David A Good

Working with Heterogeneous Item Collections in the DynamoDB Enhanced Client for Java

December 07, 2020

Photo by Kevin Mueller on Unsplash

Item collections are a core concept in DynamoDB, especially for “well-designed” applications using a single table design. However, this concept doesn’t have first class support in the SDKs. In this post, I demonstrate a strategy for querying heterogeneous item collections with the AWS SDK v2 for Java’s Enhanced Client.

Item Collection Data Model

Here’s an item collection formed of a customer and their orders. This is the main dataset we will use throughout this post. I’ve explained the details of this single-table data model here.

PK SK Type CustomerId OrderId*
CUSTOMER#123 #ORDER#2020-11-25 Order 123 2020-11-25
CUSTOMER#123 #ORDER#2020-12-01 Order 123 2020-12-01
CUSTOMER#123 #ORDER#2020-12-06 Order 123 2020-12-06
CUSTOMER#123 A Customer 123

*We use an ISO-formatted date as a simple, chronologically ordered identifier

Access Patterns

The data model shown above supports the following access patterns:

  1. Get customer by id
  2. Get customer and all their orders
  3. Get customer and their most recent N orders
  4. Get just the orders for a customer

We will focus on #3. We will use a DynamoDB Query to get the customer and their most recent order in a single request. The challenge here is not with querying the item collection but with marshalling the resulting items into their respective value classes.

That is because the Enhanced Client is based on the concept of a DynamoDbTable<T>, which allows you to preform the full range DynamoDB operations (GetItem, PutItem, etc.) in an ORM sort of way. Example:

Note: The full code from these examples is available on GitHub

// Set up DynamoDbEnhancedClient...
DynamoDbTable<Customer> customerTable = 
                TABLE_NAME, TableSchema.fromClass(Customer.class));

Customer customer = new Customer();

Customer retrievedCustomer = customerTable.getItem(customer);

In DynamoDbTable<T>, T represents a single, concrete entity, like Customer.java or Order.java. These entities are typically @DynamoDbBean-annotated value classes which use other Enhanced Client annotations such as @DynamoDbPartitionKey, @DynamoDbSortKey, and @DynamoDbAttribute.
These annotations allow the SDK to dynamically read the schema for the item, e.g. TableSchema.fromClass(Customer.class) and in turn create the table abstraction, DynamoDbTable<Customer>.

However, the DynamoDbTable<T> abstraction breaks down with heterogeneous item collections where we want to query various types of entities in a single operation (in our case, a customer and one or more orders). So how can we handle this with the Enhanced Client?

For starters, we will need a DynamoDbTable for each entity in the item collection even though all the items are in the same DynamoDB table:

DynamoDbTable<Customer> customerTable = 
                TABLE_NAME, TableSchema.fromClass(Customer.class));
DynamoDbTable<Order> orderTable = 
                TABLE_NAME, TableSchema.fromClass(Order.class));

Next we need a value class which represents the item collection:

public class CustomerItemCollection {
    private Customer customer;
    private List<Order> orders = new ArrayList<>();

    public Customer getCustomer() {
        return this.customer;

    public void setCustomer(Customer customer) {
        this.customer = customer;

    public List<Order> getOrders() {
        return this.orders;

    public void addOrder(Order order) {

With that, we’re ready to query the item collection!

CustomerItemCollection getCustomerAndRecentOrders(String customerId, 
                                                 int newestOrdersCount) {
    AttributeValue customerPk = AttributeValue.builder()
    var queryRequest = QueryRequest.builder()
        // Define aliases for the Attribute, '#pk' and the value, ':pk'
        .keyConditionExpression("#pk = :pk")
        // '#pk' refers to the Attribute 'PK'
        .expressionAttributeNames(Map.of("#pk", "PK"))
        // ':pk' refers to the customer PK of interest
        .expressionAttributeValues(Map.of(":pk", customerPk))
        // Search from "bottom to top"
        // One customer, plus N newest orders
        .limit(1 + newestOrdersCount)

    // Use the DynamoDbClient directly rather than the
    // DynamoDbEnhancedClient or DynamoDbTable
    QueryResponse queryResponse = dynamoDbClient.query(queryRequest);

    // The result is a list of items in a "DynamoDB JSON map"
    List<Map<String, AttributeValue>> items = queryResponse.items();

    var customerItemCollection = new CustomerItemCollection();

    for (Map<String, AttributeValue> item : items) {

        // Every item must have a 'Type' Attribute
        AttributeValue type = item.get("Type");
        // Switch on the 'Type' and use the respective TableSchema 
        // to marshall the DynamoDB JSON into the value class
        switch (type.s()) {
            case Customer.CUSTOMER_TYPE:
                Customer customer = 
            case Order.ORDER_TYPE:
                Order order = orderTableSchema.mapToItem(item);

    return customerItemCollection;

Here’s how we use it:

CustomerItemCollection customerItemCollection =
        app.getCustomerAndRecentOrders("123", 1)
System.out.println("Result of query item collection: " + 

The output is:

Result of query item collection: CustomerItemCollection{

See the full code from these examples on GitHub.

Software engineer crafting full-stack, cloud-native solutions for enterprise. GitHub | LinkedIn | Twitter