9 Using Container-Managed Persistence

Mapping Capabilities

Mapping refers to the ability to tie an object-based model to a relational model of data, usually the schema of a relational database. The CMP implementation provides the ability to tie a set of interrelated beans containing data and associated behaviors to the schema. This object representation of the database becomes part of the Java application. You can also customize this mapping to optimize these beans for the particular needs of an application. The result is a single data model through which both persistent database information and regular transient program data are accessed.

The mapping capabilities provided by the GlassFish Server include:

• Mapping a CMP bean to one or more tables

• Mapping CMP fields to one or more columns

• Mapping CMP fields to different column types

• Mapping tables with compound primary keys

• Mapping tables with unknown primary keys

• Mapping CMP relationships to foreign keys

• Mapping tables with overlapping primary and foreign keys

The Mapping Deployment Descriptor File

Each module with CMP beans must have the following files:

• ejb-jar.xml - The J2EE standard file for assembling enterprise beans. For a detailed description, see the Enterprise JavaBeans Specification, v2.1.

• glassfish-ejb-jar.xml - The GlassFish Server standard file for assembling enterprise beans. For a detailed description, see "The glassfish-ejb-jar.xml File" in Oracle GlassFish Server Application Deployment Guide.

• sun-cmp-mappings.xml - The mapping deployment descriptor file, which describes the mapping of CMP beans to tables in a database. For a detailed description, see "The sun-cmp-mappings.xml File" in Oracle GlassFish Server Application Deployment Guide.

The sun-cmp-mappings.xml file can be automatically generated and does not have to exist prior to deployment. For details, see Generation Options for CMP.

The sun-cmp-mappings.xml file maps CMP fields and CMR fields (relationships) to the database. A primary table must be selected for each CMP bean, and optionally, multiple secondary tables. CMP fields are mapped to columns in either the primary or secondary table(s). CMR fields are mapped to pairs of column lists (normally, column lists are the lists of columns associated with primary and foreign keys).

The sun-cmp-mappings.xml file conforms to the sun-cmp-mapping_1_2.dtd file and is packaged with the user-defined bean classes in the EJB JAR file under the META-INF directory.

The GlassFish Server creates the mappings in the sun-cmp-mappings.xml file automatically during deployment if the file is not present.

To map the fields and relationships of your entity beans manually, edit the sun-cmp-mappings.xml deployment descriptor. Only do this if you are proficient in editing XML.

The mapping information is developed in conjunction with the database schema (.dbschema) file, which can be automatically captured when you deploy the bean (see Automatic Database Schema Capture). You can manually generate the schema using the capture-schema utility (Using the capture-schema Utility).

Mapping Considerations

The following topics are addressed here:

• Join Tables and Relationships

• Automatic Primary Key Generation

• Fixed Length CHAR Primary Keys

• Managed Fields

• BLOB Support

• CLOB Support

The data types used in automatic schema generation are also suggested for manual mapping. These data types are described in Supported Data Types for CMP.

<schema-generator-properties>
   <property>
      <name>Employee.voiceGreeting.jdbc-type</name>
      <value>BLOB</value>
   </property>
   <property>
      <name>Employee.voiceGreeting.jdbc-maximum-length</name>
      <value>10240</value>
   </property>
   ...
</schema-generator-properties>

<schema-generator-properties>
   <property>
      <name>Employee.resume.jdbc-type</name>
      <value>CLOB</value>
   </property>
   <property>
      <name>Employee.resume.jdbc-maximum-length</name>
      <value>10240</value>
   </property>
   ...
</schema-generator-properties>

Supported Data Types for CMP

CMP supports a set of JDBC data types that are used in mapping Java data fields to SQL types. Supported JDBC data types are as follows: BIGINT, BIT, BLOB, CHAR, CLOB, DATE, DECIMAL, DOUBLE, FLOAT, INTEGER, NUMERIC, REAL, SMALLINT, TIME, TIMESTAMP, TINYINT, VARCHAR.

The following table contains the mappings of Java types to JDBC types when automatic mapping is used.

The following table contains the mappings of JDBC types to database vendor-specific types when automatic mapping is used. For a list of the JDBC drivers currently supported by the GlassFish Server, see the Oracle GlassFish Server Release Notes. For configurations of supported and other drivers, see "Configuration Specifics for JDBC Drivers" in Oracle GlassFish Server Administration Guide.

Generation Options for CMP

Deployment descriptor elements or asadmin command line options can control automatic schema generation by the following:

• Creating tables during deployment

• Dropping tables during undeployment

• Dropping and creating tables during redeployment

• Specifying the database vendor

• Specifying that table names are unique

• Specifying type mappings for individual CMP fields

The following optional data subelements of the cmp-resource element in the glassfish-ejb-jar.xml file control the automatic creation of database tables at deployment. For more information about the cmp-resource element, see "cmp-resource" in Oracle GlassFish Server Application Deployment Guide and Configuring the CMP Resource.

The following options of the asadmin deploy or asadmin deploydir command control the automatic creation of database tables at deployment.

If one or more of the beans in the module are manually mapped and you use any of the asadmin deploy or asadmin deploydir options, the deployment is not harmed in any way, but the options have no effect, and a warning is written to the server log.

The following options of the asadmin undeploy command control the automatic removal of database tables at undeployment.

For more information about the asadmin deploy, asadmin deploydir, and asadmin undeploy commands, see the Oracle GlassFish Server Reference Manual.

When command line and glassfish-ejb-jar.xml options are both specified, the asadmin options take precedence.

Automatic Database Schema Capture

You can configure a CMP bean in GlassFish Server to automatically capture the database metadata and save it in a .dbschema file during deployment. If the sun-cmp-mappings.xml file contains an empty <schema/> entry, the cmp-resource entry in the glassfish-ejb-jar.xml file is used to get a connection to the database, and automatic generation of the schema is performed.

Using the capture-schema Utility

You can use the capture-schema command to manually generate the database metadata (.dbschema) file. For details, see the Oracle GlassFish Server Reference Manual.

The capture-schema utility does not modify the schema in any way. Its only purpose is to provide the persistence engine with information about the structure of the database (the schema).

Keep the following in mind when using the capture-schema command:

• The name of a .dbschema file must be unique across all deployed modules in a domain.

• If more than one schema is accessible for the schema user, more than one table with the same name might be captured if the -schemaname option of capture-schema is not set.

• The schema name must be upper case.

• Table names in databases are case-sensitive. Make sure that the table name matches the name in the database.

• PostgreSQL databases internally convert all names to lower case. Before running the capture-schema command on a PostgreSQL database, make sure table and column names are lower case in the sun-cmp-mappings.xml file.

• An Oracle database user running the capture-schema command needs ANALYZE ANY TABLE privileges if that user does not own the schema. These privileges are granted to the user by the database administrator.

Version Column Consistency Checking

The version consistency feature saves the bean state at first transactional access and caches it between transactions. The state is copied from the cache instead of being read from the database. The bean state is verified by primary key and version column values at flush for custom queries (for dirty instances only) and at commit (for clean and dirty instances).

Relationship Prefetching

In many cases when an entity bean's state is fetched from the database, its relationship fields are always accessed in the same transaction. Relationship prefetching saves database round trips by fetching data for an entity bean and those beans referenced by its CMR fields in a single database round trip.

To enable relationship prefetching for a CMR field, use the default subelement of the fetched-with element in the sun-cmp-mappings.xml file. By default, these CMR fields are prefetched whenever findByPrimaryKey or a custom finder is executed for the entity, or when the entity is navigated to from a relationship. (Recursive prefetching is not supported, because it does not usually enhance performance.) See "fetched-with" in Oracle GlassFish Server Application Deployment Guide.

To disable prefetching for specific custom finders, use the prefetch-disabled element in the glassfish-ejb-jar.xml file. See "prefetch-disabled" in Oracle GlassFish Server Application Deployment Guide.

Multilevel relationship prefetching is supported for CMP 2.1 entity beans. To enable multilevel relationship prefetching, set the following property using the asadmin create-jvm-options command:

asadmin create-jvm-options -Dcom.sun.jdo.spi.persistence.support.sqlstore.MULTILEVEL_PREFETCH=true

Read-Only Beans

Another feature that the GlassFish Server provides is the read-only bean, an entity bean that is never modified by an EJB client. Read-only beans avoid database updates completely.

A read-only bean can be used to cache a database entry that is frequently accessed but rarely updated (externally by other beans). When the data that is cached by a read-only bean is updated by another bean, the read-only bean can be notified to refresh its cached data.

The GlassFish Server provides a number of ways by which a read-only bean's state can be refreshed. By setting the refresh-period-in-seconds element in the glassfish-ejb-jar.xml file and the trans-attribute element (or @TransactionAttribute annotation) in the ejb-jar.xml file, it is easy to configure a read-only bean that is one of the following:

• Always refreshed

• Periodically refreshed

• Never refreshed

• Programmatically refreshed

Access to CMR fields of read-only beans is not supported. Deployment will succeed, but an exception will be thrown at runtime if a get or set method is invoked.

Read-only beans are best suited for situations where the underlying data never changes, or changes infrequently. For further information and usage guidelines, see Using Read-Only Beans.

About JDOQL Queries

The Enterprise JavaBeans Specification, v1.1 does not specify the format of the finder method description. The GlassFish Server uses an extension of Java Data Objects Query Language (JDOQL) queries to implement finder and selector methods. You can specify the following elements of the underlying JDOQL query:

• Filter expression - A Java-like expression that specifies a condition that each object returned by the query must satisfy. Corresponds to the WHERE clause in EJB QL.

• Query parameter declaration - Specifies the name and the type of one or more query input parameters. Follows the syntax for formal parameters in the Java language.

• Query variable declaration - Specifies the name and type of one or more query variables. Follows the syntax for local variables in the Java language. A query filter might use query variables to implement joins.

• Query ordering declaration - Specifies the ordering expression of the query. Corresponds to the ORDER BY clause of EJB QL.

The GlassFish Server specific deployment descriptor (glassfish-ejb-jar.xml) provides the following elements to store the EJB 1.1 finder method settings:

query-filter
query-params
query-variables
query-ordering

The bean developer uses these elements to construct a query. When the finder method that uses these elements executes, the values of these elements are used to execute a query in the database. The objects from the JDOQL query result set are converted into primary key instances to be returned by the EJB 1.1 ejbFind method.

The JDO specification, JSR 12, provides a comprehensive description of JDOQL. The following information summarizes the elements used to define EJB 1.1 finders.

Query Filter Expression

The filter expression is a String containing a Boolean expression evaluated for each instance of the candidate class. If the filter is not specified, it defaults to true. Rules for constructing valid expressions follow the Java language, with the following differences:

• Equality and ordering comparisons between primitives and instances of wrapper classes are valid.

• Equality and ordering comparisons of Date fields and Date parameters are valid.

• Equality and ordering comparisons of String fields and String parameters are valid.

• White space (non-printing characters space, tab, carriage return, and line feed) is a separator and is otherwise ignored.

• The following assignment operators are not supported.

  • Comparison operators such as =, +=, and so on

  • Pre- and post-increment

  • Pre- and post-decrement

• Methods, including object construction, are not supported, except for these methods.

  Collection.contains(Object o)
  Collection.isEmpty()
  String.startsWith(String s)
  String.endsWith(String e)
  

  In addition, the GlassFish Server supports the following nonstandard JDOQL methods.

  String.like(String pattern)
  String.like(String pattern, char escape)
  String.substring(int start, int length)
  String.indexOf(String str)
  String.indexOf(String str, int start)
  String.length()
  Math.abs(numeric n)
  Math.sqrt(double d)
  

• Navigation through a null-valued field, which throws a NullPointerException, is treated as if the sub-expression returned false.

The following expressions are supported.

• Relational operators (==, !=,>, <,>=, <=)

• Boolean operators (&, &&, |, ||, ~, !)

• Arithmetic operators (+, -, *, /)

• String concatenation, only for String + String

• Parentheses to explicitly mark operator precedence

• Cast operator

• Promotion of numeric operands for comparisons and arithmetic operations

The rules for promotion follow the Java rules extended by BigDecimal, BigInteger, and numeric wrapper classes. See the numeric promotions of the Java language specification.

Query Parameters

The parameter declaration is a String containing one or more parameter type declarations separated by commas. This follows the Java syntax for method signatures.

Query Variables

The type declarations follow the Java syntax for local variable declarations.

JDOQL Examples

This section provides a few query examples.

name == "Michael"

<finder>
   <method-name>findPlayerByName</method-name>
   <query-filter>name == "Michael"</query-filter>
</finder>

low < price && price < high

<finder>
   <method-name>findInRange</method-name>
   <query-params>double low, double high</query-params>
   <query-filter>low &lt; price &amp;&amp; price &lt high</query-filter>
   <query-ordering>price ascending</query-ordering>
</finder>

mypackage.PlayerEJB_170160966_JDOState player

(this.salary> player.salary) && (player.name == name)

<finder>
   <method-name>findByHigherSalary</method-name>
   <query-params>java.lang.String name</query-params>
   <query-filter>
      (this.salary &gt; player.salary) &amp;&amp; (player.name == name)
   </query-filter>
   <query-variables>
      mypackage.PlayerEJB_170160966_JDOState player
   </query-variables>
</finder>

Disabling ORDER BY Validation

EJB QL as defined in the EJB 2.1 Specification defines certain restrictions for the SELECT clause of an ORDER BY query (see section 11.2.8 ORDER BY Clause). This ensures that a query does not order by a field that is not returned by the query. By default, the EJB QL compiler checks the above restriction and throws an exception if the query does not conform.

However, some databases support SQL statements with an ORDER BY column that is not included in the SELECT clause. To disable the validation of the ORDER BY clause against the SELECT clause, set the DISABLE_ORDERBY_VALIDATION JVM option as follows:

asadmin create-jvm-options 
-Dcom.sun.jdo.spi.persistence.support.ejb.ejbqlc.DISABLE_ORDERBY_VALIDATION=true

The DISABLE_ORDERBY_VALIDATION option is set to false by default. Setting it to true results in a non-portable module or application.

Setting the Heap Size on DB2

On DB2, the database configuration parameter APPLHEAPSZ determines the heap size. If you are using the Oracle or DataDirect database driver, set this parameter to at least 2048 for CMP. For more information, see http://publib.boulder.ibm.com/infocenter/db2luw/v8/index.jsp?topic=/com.ibm.db2.udb.doc/opt/tsbp2024.htm.

Eager Loading of Field State

By default, the EJB container loads the state for all persistent fields (excluding relationship, BLOB, and CLOB fields) before invoking the ejbLoad method of the abstract bean. This approach might not be optimal for entity objects with large state if most business methods require access to only parts of the state.

Use the fetched-with element in sun-cmp-mappings.xml for fields that are used infrequently. See "fetched-with" in Oracle GlassFish Server Application Deployment Guide.

Restrictions on Remote Interfaces

The following restrictions apply to the remote interface of an EJB 2.1 bean that uses CMP:

• Do not expose the get and set methods for CMR fields or the persistence collection classes that are used in container-managed relationships through the remote interface of the bean.

  However, you are free to expose the get and set methods that correspond to the CMP fields of the entity bean through the bean's remote interface.

• Do not expose the container-managed collection classes that are used for relationships through the remote interface of the bean.

• Do not expose local interface types or local home interface types through the remote interface or remote home interface of the bean.

Dependent value classes can be exposed in the remote interface or remote home interface, and can be included in the client EJB JAR file.

PostgreSQL Case Insensitivity

Case-sensitive behavior cannot be achieved for PostgreSQL databases. PostgreSQL databases internally convert all names to lower case, which makes the following workarounds necessary:

• In the CMP 2.1 runtime, PostgreSQL table and column names are not quoted, which makes these names case insensitive.

• Before running the capture-schema command on a PostgreSQL database, make sure table and column names are lower case in the sun-cmp-mappings.xml file.

No Support for lock-when-loaded on Sybase

For EJB 2.1 beans, the lock-when-loaded consistency level is implemented by placing update locks on the data corresponding to a bean when the data is loaded from the database. There is no suitable mechanism available on Sybase databases to implement this feature. Therefore, the lock-when-loaded consistency level is not supported on Sybase databases. See "consistency" in Oracle GlassFish Server Application Deployment Guide.

Sybase Finder Limitation

If a finder method with an input greater than 255 characters is executed and the primary key column is mapped to a VARCHAR column, Sybase attempts to convert type VARCHAR to type TEXT and generates the following error:

com.sybase.jdbc2.jdbc.SybSQLException: Implicit conversion from datatype 
'TEXT' to 'VARCHAR' is not allowed. Use the CONVERT function to run this query. 

To avoid this error, make sure the finder method input is less than 255 characters.

Date and Time Fields

If a field type is a Java date or time type (java.util.Date, java.sql.Date, java.sql.Time, java.sql.Timestamp), make sure that the field value exactly matches the value in the database.

For example, the following code uses a java.sql.Date type as a primary key field:

java.sql.Date myDate = new java.sql.Date(System.currentTimeMillis())
BeanA.create(myDate, ...);

For some databases, this code results in only the year, month, and date portion of the field value being stored in the database. Later if the client tries to find this bean by primary key as follows, the bean is not found in the database because the value does not match the one that is stored in the database.

myBean = BeanA.findByPrimaryKey(myDate);

Similar problems can happen if the database truncates the timestamp value while storing it, or if a custom query has a date or time value comparison in its WHERE clause.

For automatic mapping to an Oracle database, fields of type java.util.Date, java.sql.Date, and java.sql.Time are mapped to Oracle's DATE data type. Fields of type java.sql.Timestamp are mapped to Oracle's TIMESTAMP(9) data type.

Set RECURSIVE_TRIGGERS to false on MSSQL

For version consistency triggers on MSSQL, the property RECURSIVE_TRIGGERS must be set to false, which is the default. If set to true, triggers throw a java.sql.SQLException.

Set this property as follows:

EXEC sp_dboption 'database-name', 'recursive triggers', 'FALSE'
go

You can test this property as follows:

SELECT DATABASEPROPERTYEX('database-name', 'IsRecursiveTriggersEnabled')
go

MySQL Database Restrictions

The following restrictions apply when you use a MySQL database with the GlassFish Server for persistence.

• MySQL treats int1 and int2 as reserved words. If you want to define int1 and int2 as fields in your table, use `int1` and `int2` field names in your SQL file.

• When VARCHAR fields get truncated, a warning is displayed instead of an error. To get an error message, start the MySQL database in strict SQL mode.

• The order of fields in a foreign key index must match the order in the explicitly created index on the primary table.

• The CREATE TABLE syntax in the SQL file must end with the following line.

  )  Engine=InnoDB;
  

  InnoDB provides MySQL with a transaction-safe (ACID compliant) storage engine having commit, rollback, and crash recovery capabilities.

• For a FLOAT type field, the correct precision must be defined. By default, MySQL uses four bytes to store a FLOAT type that does not have an explicit precision definition. For example, this causes a number such as 12345.67890123 to be rounded off to 12345.7 during an INSERT. To prevent this, specify FLOAT(10,2) in the DDL file, which forces the database to use an eight-byte double-precision column. For more information, see http://dev.mysql.com/doc/mysql/en/numeric-types.html.

• To use || as the string concatenation symbol, start the MySQL server with the --sql-mode="PIPES_AS_CONCAT" option. For more information, see http://dev.mysql.com/doc/refman/5.0/en/server-sql-mode.html and http://dev.mysql.com/doc/mysql/en/ansi-mode.html.

• MySQL always starts a new connection when autoCommit==true is set. This ensures that each SQL statement forms a single transaction on its own. If you try to rollback or commit an SQL statement, you get an error message.

  javax.transaction.SystemException: java.sql.SQLException: 
  Can't call rollback when autocommit=true
  
  javax.transaction.SystemException: java.sql.SQLException: 
  Error open transaction is not closed
  

  To resolve this issue, add relaxAutoCommit=true to the JDBC URL. For more information, see http://forums.mysql.com/read.php?39,31326,31404.

• Change the trigger create format from the following:

  CREATE TRIGGER T_UNKNOWNPKVC1 
  BEFORE UPDATE ON UNKNOWNPKVC1
  FOR EACH ROW
          WHEN (NEW.VERSION = OLD.VERSION)
  BEGIN
          :NEW.VERSION := :OLD.VERSION + 1;
  END;
  /
  

  To the following:

  DELIMITER |
  CREATE TRIGGER T_UNKNOWNPKVC1
  BEFORE UPDATE ON UNKNOWNPKVC1
  FOR EACH ROW
          WHEN (NEW.VERSION = OLD.VERSION)
  BEGIN
          :NEW.VERSION := :OLD.VERSION + 1;
  END
  |
  DELIMITER ;
  

  For more information, see http://dev.mysql.com/doc/mysql/en/create-trigger.html.

• MySQL does not allow a DELETE on a row that contains a reference to itself. Here is an example that illustrates the issue.

  create table EMPLOYEE (
          empId   int         NOT NULL,
          salary  float(25,2) NULL,
          mgrId   int         NULL,
          PRIMARY KEY (empId),
          FOREIGN KEY (mgrId) REFERENCES EMPLOYEE (empId)
          ) ENGINE=InnoDB;
  
          insert into Employee values (1, 1234.34, 1);
          delete from Employee where empId = 1;
  

  This example fails with the following error message.

  ERROR 1217 (23000): Cannot delete or update a parent row: 
  a foreign key constraint fails
  

  To resolve this issue, change the table creation script to the following:

  create table EMPLOYEE (
          empId   int         NOT NULL,
          salary  float(25,2) NULL,
          mgrId   int         NULL,
          PRIMARY KEY (empId),
          FOREIGN KEY (mgrId) REFERENCES EMPLOYEE (empId)
          ON DELETE SET NULL
          ) ENGINE=InnoDB;
  
          insert into Employee values (1, 1234.34, 1);
          delete from Employee where empId = 1;
  

  This can be done only if the foreign key field is allowed to be null. For more information, see http://dev.mysql.com/doc/mysql/en/innodb-foreign-key-constraints.html.

• When an SQL script has foreign key constraints defined, capture-schema fails to capture the table information correctly. To work around the problem, remove the constraints and then run capture-schema. Here is an example that illustrates the issue.

  CREATE TABLE ADDRESSBOOKBEANTABLE (ADDRESSBOOKNAME VARCHAR(255) 
      NOT NULL PRIMARY KEY, 
  CONNECTEDUSERS              BLOB NULL, 
  OWNER                       VARCHAR(256), 
  FK_FOR_ACCESSPRIVILEGES     VARCHAR(256), 
  CONSTRAINT FK_ACCESSPRIVILEGE FOREIGN KEY (FK_FOR_ACCESSPRIVILEGES) 
      REFERENCES ACCESSPRIVILEGESBEANTABLE (ROOT) 
  ) ENGINE=InnoDB;  
  

  To resolve this issue, change the table creation script to the following:

  CREATE TABLE ADDRESSBOOKBEANTABLE (ADDRESSBOOKNAME VARCHAR(255) 
      NOT NULL PRIMARY KEY, 
  CONNECTEDUSERS              BLOB NULL, 
  OWNER                       VARCHAR(256), 
  FK_FOR_ACCESSPRIVILEGES     VARCHAR(256) 
  ) ENGINE=InnoDB;