19.17. What isn’t and will not be supported

Bags are not supported because they can contain non-unique elements. Persisting, a bag of Strings violates the relational database principle that each table is a set of tuples.

In case of bags, however (which require a join table), if there is a duplicate element, the two tuples corresponding to the elements will be the same. Hibernate allows this, however Envers (or more precisely: the database connector) will throw an exception when trying to persist two identical elements, because of a unique constraint violation.

There are at least two ways out if you need bag semantics:

1. use an indexed collection, with the @javax.persistence.OrderColumn annotation

2. provide a unique id for your elements with the @CollectionId annotation.

19.18. What isn’t and will be supported

1. Bag style collections with a @CollectionId identifier column (see HHH-3950).

19.19. @OneToMany with @JoinColumn

When a collection is mapped using these two annotations, Hibernate doesn’t generate a join table. Envers, however, has to do this, so that when you read the revisions in which the related entity has changed, you don’t get false results.

To be able to name the additional join table, there is a special annotation: @AuditJoinTable, which has similar semantics to JPA @JoinTable.

One special case are relations mapped with @OneToMany with @JoinColumn on the one side, and @ManyToOne and @JoinColumn( insertable=false, updatable=false) on the many side. Such relations are in fact bidirectional, but the owning side is the collection.

To properly audit such relations with Envers, you can use the @AuditMappedBy annotation. It enables you to specify the reverse property (using the mappedBy element). In case of indexed collections, the index column must also be mapped in the referenced entity (using @Column( insertable=false, updatable=false ), and specified using positionMappedBy. This annotation will affect only the way Envers works. Please note that the annotation is experimental and may change in the future.

19.20. Advanced: Audit table partitioning

19.21. Benefits of audit table partitioning

Because audit tables tend to grow indefinitely, they can quickly become really large. When the audit tables have grown to a certain limit (varying per RDBMS and/or operating system) it makes sense to start using table partitioning. SQL table partitioning offers a lot of advantages including, but certainly not limited to:

1. Improved query performance by selectively moving rows to various partitions (or even purging old rows)

2. Faster data loads, index creation, etc.

19.22. Suitable columns for audit table partitioning

Generally SQL tables must be partitioned on a column that exists within the table. As a rule it makes sense to use either the end revision or the end revision timestamp column for partitioning of audit tables.

The reason why the end revision information should be used for audit table partitioning is based on the assumption that audit tables should be partitioned on an 'increasing level of relevancy', like so:

1. A couple of partitions with audit data that is not very (or no longer) relevant. This can be stored on slow media, and perhaps even be purged eventually.

2. Some partitions for audit data that is potentially relevant.

3. One partition for audit data that is most likely to be relevant. This should be stored on the fastest media, both for reading and writing.

19.23. Audit table partitioning example

In order to determine a suitable column for the 'increasing level of relevancy', consider a simplified example of a salary registration for an unnamed agency.

Currently, the salary table contains the following rows for a certain person X:

Table 9. Salaries table

The salary for the current fiscal year (2010) is unknown. The agency requires that all changes in registered salaries for a fiscal year are recorded (i.e. an audit trail). The rationale behind this is that decisions made at a certain date are based on the registered salary at that time. And at any time it must be possible reproduce the reason why a certain decision was made at a certain date.

The following audit information is available, sorted on in order of occurrence:

Table 10. Salaries - audit table

19.24. Determining a suitable partitioning column

To partition this data, the 'level of relevancy' must be defined. Consider the following:

1. For fiscal year 2006 there is only one revision. It has the oldest revision timestamp of all audit rows, but should still be regarded as relevant because it’s the latest modification for this fiscal year in the salary table (its end revision timestamp is null).

     Also note that it would be very unfortunate if in 2011 there would be an update of the salary for fiscal year 2006 (which is possible in until at least 10 years after the fiscal year),
     and the audit information would have been moved to a slow disk (based on the age of the __revision timestamp__).
     Remember that in this case Envers will have to update the _end revision timestamp_ of the most recent audit row.
   . There are two revisions in the salary of fiscal year 2007 which both have nearly the same _revision timestamp_ and a different __end revision timestamp__.
     On first sight, it is evident that the first revision was a mistake and probably not relevant.
     The only relevant revision for 2007 is the one with _end revision timestamp_ null.

Based on the above, it is evident that only the end revision timestamp is suitable for audit table partitioning. The revision timestamp is not suitable.

19.25. Determining a suitable partitioning scheme

A possible partitioning scheme for the salary table would be as follows:

• end revision timestamp year = 2008

• This partition contains audit data that is not very (or no longer) relevant.

• end revision timestamp year = 2009

• This partition contains audit data that is potentially relevant.

• end revision timestamp year >= 2010 or null

• This partition contains the most relevant audit data.

This partitioning scheme also covers the potential problem of the update of the end revision timestamp, which occurs if a row in the audited table is modified. Even though Envers will update the end revision timestamp of the audit row to the system date at the instant of modification, the audit row will remain in the same partition (the 'extension bucket').

And sometime in 2011, the last partition (or 'extension bucket') is split into two new partitions:

1. end revision timestamp year = 2010:: This partition contains audit data that is potentially relevant (in 2011).

2. end revision timestamp year >= 2011 or null:: This partition contains the most interesting audit data and is the new 'extension bucket'.

19.26. Envers links

1. Hibernate main page

2. Forum

3. JIRA issue tracker (when adding issues concerning Envers, be sure to select the "envers" component!)

4. IRC channel

5. FAQ