Wednesday, January 25, 2012

Storing hierarchical data in MongoDB

Continuing NoSQL journey with MongoDB, I would like to touch one specific use case which comes up very often: storing hierarchical document relations. MongoDB is awesome document data store but what if documents have parent-child relationships? Can we effectively store and query such document hierarchies? The answer, for sure, is yes, we can. MongoDB has several recommendations how to store Trees in MongoDB. The one solution described there as well and quite widely used is using materialized path.

Let me explain how it works by providing very simple examples. As in previous posts, we will build Spring application using recently released version 1.0 of Spring Data MongoDB project. Our POM file contains very basic dependencies, nothing more.


    4.0.0

    mongodb
    com.example.spring
    0.0.1-SNAPSHOT
    jar

    
        UTF-8
        3.0.7.RELEASE
    

    
        
            org.springframework.data
            spring-data-mongodb
            1.0.0.RELEASE
            
                
                    org.springframework
                    spring-beans
                
                
                    org.springframework
                    spring-expression
                
            
        

        
            cglib
            cglib-nodep
            2.2
        

        
            log4j
            log4j
            1.2.16
        

        
            org.mongodb
            mongo-java-driver
            2.7.2
        

        
            org.springframework
            spring-core
            ${spring.version}
        

        
            org.springframework
            spring-context
            ${spring.version}
        

        
            org.springframework
            spring-context-support
            ${spring.version}
        
    

    
        
            
                org.apache.maven.plugins
                maven-compiler-plugin
                2.3.2
                
                    1.6
                    1.6
                
            
        
    

To properly configure Spring context, I will use configuration approach utilizing Java classes. I am more and more advocating to use this style as it provides strong typed configuration and most of the mistakes could be caught on compilation time, no need to inspect your XML files anymore. Here how it looks like:

package com.example.mongodb.hierarchical;

import org.springframework.context.annotation.Bean;
import org.springframework.context.annotation.Configuration;
import org.springframework.data.mongodb.core.MongoFactoryBean;
import org.springframework.data.mongodb.core.MongoTemplate;
import org.springframework.data.mongodb.core.SimpleMongoDbFactory;

@Configuration
public class AppConfig {
    @Bean
    public MongoFactoryBean mongo() {
        final MongoFactoryBean factory = new MongoFactoryBean();
        factory.setHost( "localhost" );
        return factory;
    }

    @Bean
    public SimpleMongoDbFactory mongoDbFactory() throws Exception{
        return new SimpleMongoDbFactory( mongo().getObject(), "hierarchical" );
    }

    @Bean
    public MongoTemplate mongoTemplate() throws Exception {
        return new MongoTemplate( mongoDbFactory() );
    }

    @Bean
    public IDocumentHierarchyService documentHierarchyService() throws Exception {
        return new DocumentHierarchyService( mongoTemplate() );
    }
}

That's pretty nice and clear. Thanks, Spring guys! Now, all boilerplate stuff is ready. Let's move to interesting part: documents. Our database will contain 'documents' collection which stores documents of type SimpleDocument. We describe this using Spring Data MongoDB annotations for SimpleDocument POJO.

package com.example.mongodb.hierarchical;

import java.util.Collection;
import java.util.HashSet;

import org.springframework.data.annotation.Id;
import org.springframework.data.annotation.Transient;
import org.springframework.data.mongodb.core.mapping.Document;
import org.springframework.data.mongodb.core.mapping.Field;

@Document( collection = "documents" )
public class SimpleDocument {
    public static final String PATH_SEPARATOR = ".";

    @Id private String id;
    @Field private String name;
    @Field private String path;

    // We won't store this collection as part of document but will build it on demand
    @Transient private Collection< SimpleDocument > documents = new HashSet< SimpleDocument >();

    public SimpleDocument() {
    }

    public SimpleDocument( final String id, final String name ) {
        this.id = id;
        this.name = name;
        this.path = id;
    }

    public SimpleDocument( final String id, final String name, final SimpleDocument parent ) {
        this( id, name );
        this.path = parent.getPath() + PATH_SEPARATOR + id;
    }

    public String getId() {
        return id;
    }

    public void setId(String id) {
        this.id = id;
    }

    public String getName() {
        return name;
    }

    public void setName(String name) {
        this.name = name;
    }

    public String getPath() {
        return path;
    }

    public void setPath(String path) {
        this.path = path;
    }

    public Collection< SimpleDocument > getDocuments() {
        return documents;
    }
}

Let me explain few things here. First, magic property path: this is a key to construct and query through our hierarchy. Path contains identifiers of all document's parents, usually divided by some kind of separator, in our case just . (dot). Storing document hierarchical relationships in this way allows quickly build hierarchy, search and navigate. Second, notice transient documents collection: this non-persistent collection is constructed by persistent provider and contains all descendant documents (which, in case, also contain own descendants). Let see it in action by looking into find method implementation:

package com.example.mongodb.hierarchical;

import java.util.Arrays;
import java.util.Collection;
import java.util.HashMap;
import java.util.Map;

import org.springframework.data.mongodb.core.MongoOperations;
import org.springframework.data.mongodb.core.query.Criteria;
import org.springframework.data.mongodb.core.query.Query;

public class DocumentHierarchyService {
    private MongoOperations template;

    public DocumentHierarchyService( final MongoOperations template ) {
        this.template = template;
    }

    @Override
    public SimpleDocument find( final String id ) {
        final SimpleDocument document = template.findOne(
            Query.query( new Criteria( "id" ).is( id ) ),
            SimpleDocument.class
        );

        if( document == null ) {
            return document;
        }

        return build(
            document,
            template.find(
                Query.query( new Criteria( "path" ).regex( "^" + id + "[.]" ) ),
                SimpleDocument.class
            )
        );
    }

    private SimpleDocument build( final SimpleDocument root, final Collection< SimpleDocument > documents ) {
        final Map< String, SimpleDocument > map = new HashMap< String, SimpleDocument >();

        for( final SimpleDocument document: documents ) {
            map.put( document.getPath(), document );
        }

        for( final SimpleDocument document: documents ) {
            map.put( document.getPath(), document );

            final String path = document
                .getPath()
                .substring( 0, document.getPath().lastIndexOf( SimpleDocument.PATH_SEPARATOR ) );

            if( path.equals( root.getPath() ) ) {
                root.getDocuments().add( document );
            } else {
                final SimpleDocument parent = map.get( path );
                if( parent != null ) {
                    parent.getDocuments().add( document );
                }
            }
        }

        return root;
    }
}

As you can see, to get single document with a whole hierarchy we need to run just two queries (but more optimal algorithm could reduce it to just one single query). Here is a sample hierarchy and the the result of reading root document from MongoDB


template.dropCollection( SimpleDocument.class );

final SimpleDocument parent = new SimpleDocument( "1", "Parent 1" );
final SimpleDocument child1 = new SimpleDocument( "2", "Child 1.1", parent );
final SimpleDocument child11 = new SimpleDocument( "3", "Child 1.1.1", child1 );
final SimpleDocument child12 = new SimpleDocument( "4", "Child 1.1.2", child1 );
final SimpleDocument child121 = new SimpleDocument( "5", "Child 1.1.2.1", child12 );
final SimpleDocument child13 = new SimpleDocument( "6", "Child 1.1.3", child1 );
final SimpleDocument child2 = new SimpleDocument( "7", "Child 1.2", parent );

template.insertAll( Arrays.asList( parent, child1, child11, child12, child121, child13, child2 ) );

...

final ApplicationContext context = new AnnotationConfigApplicationContext( AppConfig.class );
final IDocumentHierarchyService service = context.getBean( IDocumentHierarchyService.class );

final SimpleDocument document = service.find( "1" );
//  Printing document show following hierarchy:
//
//  Parent 1
//   |-- Child 1.1
//     |-- Child 1.1.1
//     |-- Child 1.1.3
//     |-- Child 1.1.2
//       |-- Child 1.1.2.1
//   |-- Child 1.2

That's it. Simple a powerful concept. Sure, adding index on a path property will speed up query significantly. There are a plenty of improvements and optimizations but basic idea should be clear now.