32. Distributed Transactions with JTA

Spring Boot supports distributed JTA transactions across multiple XA resources using either an Atomikos or Bitronix embedded transaction manager. JTA transactions are also supported when deploying to a suitable Java EE Application Server.

When a JTA environment is detected, Spring’s JtaTransactionManager will be used to manage transactions. Auto-configured JMS, DataSource and JPA beans will be upgraded to support XA transactions. You can use standard Spring idioms such as @Transactional to participate in a distributed transaction. If you are within a JTA environment and still want to use local transactions you can set the spring.jta.enabled property to false to disable the JTA auto-configuration.

32.1 Using an Atomikos transaction manager

Atomikos is a popular open source transaction manager which can be embedded into your Spring Boot application. You can use the spring-boot-starter-jta-atomikos Starter POM to pull in the appropriate Atomikos libraries. Spring Boot will auto-configure Atomikos and ensure that appropriate depends-on settings are applied to your Spring beans for correct startup and shutdown ordering.

By default Atomikos transaction logs will be written to a transaction-logs directory in your application home directory (the directory in which your application jar file resides). You can customize this directory by setting a spring.jta.log-dir property in your application.properties file. Properties starting spring.jta. can also be used to customize the Atomikos UserTransactionServiceImp. See the AtomikosProperties Javadoc for complete details.

[Note]Note

To ensure that multiple transaction managers can safely coordinate the same resource managers, each Atomikos instance must be configured with a unique ID. By default this ID is the IP address of the machine on which Atomikos is running. To ensure uniqueness in production, you should configure the spring.jta.transaction-manager-id property with a different value for each instance of your application.

32.2 Using a Bitronix transaction manager

Bitronix is another popular open source JTA transaction manager implementation. You can use the spring-boot-starter-jta-bitronix starter POM to add the appropriate Birtronix dependencies to your project. As with Atomikos, Spring Boot will automatically configure Bitronix and post-process your beans to ensure that startup and shutdown ordering is correct.

By default Bitronix transaction log files (part1.btm and part2.btm) will be written to a transaction-logs directory in your application home directory. You can customize this directory by using the spring.jta.log-dir property. Properties starting spring.jta. are also bound to the bitronix.tm.Configuration bean, allowing for complete customization. See the Bitronix documentation for details.

[Note]Note

To ensure that multiple transaction managers can safely coordinate the same resource managers, each Bitronix instance must be configured with a unique ID. By default this ID is the IP address of the machine on which Bitronix is running. To ensure uniqueness in production, you should configure the spring.jta.transaction-manager-id property with a different value for each instance of your application.

32.3 Using a Java EE managed transaction manager

If you are packaging your Spring Boot application as a war or ear file and deploying it to a Java EE application server, you can use your application servers built-in transaction manager. Spring Boot will attempt to auto-configure a transaction manager by looking at common JNDI locations (java:comp/UserTransaction, java:comp/TransactionManager etc). If you are using a transaction service provided by your application server, you will generally also want to ensure that all resources are managed by the server and exposed over JNDI. Spring Boot will attempt to auto-configure JMS by looking for a ConnectionFactory at the JNDI path java:/JmsXA or java:/XAConnectionFactory and you can use the spring.datasource.jndi-name property to configure your DataSource.

32.4 Mixing XA and non-XA JMS connections

When using JTA, the primary JMS ConnectionFactory bean will be XA aware and participate in distributed transactions. In some situations you might want to process certain JMS messages using a non-XA ConnectionFactory. For example, your JMS processing logic might take longer than the XA timeout.

If you want to use a non-XA ConnectionFactory you can inject the nonXaJmsConnectionFactory bean rather than the @Primary jmsConnectionFactory bean. For consistency the jmsConnectionFactory bean is also provided using the bean alias xaJmsConnectionFactory.

For example:

// Inject the primary (XA aware) ConnectionFactory
@Autowired
private ConnectionFactory defaultConnectionFactory;

// Inject the XA aware ConnectionFactory (uses the alias and injects the same as above)
@Autowired
@Qualifier("xaJmsConnectionFactory")
private ConnectionFactory xaConnectionFactory;

// Inject the non-XA aware ConnectionFactory
@Autowired
@Qualifier("nonXaJmsConnectionFactory")
private ConnectionFactory nonXaConnectionFactory;

32.5 Supporting an alternative embedded transaction manager

The XAConnectionFactoryWrapper and XADataSourceWrapper interfaces can be used to support alternative embedded transaction managers. The interfaces are responsible for wrapping XAConnectionFactory and XADataSource beans and exposing them as regular ConnectionFactory and DataSource beans which will transparently enroll in the distributed transaction. DataSource and JMS auto-configuration will use JTA variants as long as you have a JtaTransactionManager bean and appropriate XA wrapper beans registered within your ApplicationContext.

The BitronixXAConnectionFactoryWrapper and BitronixXADataSourceWrapper provide good examples of how to write XA wrappers.