How to find distinct values in a stream of events


How can I filter out duplicate events from a Kafka topic based on a field in the event, producing a new stream of unique events per time window?

Edit this page

Example use case:

Consider a topic with events that represent clicks on a website. Each event contains an IP address, a URL, and a timestamp. In this tutorial, we'll write a program that filters click events by the IP address within a window of time.

Code example:

Try it

Initialize the project

To get started, make a new directory anywhere you’d like for this project:

mkdir distinct-events && cd distinct-events

Next, create a directory for configuration data:

mkdir configuration

Sign up for Confluent Cloud and provision resources

Sign up for Confluent Cloud, a fully-managed Apache Kafka service. Then provision your resources:

  1. After you log in to Confluent Cloud, click on Add cloud environment and name the environment learn-kafka. Using a new environment keeps your learning resources separate from your other Confluent Cloud resources.

  2. From the Billing & payment section in the Menu, apply the promo code CC100KTS to receive an additional $100 free usage on Confluent Cloud (details).

  3. Click on LEARN and follow the instructions to launch a Kafka cluster and to enable Schema Registry.

Confluent Cloud

Create a properties file with Confluent Cloud information

From the Confluent Cloud UI, navigate to your Kafka cluster and click on Clients and then select Java.

Create new credentials for your Kafka cluster and Schema Registry, and then Confluent Cloud will show a configuration similar to below with your new credentials automatically populated (make sure show API keys is checked). Copy and paste it into a configuration/ file on your machine.

# Required connection configs for Kafka producer, consumer, and admin
bootstrap.servers={{ BOOTSTRAP_SERVERS }}
security.protocol=SASL_SSL   required username='{{ CLUSTER_API_KEY }}'   password='{{ CLUSTER_API_SECRET }}';
# Required for correctness in Apache Kafka clients prior to 2.6

# Best practice for Kafka producer to prevent data loss

# Required connection configs for Confluent Cloud Schema Registry
schema.registry.url={{ SR_URL }}
basic.auth.credentials.source=USER_INFO{{ SR_API_KEY }}:{{ SR_API_SECRET }}
Do not directly copy and paste the above configuration. You must copy it from the UI so that it includes your Confluent Cloud information and credentials.

Download and setup the Confluent Cloud CLI

Instructions for setting up Confluent Cloud CLI is available from within the Confluent Cloud UI. Navigate to your Kafka cluster, click on the CLI and tools section, and run through the steps in the CCloud CLI tab.

Configure the project

We’ll use Gradle as our build system. Create a file named build.gradle with the following contents (on macOS pbpaste > build.gradle after copying the below):

buildscript {
  repositories {
  dependencies {
    classpath "com.commercehub.gradle.plugin:gradle-avro-plugin:0.22.0"
    classpath "com.github.jengelman.gradle.plugins:shadow:4.0.2"

plugins {
  id "java"
  id "" version "2.8.0"

sourceCompatibility = "1.8"
targetCompatibility = "1.8"
version = "0.0.1"

repositories {

  maven {
    url ""

apply plugin: "com.commercehub.gradle.plugin.avro"
apply plugin: "com.github.johnrengelman.shadow"

dependencies {
  implementation "org.apache.avro:avro:1.10.2"
  implementation "org.slf4j:slf4j-simple:1.7.30"
  implementation "org.apache.kafka:kafka-streams:2.7.0"
  implementation "io.confluent:kafka-streams-avro-serde:6.1.1"
  testImplementation "org.apache.kafka:kafka-streams-test-utils:2.7.0"
  testImplementation "junit:junit:4.13.2"

test {
  testLogging {
    outputs.upToDateWhen { false }
    showStandardStreams = true
    exceptionFormat = "full"

jar {
  manifest {
        "Class-Path": configurations.compileClasspath.collect { it.getName() }.join(" "),
        "Main-Class": "io.confluent.developer.FindDistinctEvents"

shadowJar {
  archiveBaseName = "kstreams-find-distinct-standalone"
  archiveClassifier = ''

And be sure to run the following command to obtain the Gradle wrapper:

gradle wrapper

Then create a development configuration file at configuration/


Update the properties file with Confluent Cloud information

Using the command below, append the contents of configuration/ (with your Confluent Cloud configuration) to configuration/ (with the application properties).

cat configuration/ >> configuration/

Create a schema for the events

Create a directory for the schemas that represent the events in the stream:

mkdir -p src/main/avro

Then create the following Avro schema file at src/main/avro/click.avsc for the click events (on macOS, copy the text then execute in the terminal: pbpaste > src/main/avro/click.avsc):

  "namespace": "io.confluent.developer.avro",
  "type": "record",
  "name": "Click",
  "fields": [
    {"name": "ip", "type": "string"},
    {"name": "url", "type": "string"},
    {"name": "timestamp", "type": "string"}

Because this Avro schema is used in the Java code, it needs to compile it. Run the following:

./gradlew build

Create the Kafka Streams topology

Create a directory for the Java files in this project:

mkdir -p src/main/java/io/confluent/developer

Then create the following file at src/main/java/io/confluent/developer/

Focusing on the buildTopology method, note how the Kafka Streams topology relies on a ValueTransformerWithKey and a Window Store to filter out the duplicate IP addresses. Events are de-duped within a 2 minute window, and unique clicks are produced to a new topic named distinct-clicks.

Note that we are using ValueTransformerWithKey here instead of Transformer since we need keys to transform data, but there is no need to re-key the stream. Tranformer usage will lead to redundant repartitioning when grouping operations are used afterwards.

package io.confluent.developer;

import org.apache.kafka.clients.admin.AdminClient;
import org.apache.kafka.clients.admin.NewTopic;
import org.apache.kafka.common.serialization.Serdes;
import org.apache.kafka.streams.KafkaStreams;
import org.apache.kafka.streams.StreamsBuilder;
import org.apache.kafka.streams.StreamsConfig;
import org.apache.kafka.streams.Topology;
import org.apache.kafka.streams.kstream.Consumed;
import org.apache.kafka.streams.kstream.KeyValueMapper;
import org.apache.kafka.streams.kstream.Produced;
import org.apache.kafka.streams.kstream.ValueTransformerWithKey;
import org.apache.kafka.streams.processor.ProcessorContext;
import org.apache.kafka.streams.state.StoreBuilder;
import org.apache.kafka.streams.state.Stores;
import org.apache.kafka.streams.state.WindowStore;
import org.apache.kafka.streams.state.WindowStoreIterator;

import java.time.Duration;
import java.util.ArrayList;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Properties;
import java.util.concurrent.CountDownLatch;

import io.confluent.common.utils.TestUtils;
import io.confluent.developer.avro.Click;
import io.confluent.kafka.streams.serdes.avro.SpecificAvroSerde;

import static io.confluent.kafka.serializers.AbstractKafkaSchemaSerDeConfig.SCHEMA_REGISTRY_URL_CONFIG;

public class FindDistinctEvents {

    private static final String storeName = "eventId-store";

     * Discards duplicate click events from the input stream by ip address
     * <p>
     * Duplicate records are detected based on ip address
     * The transformer remembers known ip addresses within an associated window state
     * store, which automatically purges/expires IPs from the store after a certain amount of
     * time has passed to prevent the store from growing indefinitely.
     * <p>
     * Note: This code is for demonstration purposes and was not tested for production usage.
    private static class DeduplicationTransformer<K, V, E> implements ValueTransformerWithKey<K, V, V> {

        private ProcessorContext context;

         * Key: ip address
         * Value: timestamp (event-time) of the corresponding event when the event ID was seen for the
         * first time
        private WindowStore<E, Long> eventIdStore;

        private final long leftDurationMs;
        private final long rightDurationMs;

        private final KeyValueMapper<K, V, E> idExtractor;

         * @param maintainDurationPerEventInMs how long to "remember" a known ip address
         *                                     during the time of which any incoming duplicates
         *                                     will be dropped, thereby de-duplicating the
         *                                     input.
         * @param idExtractor                  extracts a unique identifier from a record by which we de-duplicate input
         *                                     records; if it returns null, the record will not be considered for
         *                                     de-duping but forwarded as-is.
        DeduplicationTransformer(final long maintainDurationPerEventInMs, final KeyValueMapper<K, V, E> idExtractor) {
            if (maintainDurationPerEventInMs < 1) {
                throw new IllegalArgumentException("maintain duration per event must be >= 1");
            leftDurationMs = maintainDurationPerEventInMs / 2;
            rightDurationMs = maintainDurationPerEventInMs - leftDurationMs;
            this.idExtractor = idExtractor;

        public void init(final ProcessorContext context) {
            this.context = context;
            eventIdStore = (WindowStore<E, Long>) context.getStateStore(storeName);

        public V transform(final K key, final V value) {
            final E eventId = idExtractor.apply(key, value);
            if (eventId == null) {
                return value;
            } else {
                final V output;
                if (isDuplicate(eventId)) {
                    output = null;
                    updateTimestampOfExistingEventToPreventExpiry(eventId, context.timestamp());
                } else {
                    output = value;
                    rememberNewEvent(eventId, context.timestamp());
                return output;

        private boolean isDuplicate(final E eventId) {
            final long eventTime = context.timestamp();
            final WindowStoreIterator<Long> timeIterator = eventIdStore.fetch(
                    eventTime - leftDurationMs,
                    eventTime + rightDurationMs);
            final boolean isDuplicate = timeIterator.hasNext();
            return isDuplicate;

        private void updateTimestampOfExistingEventToPreventExpiry(final E eventId, final long newTimestamp) {
            eventIdStore.put(eventId, newTimestamp, newTimestamp);

        private void rememberNewEvent(final E eventId, final long timestamp) {
            eventIdStore.put(eventId, timestamp, timestamp);

        public void close() {
            // Note: The store should NOT be closed manually here via `eventIdStore.close()`!
            // The Kafka Streams API will automatically close stores when necessary.


    private SpecificAvroSerde<Click> buildClicksSerde(final Properties allProps) {
        final SpecificAvroSerde<Click> serde = new SpecificAvroSerde<>();
        final Map<String, String> config = (Map)allProps;
        serde.configure(config, false);
        return serde;

    public Topology buildTopology(Properties allProps,
                                  final SpecificAvroSerde<Click> clicksSerde) {
        final StreamsBuilder builder = new StreamsBuilder();

        final String inputTopic = allProps.getProperty("");
        final String outputTopic = allProps.getProperty("");

        // How long we "remember" an event.  During this time, any incoming duplicates of the event
        // will be, well, dropped, thereby de-duplicating the input data.
        // The actual value depends on your use case.  To reduce memory and disk usage, you could
        // decrease the size to purge old windows more frequently at the cost of potentially missing out
        // on de-duplicating late-arriving records.
        final Duration windowSize = Duration.ofMinutes(2);

        // retention period must be at least window size -- for this use case, we don't need a longer retention period
        // and thus just use the window size as retention time
        final Duration retentionPeriod = windowSize;

        final StoreBuilder<WindowStore<String, Long>> dedupStoreBuilder = Stores.windowStoreBuilder(


                .stream(inputTopic, Consumed.with(Serdes.String(), clicksSerde))
                .transformValues(() -> new DeduplicationTransformer<>(windowSize.toMillis(), (key, value) -> value.getIp()), storeName)
                .filter((k, v) -> v != null)
                .to(outputTopic, Produced.with(Serdes.String(), clicksSerde));


    public void createTopics(Properties allProps) {
        AdminClient client = AdminClient.create(allProps);

        List<NewTopic> topics = new ArrayList<>();
        topics.add(new NewTopic(
        topics.add(new NewTopic(


    public static Properties loadEnvProperties(String fileName) throws IOException {
        Properties allProps = new Properties();
        FileInputStream input = new FileInputStream(fileName);

        return allProps;

    public static void main(String[] args) throws IOException {
        if (args.length < 1) {
            throw new IllegalArgumentException(
                    "This program takes one argument: the path to an environment configuration file.");

        new FindDistinctEvents().runRecipe(args[0]);

    private void runRecipe(final String configPath) throws IOException {
        final Properties allProps = new Properties();
        try (InputStream inputStream = new FileInputStream(configPath)) {
        allProps.put(StreamsConfig.APPLICATION_ID_CONFIG, allProps.getProperty(""));
        allProps.put(StreamsConfig.STATE_DIR_CONFIG, TestUtils.tempDirectory().getPath());

        final Topology topology = this.buildTopology(allProps, this.buildClicksSerde(allProps));


        final KafkaStreams streams = new KafkaStreams(topology, allProps);
        final CountDownLatch latch = new CountDownLatch(1);

        // Attach shutdown handler to catch Control-C.
        Runtime.getRuntime().addShutdownHook(new Thread("streams-shutdown-hook") {
            public void run() {

        try {
        } catch (Throwable e) {


Compile and run the Kafka Streams program

In your terminal, run:

./gradlew shadowJar

Now that an uberjar for the Kafka Streams application has been built, you can launch it locally. When you run the following, the prompt won’t return, because the application will run until you exit it (Ctrl+C). Run the application in a separate terminal in order to see any output as well as continue with the remaining instructions in the current terminal. The Kafka Streams application should produce a log entry similar to this State transition from REBALANCING to RUNNING to indicate it’s functioning correctly.

java -jar build/libs/kstreams-find-distinct-standalone-0.0.1.jar configuration/

Produce sample click events to the input topic

In a new terminal window, run the following command to start a Confluent Cloud CLI producer:

ccloud kafka topic produce clicks \
  --parse-key \
  --delimiter ":" \
  --value-format avro \
  --schema src/main/avro/click.avsc

You will be prompted for the Confluent Cloud Schema Registry credentials as shown below, which you can find in the configuration/ configuration file. Look for the configuration parameter, whereby the ":" is the delimiter between the key and secret.

Enter your Schema Registry API key:
Enter your Schema Registry API secret:

When the producer starts, it will log some messages and hang, waiting for your input. Each line represents input data for the Kafka Streams application. To send all of the events below, paste the following into the prompt and press enter:


Enter Ctrl+C to exit.

In the next steps we will run a consumer to observe the distinct click events. You can experiment with various orderings of the records in order to observe what makes a click event distinct. By default the distinct event window store looks for distinct clicks over a 2-minute duration.

Consume distinct events from the output topic

Run the following command to start a Confluent Cloud CLI consumer to view the distinct click events:

ccloud kafka topic consume distinct-clicks -b --value-format avro

Depending on the cadence and values you produce in the steps above, you should see messages similar to the following:


Enter Ctrl+C to exit.

Teardown Confluent Cloud resources

You may try another Kafka tutorial, but if you don’t plan on doing other tutorials, use the Confluent Cloud UI or CLI to destroy all the resources you created. Verify they are destroyed to avoid unexpected charges.

Test it

Create a test configuration file

First, create a test file at configuration/

Write a test

Then, create a directory for the tests to live in:

mkdir -p src/test/java/io/confluent/developer

Create the following test file at src/test/java/io/confluent/developer/ Testing a Kafka Streams application requires a bit of test harness code, but happily the org.apache.kafka.streams.TopologyTestDriver class makes this much more pleasant than it would otherwise be.

There is a test method in FindDistinctEventsTest annotated with @Test: shouldFilterDistinctEvents() which follows the common Arrange Act Assert (AAA) pattern. This is a simple method that runs our Streams topology using the TopologyTestDriver and some mocked data that is set up inside the test method.

package io.confluent.developer;

import org.apache.kafka.common.serialization.Deserializer;
import org.apache.kafka.common.serialization.Serdes;
import org.apache.kafka.common.serialization.Serializer;
import org.apache.kafka.streams.StreamsConfig;
import org.apache.kafka.streams.TestInputTopic;
import org.apache.kafka.streams.Topology;
import org.apache.kafka.streams.TopologyTestDriver;
import org.junit.Assert;
import org.junit.Test;

import java.nio.file.Path;
import java.nio.file.Paths;
import java.util.HashMap;
import java.util.List;
import java.util.Map;
import java.util.Objects;
import java.util.Properties;

import io.confluent.developer.avro.Click;
import io.confluent.kafka.streams.serdes.avro.SpecificAvroSerde;

import static java.util.Arrays.asList;

public class FindDistinctEventsTest {

  private final static String TEST_CONFIG_FILE = "configuration/";
  private final static Path STATE_DIR =
      Paths.get(System.getProperty("user.dir"), "build");

  private final Properties allProps;

  public FindDistinctEventsTest() throws IOException {
    allProps = FindDistinctEvents.loadEnvProperties(TEST_CONFIG_FILE);
    allProps.put(StreamsConfig.STATE_DIR_CONFIG, STATE_DIR.toString());

  private static SpecificAvroSerde<Click> makeSerializer(Properties allProps) {
    SpecificAvroSerde<Click> serde = new SpecificAvroSerde<>();

    Map<String, String> config = new HashMap<>();
    config.put("schema.registry.url", allProps.getProperty("schema.registry.url"));
    serde.configure(config, false);

    return serde;

  public void shouldFilterDistinctEvents() {

    final FindDistinctEvents distinctifier = new FindDistinctEvents();

    String inputTopic = allProps.getProperty("");
    String outputTopic = allProps.getProperty("");

    final SpecificAvroSerde<Click> clickSerde = makeSerializer(allProps);

    Topology topology = distinctifier.buildTopology(allProps, clickSerde);
    final List<Click> expectedOutput;
    List<Click> actualOutput;
    try (TopologyTestDriver testDriver = new TopologyTestDriver(topology, allProps)) {

      Serializer<String> keySerializer = Serdes.String().serializer();

      final List<Click> clicks = asList(
          new Click("",
          new Click("",
          new Click("",
          new Click("",
          new Click("",
          new Click("",

      final TestInputTopic<String, Click>
          testDriverInputTopic =
          testDriver.createInputTopic(inputTopic, keySerializer, clickSerde.serializer());

      clicks.forEach(clk -> testDriverInputTopic.pipeInput(clk.getIp(), clk));

      expectedOutput = asList(clicks.get(0), clicks.get(1), clicks.get(2));

      Deserializer<String> keyDeserializer = Serdes.String().deserializer();
      actualOutput =
          testDriver.createOutputTopic(outputTopic, keyDeserializer, clickSerde.deserializer()).readValuesToList()

    Assert.assertEquals(expectedOutput, actualOutput);

Invoke the tests

Now run the test, which is as simple as:

./gradlew test

Take it to production

Create a production configuration file

First, create a new configuration file at configuration/ with the following content. Be sure to fill in the addresses of your production hosts and change any other parameters that make sense for your setup.
bootstrap.servers=<< FILL ME IN >>
schema.registry.url=<< FILL ME IN >>
input.topic.partitions=<< FILL ME IN >>
input.topic.replication.factor=<< FILL ME IN >>
output.topic.partitions=<< FILL ME IN >>
output.topic.replication.factor=<< FILL ME IN >>

Build a Docker image

In your terminal, execute the following to invoke the Jib plugin to build an image:

gradle jibDockerBuild --image=io.confluent.developer/kstreams-find-distinct:0.0.1

Launch the container

Finally, launch the container using your preferred container orchestration service. If you want to run it locally, you can execute the following:

docker run -v $PWD/configuration/ io.confluent.developer/kstreams-find-distinct:0.0.1