Coming from the relational DB world, we identify objects by their relationships. A one-to-one relationship could be a person with an address. Modeling it in a relational database would most probably require two tables: a Person and an Address table with a foreign key person_id in the Address table, as shown in the following diagram:

The perfect analogy in MongoDB would be two collections, Person and Address, as shown in the following code:

> db.Person.findOne()
{
"_id" : ObjectId("590a530e3e37d79acac26a41"), "name" : "alex"
}
> db.Address.findOne()
{
"_id" : ObjectId("590a537f3e37d79acac26a42"),
"person_id" : ObjectId("590a530e3e37d79acac26a41"),
"address" : "N29DD"
}

Now, we can use the same pattern as we do in a relational database to find Person from address, as shown in the following example:

> db.Person.find({"_id": db.Address.findOne({"address":"N29DD"}).person_id})
{
"_id" : ObjectId("590a530e3e37d79acac26a41"), "name" : "alex"
}

This pattern is well known and works in the relational world.

In MongoDB, we don't have to follow this pattern, as there are more suitable ways to model these kinds of relationship.

One way in which we would typically model one-to-one or one-to-many relationships in MongoDB would be through embedding. If the person has two addresses, then the same example would then be shown in the following way: 

{ "_id" : ObjectId("590a55863e37d79acac26a43"), "name" : "alex", "address" : [ "N29DD", "SW1E5ND" ] }

Using an embedded array, we can have access to every address this user has. Embedding querying is rich and flexible so that we can store more information in each document, as shown in the following example:

{ "_id" : ObjectId("590a56743e37d79acac26a44"),
"name" : "alex",
"address" : [ { "description" : "home", "postcode" : "N29DD" },
{ "description" : "work", "postcode" : "SW1E5ND" } ] }

The advantages of this approach are as follows:

• No need for two queries across different collections
• It can exploit atomic updates to make sure that updates in the document will be all-or-nothing from the perspective of other readers of this document
• It can embed attributes in multiple nest levels creating complex structures

The most notable disadvantage is that the maximum size of the document is 16 MB so this approach cannot be used for an arbitrary, ever-growing number of attributes. Storing hundreds of elements in embedded arrays will also degrade performance.