How to aggregate Kafka messages over tumbling windows in Java using Flink's Table API for Confluent Cloud

In this tutorial, you will learn how to aggregate messages over tumbling windows using Flink's Table API. Tumbling windows are fixed-size, non-overlapping, and contiguous time intervals.

Prerequisites

• A Confluent Cloud account
• The Confluent CLI installed on your machine
• Java 21, e.g., follow the OpenJDK installation instructions here if you don't have Java. Validate that java -version shows version 21.
• Clone the confluentinc/tutorials GitHub repository (if you haven't already) and navigate to the tutorials directory:

  git clone git@github.com:confluentinc/tutorials.git
  cd tutorials

Provision Confluent Cloud infrastructure

If you already have the Confluent Cloud resources required to populate a Table API client configuration file, e.g., from running a different tutorial, you may skip to the next step after creating or copying the properties file as documented here to tumbling-windows/flink_table_api_java/src/main/resources/cloud.properties within the top-level tutorials directory.

If you need to create the Confluent Cloud infrastructure needed to run this tutorial, the confluent-flink-quickstart CLI plugin creates the resources that you need to get started with Confluent Cloud for Apache Flink. Install it by running:

confluent plugin install confluent-flink-quickstart

Run the plugin as follows to create the Confluent Cloud resources needed for this tutorial and generate a Table API client configuration file. Note that you may specify a different cloud provider (gcp or azure) or region. You can find supported regions in a given cloud provider by running confluent flink region list --cloud <CLOUD>.

confluent flink quickstart \
    --name flink_table_api_tutorials \
    --max-cfu 10 \
    --region us-east-1 \
    --cloud aws \
    --table-api-client-config-file ./tumbling-windows/flink_table_api_java/src/main/resources/cloud.properties

The plugin should complete in under a minute and will generate a properties file as documented here.

Inspect the code

Dependencies

Before digging into Java source code, first check out the two dependencies required to use the Flink Table API for Confluent Cloud. These are defined in the dependencies section of the tumbling-windows/flink_table_api_java/build.gradle file. (For Maven, see the analogous pom.xml snippet here.)

• org.apache.flink:flink-table-api-java: This is the Apache Flink Table API implementation dependency. It contains, for example, the classes implementing the Table API DSL (domain-specific language).
• io.confluent.flink:confluent-flink-table-api-java-plugin: This dependency contains the "glue" for instantiating an Apache Flink Table API table environment against Confluent Cloud (i.e., an implementation of the org.apache.flink.table.api.TableEnvironment interface), as well as other helper utilities that we will use in this tutorial.

Java source

Take a look at the source code in tumbling-windows/flink_table_api_java/FlinkTableApiTumblingWindows.java. These two lines instantiate a table environment for executing Table API programs against Confluent Cloud:

EnvironmentSettings envSettings = ConfluentSettings.fromResource("/cloud.properties");
TableEnvironment tableEnv = TableEnvironment.create(envSettings);

Let's aggregate one of Confluent Cloud's example tables. You can find these tables in the read-only marketplace database of the examples catalog. The source code in this example uses the Table API's Table.window and GroupWindowedTable.groupBy methods to aggregate over 2 second windows. The aggregation is a simple count().

TableResult tableResult = tableEnv.from("examples.marketplace.orders")
    .window(
            Tumble.over(lit(2).seconds())
                    .on($("$rowtime"))
                    .as("window")
    ).groupBy(
            $("window")
    ).select(
            $("customer_id").count().as("count"),
            $("window").start().as("window_start"),
            $("window").end().as("window_end")
    )
    .execute();

Given the table result, we can then materialize (in memory) the rows in the resulting stream by calling ConfluentTools.collectMaterialized or ConfluentTools.printMaterialized. This line materializes and prints 2 rows from the table result:

ConfluentTools.printMaterialized(tableResult, 2);

Alternatively, we can use the Table API's TableResult interface directly to collect rows. For example, to print the count for one window:

try (CloseableIterator<Row> it = tableResult.collect()) {
    if (it.hasNext()) {
        Row row = it.next();
        System.out.println(row.getField("count"));
        System.out.println(row.getField("window_start"));
        System.out.println(row.getField("window_end"));
    }
}

Run the program

You can run the example program directly in your IDE by opening the Gradle project located at tumbling-windows/flink_table_api_java/, or via the command line from the top-level tutorials directory:

./gradlew tumbling-windows:flink_table_api_java:run

The program will output 2 rows materialized via printMaterialized, and then an additional count and window start and end. Note that the same TableResult (and its underlying iterator) is used, so the last window that is printed comes right after the first two windows printed.

+-------+-------------------------+-------------------------+
| count |            window_start |              window_end |
+-------+-------------------------+-------------------------+
|    51 | 2025-03-03 10:28:18.000 | 2025-03-03 10:28:20.000 |
|   102 | 2025-03-03 10:28:20.000 | 2025-03-03 10:28:22.000 |
+-------+-------------------------+-------------------------+
2 rows in set
100
2025-03-03T10:28:22
2025-03-03T10:28:24

Tear down Confluent Cloud infrastructure

When you are done, be sure to clean up any Confluent Cloud resources created for this tutorial. Since you created all resources in a Confluent Cloud environment, you can simply delete the environment and most of the resources created for this tutorial (e.g., the Kafka cluster and Flink compute pool) will be deleted. Run the following command in your terminal to get the environment ID of the form env-123456 corresponding to the environment named flink_table_api_tutorials_environment:

confluent environment list

Delete the environment:

confluent environment delete <ENVIRONMENT_ID>

Next, delete the Flink API key. This API key isn't associated with the deleted environment so it needs to be deleted separately. Find the key:

confluent api-key list --resource flink --current-user

And then copy the 16-character alphanumeric key and delete it:

confluent api-key delete <KEY>

Finally, for the sake of housekeeping, delete the Table API client configuration file:

rm tumbling-windows/flink_table_api_java/src/main/resources/cloud.properties