Using DBRefs

The mapping framework does not have to store child objects embedded within the document. You can also store them separately and use a DBRef to refer to that document. When the object is loaded from MongoDB, those references are eagerly resolved so that you get back a mapped object that looks the same as if it had been stored embedded within your top-level document.

The following example uses a DBRef to refer to a specific document that exists independently of the object in which it is referenced (both classes are shown in-line for brevity’s sake):

@Document
public class Account {

  @Id
  private ObjectId id;
  private Float total;
}

@Document
public class Person {

  @Id
  private ObjectId id;
  @Indexed
  private Integer ssn;
  @DBRef
  private List<Account> accounts;
}

You need not use @OneToMany or similar mechanisms because the List of objects tells the mapping framework that you want a one-to-many relationship. When the object is stored in MongoDB, there is a list of DBRefs rather than the Account objects themselves. When it comes to loading collections of DBRefs it is advisable to restrict references held in collection types to a specific MongoDB collection. This allows bulk loading of all references, whereas references pointing to different MongoDB collections need to be resolved one by one.

The mapping framework does not handle cascading saves. If you change an Account object that is referenced by a Person object, you must save the Account object separately. Calling save on the Person object does not automatically save the Account objects in the accounts property.

DBRefs can also be resolved lazily. In this case the actual Object or Collection of references is resolved on first access of the property. Use the lazy attribute of @DBRef to specify this. Required properties that are also defined as lazy loading DBRef and used as constructor arguments are also decorated with the lazy loading proxy making sure to put as little pressure on the database and network as possible.

Lazily loaded DBRefs can be hard to debug. Make sure tooling does not accidentally trigger proxy resolution by e.g. calling toString() or some inline debug rendering invoking property getters. Please consider to enable trace logging for org.springframework.data.mongodb.core.convert.DefaultDbRefResolver to gain insight on DBRef resolution.

Lazy loading may require class proxies, that in turn, might need access to jdk internals, that are not open, starting with Java 16+, due to JEP 396: Strongly Encapsulate JDK Internals by Default. For those cases please consider falling back to an interface type (eg. switch from ArrayList to List) or provide the required --add-opens argument.

Using Document References

Using @DocumentReference offers a flexible way of referencing entities in MongoDB. While the goal is the same as when using DBRefs, the store representation is different. DBRef resolves to a document with a fixed structure as outlined in the MongoDB Reference documentation.
Document references, do not follow a specific format. They can be literally anything, a single value, an entire document, basically everything that can be stored in MongoDB. By default, the mapping layer will use the referenced entities id value for storage and retrieval, like in the sample below.

@Document
class Account {

  @Id
  String id;
  Float total;
}

@Document
class Person {

  @Id
  String id;

  @DocumentReference                                   (1)
  List<Account> accounts;
}

Account account = …

template.insert(account);                               (2)

template.update(Person.class)
  .matching(where("id").is(…))
  .apply(new Update().push("accounts").value(account)) (3)
  .first();

{
  "_id" : …,
  "accounts" : [ "6509b9e" … ]                        (4)
}

1	Mark the collection of `Account` values to be referenced.
2	The mapping framework does not handle cascading saves, so make sure to persist the referenced entity individually.
3	Add the reference to the existing entity.
4	Referenced `Account` entities are represented as an array of their `_id` values.

The sample above uses an _id-based fetch query ({ '_id' : ?#{#target} }) for data retrieval and resolves linked entities eagerly. It is possible to alter resolution defaults (listed below) using the attributes of @DocumentReference

Table 1. @DocumentReference defaults
Attribute	Description	Default
`db`	The target database name for collection lookup.	`MongoDatabaseFactory.getMongoDatabase()`
`collection`	The target collection name.	The annotated property’s domain type, respectively the value type in case of `Collection` like or `Map` properties, collection name.
`lookup`	The single document lookup query evaluating placeholders via SpEL expressions using `#target` as the marker for a given source value. `Collection` like or `Map` properties combine individual lookups via an `$or` operator.	An `_id` field based query (`{ '_id' : ?#{#target} }`) using the loaded source value.
`sort`	Used for sorting result documents on server side.	None by default. Result order of `Collection` like properties is restored based on the used lookup query on a best-effort basis.
`lazy`	If set to `true` value resolution is delayed upon first access of the property.	Resolves properties eagerly by default.

@DocumentReference(lookup) allows defining filter queries that can be different from the _id field and therefore offer a flexible way of defining references between entities as demonstrated in the sample below, where the Publisher of a book is referenced by its acronym instead of the internal id.

@Document
class Book {

  @Id
  ObjectId id;
  String title;
  List<String> author;

  @Field("publisher_ac")
  @DocumentReference(lookup = "{ 'acronym' : ?#{#target} }") (1)
  Publisher publisher;
}

@Document
class Publisher {

  @Id
  ObjectId id;
  String acronym;                                            (1)
  String name;

  @DocumentReference(lazy = true)                            (2)
  List<Book> books;

}

Book document

{
  "_id" : 9a48e32,
  "title" : "The Warded Man",
  "author" : ["Peter V. Brett"],
  "publisher_ac" : "DR"
}

Publisher document

{
  "_id" : 1a23e45,
  "acronym" : "DR",
  "name" : "Del Rey",
  …
}

1	Use the `acronym` field to query for entities in the `Publisher` collection.
2	Lazy load back references to the `Book` collection.

The above snippet shows the reading side of things when working with custom referenced objects. Writing requires a bit of additional setup as the mapping information do not express where #target stems from. The mapping layer requires registration of a Converter between the target document and DocumentPointer, like the one below:

@WritingConverter
class PublisherReferenceConverter implements Converter<Publisher, DocumentPointer<String>> {

	@Override
	public DocumentPointer<String> convert(Publisher source) {
		return () -> source.getAcronym();
	}
}

If no DocumentPointer converter is provided the target reference document can be computed based on the given lookup query. In this case the association target properties are evaluated as shown in the following sample.

@Document
class Book {

  @Id
  ObjectId id;
  String title;
  List<String> author;

  @DocumentReference(lookup = "{ 'acronym' : ?#{acc} }") (1) (2)
  Publisher publisher;
}

@Document
class Publisher {

  @Id
  ObjectId id;
  String acronym;                                        (1)
  String name;

  // ...
}

{
  "_id" : 9a48e32,
  "title" : "The Warded Man",
  "author" : ["Peter V. Brett"],
  "publisher" : {
    "acc" : "DOC"
  }
}

1	Use the `acronym` field to query for entities in the `Publisher` collection.
2	The field value placeholders of the lookup query (like `acc`) is used to form the reference document.

It is also possible to model relational style One-To-Many references using a combination of @ReadonlyProperty and @DocumentReference. This approach allows link types without storing the linking values within the owning document but rather on the referencing document as shown in the example below.

@Document
class Book {

  @Id
  ObjectId id;
  String title;
  List<String> author;

  ObjectId publisherId;                                        (1)
}

@Document
class Publisher {

  @Id
  ObjectId id;
  String acronym;
  String name;

  @ReadOnlyProperty                                            (2)
  @DocumentReference(lookup="{'publisherId':?#{#self._id} }")  (3)
  List<Book> books;
}

Book document

{
  "_id" : 9a48e32,
  "title" : "The Warded Man",
  "author" : ["Peter V. Brett"],
  "publisherId" : 8cfb002
}

Publisher document

{
  "_id" : 8cfb002,
  "acronym" : "DR",
  "name" : "Del Rey"
}

1	Set up the link from `Book` (reference) to `Publisher` (owner) by storing the `Publisher.id` within the `Book` document.
2	Mark the property holding the references to be readonly. This prevents storing references to individual `Book`s with the `Publisher` document.
3	Use the `#self` variable to access values within the `Publisher` document and in this retrieve `Books` with matching `publisherId`.

With all the above in place it is possible to model all kind of associations between entities. Have a look at the non-exhaustive list of samples below to get feeling for what is possible.

Example 1. Simple Document Reference using id field

class Entity {
  @DocumentReference
  ReferencedObject ref;
}

// entity
{
  "_id" : "8cfb002",
  "ref" : "9a48e32" (1)
}

// referenced object
{
  "_id" : "9a48e32" (1)
}

1	MongoDB simple type can be directly used without further configuration.

Example 2. Simple Document Reference using id field with explicit lookup query

class Entity {
  @DocumentReference(lookup = "{ '_id' : '?#{#target}' }") (1)
  ReferencedObject ref;
}

// entity
{
  "_id" : "8cfb002",
  "ref" : "9a48e32"                                        (1)
}

// referenced object
{
  "_id" : "9a48e32"
}

1	target defines the reference value itself.

Example 3. Document Reference extracting the refKey field for the lookup query

class Entity {
  @DocumentReference(lookup = "{ '_id' : '?#{refKey}' }")  (1) (2)
  private ReferencedObject ref;
}

@WritingConverter
class ToDocumentPointerConverter implements Converter<ReferencedObject, DocumentPointer<Document>> {
	public DocumentPointer<Document> convert(ReferencedObject source) {
		return () -> new Document("refKey", source.id);    (1)
	}
}

// entity
{
  "_id" : "8cfb002",
  "ref" : {
    "refKey" : "9a48e32"                                   (1)
  }
}

// referenced object
{
  "_id" : "9a48e32"
}

1	The key used for obtaining the reference value must be the one used during write.
2	`refKey` is short for `target.refKey`.

Example 4. Document Reference with multiple values forming the lookup query

class Entity {
  @DocumentReference(lookup = "{ 'firstname' : '?#{fn}', 'lastname' : '?#{ln}' }") (1) (2)
  ReferencedObject ref;
}

// entity
{
  "_id" : "8cfb002",
  "ref" : {
    "fn" : "Josh",           (1)
    "ln" : "Long"            (1)
  }
}

// referenced object
{
  "_id" : "9a48e32",
  "firstname" : "Josh",      (2)
  "lastname" : "Long",       (2)
}

1	Read/write the keys `fn` & `ln` from/to the linkage document based on the lookup query.
2	Use non id fields for the lookup of the target documents.

Example 5. Document Reference reading from a target collection

class Entity {
  @DocumentReference(lookup = "{ '_id' : '?#{id}' }", collection = "?#{collection}") (2)
  private ReferencedObject ref;
}

@WritingConverter
class ToDocumentPointerConverter implements Converter<ReferencedObject, DocumentPointer<Document>> {
	public DocumentPointer<Document> convert(ReferencedObject source) {
		return () -> new Document("id", source.id)                                   (1)
                           .append("collection", … );                                (2)
	}
}

// entity
{
  "_id" : "8cfb002",
  "ref" : {
    "id" : "9a48e32",                                                                (1)
    "collection" : "…"                                                               (2)
  }
}

1	Read/write the keys `_id` from/to the reference document to use them in the lookup query.
2	The collection name can be read from the reference document using its key.

We know it is tempting to use all kinds of MongoDB query operators in the lookup query and this is fine. But there a few aspects to consider:

Make sure to have indexes in place that support your lookup.
Mind that resolution requires a server rountrip inducing latency, consider a lazy strategy.
A collection of document references is bulk loaded using the $or operator.
The original element order is restored in memory on a best-effort basis. Restoring the order is only possible when using equality expressions and cannot be done when using MongoDB query operators. In this case results will be ordered as they are received from the store or via the provided @DocumentReference(sort) attribute.

A few more general remarks:

Do you use cyclic references? Ask your self if you need them.
Lazy document references are hard to debug. Make sure tooling does not accidentally trigger proxy resolution by e.g. calling toString().
There is no support for reading document references using reactive infrastructure.