Package Specification

• Introduction
• The Entity Model
• A brief example
• Which API to use?
• Java 1.5 dependencies
  • Generic Types
  • Annotations
• Bytecode Enhancement

The Direct Persistence Layer (DPL) was designed to meet the following requirements.

• A type safe and convenient API is provided for accessing persistent objects. The use of Java generic types, although optional, is fully exploited to provide type safety. For example:

  {@literal PrimaryIndex employerById = ...;}
  long employerId = ...;
  Employer employer = employerById.get(employerId);

• All Java types are allowed to be persistent without requiring that they implement special interfaces. Persistent fields may be {@code private}, package-private (default access), {@code protected}, or {@code public}. No hand-coding of bindings is required. However, each persistent class must have a default constructor. For example:

  {@literal @Persistent}
  class Address {
      String street;
      String city;
      String state;
      int zipCode;
      private Address() {}
  }

• Bytecode enhancement provides fully optimized bindings that do not use Java reflection.
• It is easy to define primary and secondary keys. No external schema is required and Java annotations may be used for defining all metadata. Extensions may derive metadata from other sources. For example, the following Employer class is defined as a persistent entity with a primary key field {@code id} and the secondary key field {@code name}:

{@literal @Entity}
class Employer {

    {@literal @PrimaryKey(sequence="ID")}
    long id;

    {@literal @SecondaryKey(relate=ONE_TO_ONE)}
    String name;

    Address address;

    private Employer() {}
}

• Interoperability with external components is supported via the Java collections framework. Any primary or secondary index can be accessed using a standard java.util collection. For example:

  {@literal java.util.SortedMap map = employerByName.sortedMap();}

• Class evolution is explicitly supported. Compatible changes (adding fields and type widening) are performed automatically and transparently. For example, without any special configuration a {@code street2} field may be added to the {@code Address} class and the type of the {@code zipCode} field may be changed from {@code int} to {@code long}:

  {@literal @Persistent}
  class Address {
      String street;
      String street2;
      String city;
      String state;
      long zipCode;
      private Address() {}
  }

  Many incompatible class changes, such as renaming fields or refactoring a single class, can be performed using {@link com.sleepycat.persist.evolve.Mutations Mutations}. Mutations are automatically applied lazily as data is accessed, avoiding downtime to convert large databases during a software upgrade.

  Complex refactoring involving multiple classes may be performed using the a store conversion. The DPL always provides access to your data via a {@code RawStore}, no matter what changes have been made to persistent classes.



• The performance of the Berkeley DB transactional engine is not compromised. Operations are internally mapped directly to the engine API, object bindings are lightweight, and all engine tuning parameters are available. For example, a "dirty read" may be performed using an optional {@link com.sleepycat.db.LockMode LockMode} parameter:

  Employer employer = employerByName.get(null, "Gizmo Inc", LockMode.READ_UNCOMMITTED);

  For high performance applications, {@link com.sleepycat.db.DatabaseConfig DatabaseConfig} parameters may be used to tune the performance of the Berkeley DB engine. For example, the size of an internal Btree node can be specified as follows:

  DatabaseConfig config = store.getPrimaryConfig(Employer.class);
  config.setNodeMaxEntries(64);
  store.setPrimaryConfig(config);

The DPL is intended for applications that represent persistent domain objects using Java classes. An entity class is an ordinary Java class that has a primary key and is stored and accessed using a primary index. It may also have any number of secondary keys, and entities may be accessed by secondary key using a secondary index.

An entity class may be defined with the {@link com.sleepycat.persist.model.Entity Entity} annotation. For each entity class, its primary key may be defined using the {@link com.sleepycat.persist.model.PrimaryKey PrimaryKey} annotation and any number of secondary keys may be defined using the {@link com.sleepycat.persist.model.SecondaryKey SecondaryKey} annotation.

In the following example, the {@code Person.ssn} (social security number) field is the primary key and the {@code Person.employerIds} field is a many-to-many secondary key.

{@literal @Entity}
class Person {

    {@literal @PrimaryKey}
    String ssn;

    String name;
    Address address;

    {@literal @SecondaryKey(relate=MANY_TO_MANY, relatedEntity=Employer.class)}
    {@literal Set employerIds = new HashSet();}

    private Person() {} // For bindings
}

A set of entity classes constitutes an entity model. In addition to isolated entity classes, an entity model may contain relationships between entities. Relationships may be defined using the {@link com.sleepycat.persist.model.SecondaryKey SecondaryKey} annotation. Many-to-one, one-to-many, many-to-many and one-to-one relationships are supported, as well as foreign key constraints.

In the example above, a relationship between the {@code Person} and {@code Employer} entities is defined via the {@code Person.employerIds} field. The {@code relatedEntity=Employer.class} annotation property establishes foreign key constraints to guarantee that every element of the {@code employerIds} set is a valid {@code Employer} primary key.

For more information on the entity model, see the {@link com.sleepycat.persist.model.AnnotationModel AnnotationModel} and the {@link com.sleepycat.persist.model.Entity Entity} annotation.

The root object in the DPL is the {@link com.sleepycat.persist.EntityStore EntityStore}. An entity store manages any number of objects for each entity class defined in the model. The store provides access to the primary and secondary indices for each entity class, for example:

EntityStore store = new EntityStore(...);

{@literal PrimaryIndex personBySsn} =
    store.getPrimaryIndex(String.class, Person.class);

The following example shows how to define an entity model and how to store and access persistent objects. Exception handling is omitted for brevity.

import java.io.File;
import java.util.HashSet;
import java.util.Set;

import com.sleepycat.db.DatabaseException;
import com.sleepycat.db.Environment;
import com.sleepycat.db.EnvironmentConfig;
import com.sleepycat.persist.EntityCursor;
import com.sleepycat.persist.EntityIndex;
import com.sleepycat.persist.EntityStore;
import com.sleepycat.persist.PrimaryIndex;
import com.sleepycat.persist.SecondaryIndex;
import com.sleepycat.persist.StoreConfig;
import com.sleepycat.persist.model.Entity;
import com.sleepycat.persist.model.Persistent;
import com.sleepycat.persist.model.PrimaryKey;
import com.sleepycat.persist.model.SecondaryKey;
import static com.sleepycat.persist.model.DeleteAction.NULLIFY;
import static com.sleepycat.persist.model.Relationship.ONE_TO_ONE;
import static com.sleepycat.persist.model.Relationship.ONE_TO_MANY;
import static com.sleepycat.persist.model.Relationship.MANY_TO_ONE;
import static com.sleepycat.persist.model.Relationship.MANY_TO_MANY;

// An entity class.
//
{@literal @Entity}
class Person {

    {@literal @PrimaryKey}
    String ssn;

    String name;
    Address address;

    {@literal @SecondaryKey(relate=MANY_TO_ONE, relatedEntity=Person.class)}
    String parentSsn;

    {@literal @SecondaryKey(relate=ONE_TO_MANY)}
    {@literal Set emailAddresses = new HashSet();}

    {@code @SecondaryKey(relate=MANY_TO_MANY, relatedEntity=Employer.class,
                                       onRelatedEntityDelete=NULLIFY)}
    {@code Set employerIds = new HashSet();}

    Person(String name, String ssn, String parentSsn) {
        this.name = name;
        this.ssn = ssn;
        this.parentSsn = parentSsn;
    }

    private Person() {} // For bindings
}

// Another entity class.
//
{@literal @Entity}
class Employer {

    {@literal @PrimaryKey(sequence="ID")}
    long id;

    {@literal @SecondaryKey(relate=ONE_TO_ONE)}
    String name;

    Address address;

    Employer(String name) {
        this.name = name;
    }

    private Employer() {} // For bindings
}

// A persistent class used in other classes.
//
{@literal @Persistent}
class Address {
    String street;
    String city;
    String state;
    int zipCode;
    private Address() {} // For bindings
}

// The data accessor class for the entity model.
//
class PersonAccessor {

    // Person accessors
    //
    {@literal PrimaryIndex personBySsn;}
    {@literal SecondaryIndex personByParentSsn;}
    {@literal SecondaryIndex personByEmailAddresses;}
    {@literal SecondaryIndex personByEmployerIds;}

    // Employer accessors
    //
    {@literal PrimaryIndex employerById;}
    {@literal SecondaryIndex employerByName;}

    // Opens all primary and secondary indices.
    //
    public PersonAccessor(EntityStore store)
        throws DatabaseException {

        personBySsn = store.getPrimaryIndex(
            String.class, Person.class);

        personByParentSsn = store.getSecondaryIndex(
            personBySsn, String.class, "parentSsn");

        personByEmailAddresses = store.getSecondaryIndex(
            personBySsn, String.class, "emailAddresses");

        personByEmployerIds = store.getSecondaryIndex(
            personBySsn, Long.class, "employerIds");

        employerById = store.getPrimaryIndex(
            Long.class, Employer.class);

        employerByName = store.getSecondaryIndex(
            employerById, String.class, "name"); 
    }
}

// Open a transactional Berkeley DB engine environment.
//
EnvironmentConfig envConfig = new EnvironmentConfig();
envConfig.setAllowCreate(true);
envConfig.setTransactional(true);
Environment env = new Environment(new File("/my/data"), envConfig);

// Open a transactional entity store.
//
StoreConfig storeConfig = new StoreConfig();
storeConfig.setAllowCreate(true);
storeConfig.setTransactional(true);
EntityStore store = new EntityStore(env, "PersonStore", storeConfig);

// Initialize the data access object.
//
PersonAccessor dao = new PersonAccessor(store);

// Add a parent and two children using the Person primary index.  Specifying a
// non-null parentSsn adds the child Person to the sub-index of children for
// that parent key.
//
dao.personBySsn.put(new Person("Bob Smith", "111-11-1111", null));
dao.personBySsn.put(new Person("Mary Smith", "333-33-3333", "111-11-1111"));
dao.personBySsn.put(new Person("Jack Smith", "222-22-2222", "111-11-1111"));

// Print the children of a parent using a sub-index and a cursor.
//
{@literal EntityCursor children =}
    dao.personByParentSsn.subIndex("111-11-1111").entities();
try {
    for (Person child : children) {
        System.out.println(child.ssn + ' ' + child.name);
    }
} finally {
    children.close();
}

// Get Bob by primary key using the primary index.
//
Person bob = dao.personBySsn.get("111-11-1111");
assert bob != null;

// Create two employers.  Their primary keys are assigned from a sequence.
//
Employer gizmoInc = new Employer("Gizmo Inc");
Employer gadgetInc = new Employer("Gadget Inc");
dao.employerById.put(gizmoInc);
dao.employerById.put(gadgetInc);

// Bob has two jobs and two email addresses.
//
bob.employerIds.add(gizmoInc.id);
bob.employerIds.add(gadgetInc.id);
bob.emailAddresses.add("bob@bob.com");
bob.emailAddresses.add("bob@gmail.com");

// Update Bob's record.
//
dao.personBySsn.put(bob);

// Bob can now be found by both email addresses.
//
bob = dao.personByEmailAddresses.get("bob@bob.com");
assert bob != null;
bob = dao.personByEmailAddresses.get("bob@gmail.com");
assert bob != null;

// Bob can also be found as an employee of both employers.
//
{@literal EntityIndex employees;}
employees = dao.personByEmployerIds.subIndex(gizmoInc.id);
assert employees.contains("111-11-1111");
employees = dao.personByEmployerIds.subIndex(gadgetInc.id);
assert employees.contains("111-11-1111");

// When an employer is deleted, the onRelatedEntityDelete=NULLIFY for the
// employerIds key causes the deleted ID to be removed from Bob's employerIds.
//
dao.employerById.delete(gizmoInc.id);
bob = dao.personBySsn.get("111-11-1111");
assert !bob.employerIds.contains(gizmoInc.id);

store.close();
env.close();

The example illustrates several characteristics of the DPL:

• Persistent data and keys are defined in terms of instance fields. For brevity the example does not show getter and setter methods, although these would normally exist to provide encapsulation. The DPL accesses fields during object serialization and deserialization, rather than calling getter/setter methods, leaving business methods free to enforce arbitrary validation rules. For example:

  {@literal @Persistent}
  public class ConstrainedValue {
  
      private int min;
      private int max;
      private int value;
  
      private ConstrainedValue() {} // For bindings
  
      public ConstrainedValue(int min, int max) {
          this.min = min;
          this.max = max;
          value = min;
      }
  
      public setValue(int value) {
          if (value < min || value > max) {
              throw new IllegalArgumentException("out of range");
          }
          this.value = value;
      }
  }
  

  The above {@code setValue} method would not work if it were called during object deserialization, since the order of setting fields is arbitrary. The {@code min} and {@code max} fields may not be set before the {@code value} is set.


• The example creates a transactional store and therefore all operations are transaction protected. Because no explicit transactions are used, auto-commit is used implicitly.

  Explicit transactions may also be used to group multiple operations in a single transaction, and all access methods have optional transaction parameters. For example, the following two operations are performed atomically in a transaction:

  Transaction txn = env.beginTransaction(null, null);
  dao.employerById.put(txn, gizmoInc);
  dao.employerById.put(txn, gadgetInc);
  txn.commit();
  

• To provide maximum performance, the DPL operations map directly to the Btree operations of the Berkeley DB engine. Unlike other persistence approaches, keys and indices are exposed for direct access and performance tuning.

  Queries are implemented by calling methods of the primary and secondary indices. An {@link com.sleepycat.persist.EntityJoin EntityJoin} class is also available for performing equality joins. For example, the following code queries all of Bob's children that work for Gizmo Inc:

  {@literal EntityJoin join = new EntityJoin(dao.personBySsn);}
  
  join.addCondition(dao.personByParentSsn, "111-11-1111");
  join.addCondition(dao.personByEmployerIds, gizmoInc.id);
  
  {@literal ForwardCursor results = join.entities();}
  try {
      for (Person person : results) {
          System.out.println(person.ssn + ' ' + person.name);
      }
  } finally {
      results.close();
  }
  

• Object relationships are based on keys. When a {@code Person} with a given employer ID in its {@code employerIds} set is stored, the {@code Person} object becomes part of the collection of employees for that employer. This collection of employees is accessed using a {@link com.sleepycat.persist.SecondaryIndex#subIndex SecondaryIndex.subIndex} for the employer ID, as shown below:

  {@literal EntityCursor employees =}
      dao.personByEmployerIds.subIndex(gizmoInc.id).entities();
  try {
      for (Person employee : employees) {
          System.out.println(employee.ssn + ' ' + employee.name);
      }
  } finally {
      employees.close();
  }
  

• Note that when Bob's employer is deleted in the example, the {@code Person} object for Bob is refetched to see the change to its {@code employerIds}. This is because objects are accessed by value, not by reference. In other words, no object cache or "persistence context" is maintained by the DPL. The low level caching of the embedded Berkeley DB engine, combined with lightweight object bindings, provides maximum performance.

The Berkeley DB engine has a {@link com.sleepycat.db Base API}, a {@link com.sleepycat.collections Collections API} and a {@link com.sleepycat.persist Direct Persistence Layer (DPL)}. Follow these guidelines if you are not sure which API to use:

• When Java classes are used to represent domain objects in an application, the DPL is recommended. The more domain classes, the more value there is in using annotations to define your schema.


• When porting an application between Berkeley DB and Berkeley DB Java Edition, or when you've chosen not to use Java classes to represent domain objects, then the Base API is recommended. You may also prefer to use this API if you have very few domain classes.


• The Collections API is useful for interoperating with external components because it conforms to the standard Java Collections Framework. It is therefore useful in combination with both the Base API and the DPL. You may prefer this API because it provides the familiar Java Collections interface.

The DPL uses two features of Java 1.5: generic types and annotations. If you wish to avoid using these two Java 1.5 features, the DPL provides options for doing so.

Generic types are used to provide type safety, especially for the {@link com.sleepycat.persist.PrimaryIndex PrimaryIndex}, {@link com.sleepycat.persist.SecondaryIndex SecondaryIndex}, and {@link com.sleepycat.persist.EntityCursor EntityCursor} classes. If you don't wish to use generic types, you can simply not declare your index and cursor objects using generic type parameters. This is the same as using the Java 1.5 Collections Framework without using generic types.

If you don't wish to use annotations, you can provide another source of metadata by implementing an {@link com.sleepycat.persist.model.EntityModel EntityModel} class. For example, naming conventions, static members, or an XML configuration file might be used as a source of metadata. However, if you don't use annotations then you won't be able to use bytecode enhancement, which is described next.

The persistent fields of a class may be private, package-private, protected or public. The DPL can access persistent fields either by bytecode enhancement or by reflection.

Bytecode enhancement may be used to fully optimize binding performance and to avoid the use of Java reflection. In applications that are CPU bound, avoiding Java reflection can have a significant performance impact.

Bytecode enhancement may be performed either at runtime or at build time (offline). When enhancement is performed at runtime, persistent classes are enhanced as they are loaded. When enhancement is performed offline, class files are enhanced during a post-compilation step. Enhanced classes are used to efficiently access all fields and default constructors, including non-public members.

See {@link com.sleepycat.persist.model.ClassEnhancer ClassEnhancer} for bytecode enhancement configuration details.

If bytecode enhancement is not used as described above, the DPL will use reflection for accessing persistent fields and the default constructor. The {@link java.lang.reflect.AccessibleObject#setAccessible AccessibleObject.setAccessible} method is called by the DPL to enable access to non-public fields and constructors. If you are running under a Java security manager you must configure your security policy to allow the following permission:

{@code permission java.lang.reflect.ReflectPermission "suppressAccessChecks";}

There are three cases where setting the above permission is not required:

1. If you are not running under a Java Security Manager, then access to non-public members via reflection is not restricted. This is the default for J2SE.


2. If all persistent fields and default constructors are {@code public} then they can be accessed via reflection without special permissions, even when running under a Java Security Manager. However, declaring {@code public} instance fields is not recommended because it discourages encapsulation.


3. If bytecode enhancement is used as described above, then reflection will not be used.

It is well known that executing generated code is faster than reflection. However, this performance difference may or may not impact a given application since it may be overshadowed by other factors. Performance testing in a realistic usage scenario is the best way to determine the impact. If you are determined to avoid the use of reflection then option 3 above is recommended.