Skip to main content

Integration With Spring

The rdf4j-spring module allows for using an RDF4J repository as the data backend of a spring application.

New in RDF4J 4.0 Milestone 1 Experimental

A self-contained demo application can be found at rdf4j-spring-demo

Getting Started

To use RDF as the data backend of a spring application built with maven, use these dependencies:


… setting the property rdf4j.version is set to the RDF4J version you want (minimum 4.0.0).

In order for the application to run, a repository has to be configured:

To configure the application to access an existing repository, set the following configuration properties, e.g. in


To use an in-memory repository (for example, for unit tests), use


Programming with RDF4J-Spring

The main purpose of rdf4j-spring is to support accessing an RDF4J repository using the DAO pattern. DAOs are subclasses of RDF4JDao and use the RDF4JTemplate for accessing the RDF4J repository configured for the application.


The RDF4JTemplate is the class used to access a Repository in rdf4j-spring. A bean of this type is configured at start up and available for wiring into beans. The RDF4JTemplate accesses the Repository through a RepositoryConnection that it obtains from a RepositoryConnectionFactory . This indirection allows for using a connection pool, connect RDF4J to spring’s transaction management, and provide query logging to a file or exposing query statistics via JMX. These features can be enabled/disabled using configuration properties (see Configuration)

Wiring into a spring bean

To use the RDF4JTemplate in a bean, define that bean in the spring application’s configuration and wire the RDF4JTemplate in:

public class MyAppConfig {
	public MyBeanClass getMyBean(@Autowired RDF4JTemplate template){
        return new MyBeanClass(template);
public class MyBeanClass {

	private RDF4JTemplate rdf4JTemplate;
    public MyBeanClass(RDF4JTemplate template){
        this.rdf4jTemplate = template;

Evaluating queries and executing updates

The RDF4JTemplate offers various ways to access the repository. For example, to evaluate a TupleQuery using the RDF4JTemplate (in this case, counting all triples):

int count = rdf4JTemplate
				.tupleQuery("SELECT (count(?a) as ?cnt) WHERE { ?a ?b ?c }")
				.toSingleton(bs -> TypeMappingUtils.toInt(QueryResultUtils.getValue(bs, "cnt")));

The query, provided through the tupleQuery method, is executed with the call to evaluateAndConvert(), which returns a TupleQueryResultConverter . The latter provides methods for converting the TupleQueryResult of the query into an object, an Optional, a Map, Set, List, or Stream. In the example, we are just interested in the count as an int - one single object - so we use the toSingleton() method and convert the value of the projection variable to an int. The conversion is done using TypeMappingUtils ; the extraction of the variable’s value from the BindingSet bs is done using QueryResultUtils .

Pre-binding variables

For binding variables before executing a query or update, use the OperationBuilder returned by the tupleQuery(), graphQuery, or update methods. It provides various withBinding() methods following the builder pattern, allowing for binding variables, as illustrated in the following example.

Set<IRI> artists = rdf4JTemplate
                    .tupleQuery("PREFIX ex: <>"
				            + "SELECT distinct ?artist "
				            + "WHERE { ?artist a ?type }")
				    .withBinding("type", EX.Artist)
				    .toSet(bs -> QueryResultUtils.getIRI(bs, "artist"));

Using the RepositoryConnection directly

For using the RepostoryConnection directly, without the need to generate a result, the consumeConnection() method is used:

rdf4JTemplate.consumeConnection(con -> con.remove(EX.Picasso, RDF.TYPE, EX.Artist);

Alternatively, to generate a result, the applyToConnection() method is used:

boolean isPresent = rdf4JTemplate.applyToConnection(
                        con -> con.hasStatement(EX.Picasso, RDF.TYPE, EX.Artist, true);

Using SPARQL queries/updates from external files

For running queries or updates from external resources, the [(tupleQuery|graphQuery|update)FromResource] methods can be used.

For example, the sparql/construct-artists.rq file on the classpath might contain this query:

PREFIX ex: <>
CONSTRUCT {?artist ?p ?o } WHERE { ?artist a ex:Artist; ?p ?o }

and could be evaluated using

Model model = rdf4JTemplate.graphQueryFromResource(

The resource to be read is resolved by spring’s ResourceLoader, which supports fully qualified URLs (e.g., file:// URLs, relative paths and classpath: pseudo-URLs.)

Implementing a DAO

Any spring bean that uses the RDF4JTemplate can be seen as a DAO and participates in transactionality, query logging, caching, etc. However, rdf4j-spring provides a few base classes that provide frequently used functionality.


RDF4JDao is a suitable base class for a general-purpose DAO. It provides two functionalities to subclasses:

  • The RDF4JTemplate is automatically wired into the bean and it is available through getRDF4JTemplate()

  • It provides a simple management facility for SPARQL query/update strings. This allows for SPARQL queries being generated only once (by String concatenation, read from a file, or built with the SparqlBuilder). The queries are prepared in the template method prepareNamedSparqlSuppliers():

In the following example, we

  • create a DAO, extending RDF4JDao
  • annotate it with @Component so it gets auto-detected during Spring’s component scan
  • create an inner class, QUERY_KEYS, as a container for String constants we use for query keys
  • implement the prepareNamedSparqlSuppliers method and add one query
  • use the prepared query in a DAO method (getArtistsWithoutPaintings()). We access the prepared query with getNamedTupleQuery(String) , passing the constant we defined in QUERY_KEYS.

@Component // make the DAO a spring component so it's auto-detected in the classpath scan
public class ArtistDao extends RDF4JDao {
    // constructor, other methods etc

    // recommended: encapsulate the keys for queries in an object
    // so it's easier to find them when you need them
	static abstract class QUERY_KEYS {
		public static final String ARTISTS_WITHOUT_PAINTINGS = "artists-without-paintings";

    // prepare the named queries, assigning each one of the keys
	protected NamedSparqlSupplierPreparer prepareNamedSparqlSuppliers(NamedSparqlSupplierPreparer preparer) {
		return preparer
                                .and(ARTIST_ID.has(iri(EX.creatorOf), Painting.PAINTING_ID).optional())

    // use the named query with getNamedTupleQuery(String)
	public Set<Artist> getArtistsWithoutPaintings(){
						.map(bs -> QueryResultUtils.getIRI(bs, ARTIST_ID))
						.map(iri -> getById(iri))
    // ...



The SimpleRDF4JCRUDDao is a suitable base class for a DAO for creating, reading, updating, and deleting one class of entities. It requires two type parameters, ENTITY and ID. It provides create, read, update, and delete functionality for the ENTITY class, using the ID class wherever the entity’s identifier is required.

Subclasses of SimpleRDF4JCRUDDao must implement a couple of template methods in order to customize the generic behaviour for the specific entity and id classes.

In the following, we use the entity Artist (as used in the demo application) as an example. Note that we define public constants of type Variable , one corresponding to each of the entity’s fields.

public class Artist {
    // recommended pattern: use a public Variable constant for each of the entities fields 
    // for use in queries and result processing. 
	public static final Variable ARTIST_ID = SparqlBuilder.var("artist_id"); 
	public static final Variable ARTIST_FIRST_NAME = SparqlBuilder.var("artist_firstName");
	public static final Variable ARTIST_LAST_NAME = SparqlBuilder.var("artist_lastName");
	private IRI id;
	private String firstName;
	private String lastName;
    // getter, setter, constructor, ...
    // be sure to implement equals() and hashCode() for proper behaviour of collections!

The ArtistDao is shown in the following code snippets.

We recommend to use @Component for auto-detection. Implementing the constructor is required.

@Component // again, make it a component (see above) 
public class ArtistDao extends SimpleRDF4JCRUDDao<Artist, IRI> {

	public ArtistDao(RDF4JTemplate rdf4JTemplate) {

The populateIdBindings method is called by the superclass to bind the id to variable(s) in a SPARQL query.

	protected void populateIdBindings(MutableBindings bindingsBuilder, IRI iri) {
		bindingsBuilder.add(ARTIST_ID, iri);

The populateBindingsForUpdate method is called by the superclass to bind all non-id variables when performing an update.

	protected void populateBindingsForUpdate(MutableBindings bindingsBuilder, Artist artist) {
				.add(ARTIST_FIRST_NAME, artist.getFirstName())
				.add(ARTIST_LAST_NAME, artist.getLastName());

The mapSolution method converts a query solution, i.e., a BindingSet , to an instance of the entity.

	protected Artist mapSolution(BindingSet querySolution) {
		Artist artist = new Artist();
		artist.setId(QueryResultUtils.getIRI(querySolution, ARTIST_ID));
		artist.setFirstName(QueryResultUtils.getString(querySolution, ARTIST_FIRST_NAME));
		artist.setLastName(QueryResultUtils.getString(querySolution, ARTIST_LAST_NAME));
		return artist;

The getReadQuery method provides the SPARQL string used to read one entity. Note that the variable names must be the same ones used in mapSolution(BindingSet). It may be cleaner to use the SparqlBuilder for generating this string.

	protected String getReadQuery() {
		return "prefix foaf: <> "
				+ "prefix ex: <> "
				+ "SELECT ?artist_id ?artist_firstName ?artist_lastName where {"
				+ "?artist_id a ex:Artist; "
				+ "    foaf:firstName ?artist_firstName; "
				+ "    foaf:surname ?artist_lastName ."
				+ " } ";

The getInsertSparql(ENTITY) method provides the SPARQL string for inserting a new instance. This SPARQL operation will also be used for updates. If updates require a different operation from inserts, it must be provided by implementing getUpdateSparql(ENTITY).

	protected NamedSparqlSupplier getInsertSparql(Artist artist) {
		return NamedSparqlSupplier.of("insert", () -> Queries.INSERT(ARTIST_ID.isA(iri(EX.Artist))

The getInputId(ENTITY) method is used to generate the id of an entity to be inserted. Here, we use the id of the specified artist object; if it is null we generate a new IRI using getRdf4JTemplate().getNewUUID().

	protected IRI getInputId(Artist artist) {
		if (artist.getId() == null) {
			return getRdf4JTemplate().getNewUUID();
		return artist.getId();
Composite Keys

If the entity uses a composite key, a class implementing CompositeKey must be used for the ID type parameter. For a key consisting of two components, the CompositeKey2 class is available. If more components are needed, the key class can be modeled after that one.


It is not uncommon for an application to read a relation present in the repository data into a Map. For example, we might want to group painting ids by artist id. The RelationMapBuilder provides the necesary functionality for such cases:

RelationMapBuilder b = new RelationMapBuilder(getRDF4JTemplate(), EX.creatorOf);
Map<IRI, Set<IRI>> paintingsByArtists = b.buildOneToMany();

Additional Functionality:

  • The constraints(GraphPattern) method restricts the relation
  • The relationIsOptional() method allows for the object to be missing, in which case an empty set is generated for the subject key.
  • The useRelationObjectAsKey() method flips the map such that the objects of the relation are used as keys and the subjects are aggregated.
  • The buildOneToOne() method returns a one to one mapping, which dies horribly if the data is not 1:1


The RDF4JCRUDDao is essentially the same as the SimpleRDF4JCRUDDao, with the one difference that it has three type parameters, ENTITY, INPUT, and ID. The class thus allows different classes for input and output: creation and updates use INPUT, e.g. save(INPUT), reading methods use ENTITY, e.g. ENTITY getById(ID).

Service Layer

Usually, the functionality offered by DAOs is rather narrow, e.g. CRUD methods for one entity class. They are combined to provide a wider range of functionality in the servcie layer. The only thing one needs to know when implementing the service layer with rdf4j-spring DAOs is that its methods need to participate in spring’s transaction management. The most straightforward way to do this is to use the @Transactional method annotation, causing the service object to be wrapped with a proxy that takes care of transactionality.

The following code snippet, taken from the demo’s ArtService class, shows part of a simple service.

public class ArtService {
	private ArtistDao artistDao;

	private PaintingDao paintingDao;

	public Artist createArtist(String firstName, String lastName) {
		Artist artist = new Artist();

	public Painting createPainting(String title, String technique, IRI artist) {
		Painting painting = new Painting();

	public List<Painting> getPaintings() {
		return paintingDao.list();

	public List<Artist> getArtists() {
		return artistDao.list();

	public Set<Artist> getArtistsWithoutPaintings(){
		return artistDao.getArtistsWithoutPaintings();
    // ...


Testing with Junit 5

Testing an application built with rdf4j-spring can be done at the DAO layer as well as on the service layer. Generally, applications will have more than one test classes.

The common approach is to have a configuration for tests that is shared by all tests, and this configuration prepares the spring context with all the required facilities. A minimal, shared test configuration is the following. Note that it imports RDF4JTestConfig :

public class TestConfig {

    DataInserter getDataInserter() {
        return new DataInserter();

With this configuration, a test class can use the dataInserter bean to insert data into an inmemory repository before each test:

@ContextConfiguration(classes = { TestConfig.class })
		properties = {
public class ArtistDaoTests {

    private ArtistDao artistDao;       

	public static void insertTestData(
			@Autowired DataInserter dataInserter,
			@Value("classpath:/data/my-testdata.ttl") Resource dataFile) {
    public void testReadArtist(){
         // ...          


Testing against a local database

The inmemory repository is likely to behave differently from any database used in production in some edge cases. It is recommended to test against a local installation of the database that is used in production in addition to testing against the inmemory repository.

With rdf4j-spring this is quite straightforward:

  1. install the database locally and create a repository for the tests
  2. provide a property file on the classpath with the necessary properties to connect to that repository (rdf4j.spring.repository.remote.* properties)
  3. Create a subclass of your test class and provide the properties file through the @TestPropertySource annotation
  4. Use the @Tag annotation, so you can easily switch the test on or off using the configuration of your test environment (most likely the Maven Surefire Plugin), as the local database installation will not be present in many build environments.


public class ArtistDaoDbTests extends ArtistDaoTests {
// no code needed, the class is just created to run your ArtistDaoTests with a different configuration     


In addition to query logging, if you need to get a close look at what’s happening inside the rdf4j-spring code, set the loglevel for org.eclipse.rdf4j.spring to DEBUG. Sometimes it may be required to look into what spring is doing. In this case, set org.springframework to DEBUG or even TRACE.

One way to do this is to provide a logback.xml file on the classpath, as can be found in the source at rdf4j-spring/src/test/resources/logback.xml.

Another way to set the loglevel is to provide an application property starting with logging.level., e.g.

which can be provided in an (for details and other ways to do that, have a look at the documentation on Externalied Configuration in Spring).


rdf4j-spring makes use of the auto-configuration feature in Spring (configured in the source file rdf4j-spring/META-INF/spring.factories). That means that bean creation at start up is governed by configuration properties, all of which are prefixed by rdf4j.spring.

The following table shows all subsystems with their property prefixes, the packages they reside in, and the class holding their properties.

Subsystem property prefix package (links to reference) Properties class
Repository rdf4j.spring.repository. org.eclipse.rdf4j.spring.repository RemoteRepositoryProperties and InMemoryRepositoryProperties
Transaction management rdf4j.spring.tx. org.eclipse.rdf4j.spring.tx TxProperties
Connection Pooling rdf4j.spring.pool. org.eclipse.rdf4j.spring.pool PoolProperties
Operation caching rdf4j.spring.operationcache. org.eclipse.rdf4j.spring.operationcache OperationCacheProperties
Operation logging rdf4j.spring.operationlog. org.eclipse.rdf4j.spring.operationlog OperationLogProperties and OperationLogJmxProperties
Query result caching rdf4j.spring.resultcache. org.eclipse.rdf4j.spring.resultcache ResultCacheProperties
UUIDSource rdf4j.spring.uuidsource. org.eclipse.rdf4j.spring.uuidsource SimpleUUIDSourceProperties , NoveltyCheckingUUIDSourceProperties , UUIDSequenceProperties , and PredictableUUIDSourceProperties

These subsystems and their configuration are described in more detail below.


As stated in the Getting Started section, to configure the application to access an existing repository, set the following configuration properties, e.g. in


To use an in-memory repository (for example, for unit tests), use


Transaction management

By default, rdf4j-spring connects with Spring’s PlatformTransactionManager. To disable this connection, use


Connection Pooling

Creating a RepositoryConnection has a certain overhead that many applications wish to avoid. rdf4j-spring allows for pooling of such connections. Several configuration options, such as the maximum number of connections, are available (see PoolProperties ).

To enable, use


Operation caching

SPARQL operations (queries and updates) require some computation time to prepare from the SPARQL string they are based on. In rdf4j-spring, this process is hidden from clients and happens in the RDF4JTemplate. By default, operations are not cached, and the same operation executed multiple times always has the overhead of parsing the SPARQL string and generating the operation. If this feature is enabled, operations are cached per connection.

Note: If connection pooling is enabled, it is possible that operations created in different threads will use different connections and will therefore all generate their own instance of the SPARQL operation, thus reducing the speedup incurred by operation caching.

To enable, use


Operation logging

(aka Query logging)

Two options are available for logging operations (queries and updates) sent to the repository:

Operation logging via SLF4J

Each operation is written to the logger org.eclipse.rdf4j.spring.operationlog.log.slf4 with loglevel DEBUG.

To enable, use


Operation logging via JMX

Each operation is recorded (if identical operations are executed, statistics are aggregated) and exposed via JMX.

To enable, use


Query result caching

Applications that frequently execute the same queries might profit from result caching. If enabled, query results are cached on a per-connection basis. By default, this cache is cleared at the end of the ongoing transaction. The performance impact of result caching is application-specific and is not unlikely to be negative. Measure carefully!

However, if the application is the only one using the repository, and therefore no updates are possible that the application does not know about, the property rdf4j.spring.resultcache.assumeNoOtherRepositoryClients=true can be set. In this case, results are copied to a global cache that all connections have access to, and which is only cleared when the application executes an update.

To enable result caching, use



Using UUIDs as identifiers for entities is a common strategy for applications using an RDF store as their backend. Doing this requires a source of new, previously unused UUIDs for new entities created by the application. Conversely, in unit tests, it is sometimes required that the UUIDs are generated in a predictable manner, so that actual results can be compared with expected results containing generated UUIDs.

The UUIDSource subsystem provides different implementations of the UUIDSource interface. The configuration of this subsystem determines which implementation is wired into the RDF4JTemplate at start up and gets used by the application.

In our opinion, the default implementation, DefaultUUIDSource is sufficient for generating previously unused UUIDs. Collisions are possible but sufficiently unlikely, so using any one of noveltychecking, sequence, and simple subsystems should not be necessary.

For using the predictable UUIDSource, which always produces the same sequence of UUIDs, use



The RDF4J-Spring module, the RDF4J-Spring-Demo, and this documentation have been developed in the project ‘BIM-Interoperables Merkmalservice’, funded by the Austrian Research Promotion Agency and Österreichische Bautechnik Veranstaltungs GmbH.

Back to the top