jpa - How to use Spring AbstractRoutingDataSource with dynamic datasources?

Question

I am working in a project using Spring, Spring Data JPA, Spring Security, Primefaces...

I was following this tutorial about dynamic datasource routing with spring.

In this tutorial, you can only achieve dynamic datasource switching between a pre-defined datasources.

Here is a snippet of my code :

springContext-jpa.xml

<bean id="dsCgWeb1" class="org.apache.commons.dbcp.BasicDataSource">
    <property name="driverClassName" value="${jdbc.driverClassName.Cargest_web}"></property>
    <property name="url" value="${jdbc.url.Cargest_web}"></property>
    <property name="username" value="${jdbc.username.Cargest_web}"></property>
    <property name="password" value="${jdbc.password.Cargest_web}"></property>
</bean>

<bean id="dsCgWeb2" class="org.apache.commons.dbcp.BasicDataSource">
    // same properties, different values ..
</bean>

<!--  Generic Datasource [Default : dsCargestWeb1]  -->
<bean id="dsCgWeb" class="com.cargest.custom.CargestRoutingDataSource">
    <property name="targetDataSources">
        <map>
            <entry key="1" value-ref="dsCgWeb1" />
            <entry key="2" value-ref="dsCgWeb2" />
        </map>
    </property>
    <property name="defaultTargetDataSource" ref="dsCgWeb1" />
</bean>

What i want to do is to make the targetDataSources map dynamic same as its elements too.

In other words, i want to fetch a certain database table, use properties stored in that table to create my datasources then put them in a map like targetDataSources.

Is there a way to do this ?

Answer 1

Nothing in AbstractRoutingDataSource forces you to use a static map of DataSourceS. It is up to you to contruct a bean implementing Map<Object, Object>, where key is what you use to select the DataSource, and value is a DataSource or (by default) a String referencing a JNDI defined data source. You can even modify it dynamically since, as the map is stored in memory, AbstractRoutingDataSource does no caching.

I have no full example code. But here is what I can imagine. In a web application, you have one database per client, all with same structure - ok, it would be a strange design, say it is just for the example. At login time, the application creates the datasource for the client and stores it in a map indexed by sessionId - The map is a bean in root context named dataSources

@Autowired
@Qualifier("dataSources");
Map<String, DataSource> sources;

// I assume url, user and password have been found from connected user
// I use DriverManagerDataSource for the example because it is simple to setup
DataSource dataSource = new DriverManagerDataSource(url, user, password);
sources.put(request.getSession.getId(), dataSource);

You also need a session listener to cleanup dataSources in its destroy method

@Autowired
@Qualifier("dataSources");
Map<String, DataSource> sources;

public void sessionDestroyed(HttpSessionEvent se)  {
    // eventually cleanup the DataSource if appropriate (nothing to do for DriverManagerDataSource ...)
    sources.remove(se.getSession.getId());
}

The routing datasource could be like :

public class SessionRoutingDataSource extends AbstractRoutingDataSource {

    @Override
    protected Object determineCurrentLookupKey() {
        HttpServletRequest request = ((ServletRequestAttributes)
                RequestContextHolder.getRequestAttributes()).getRequest();
        return request.getSession().getId();
    }

    @Autowired
    @Qualifier("dataSources")
    public void setDataSources(Map<String, DataSource> dataSources) {
        setTargetDataSources(dataSources);
}

I have not tested anything because it would be a lot of work to setting the different database, but I thing that it should be Ok. In real world there would not be a different data source per session but one per user with a count of session per user but as I said it is an over simplified example.