Skip to main content
Version: 3.3.x

Message deduplication

Message deduplication could affect the performance of the brokers during informational snapshots.

To use message deduplication in Pulsar, you need to configure your Pulsar brokers, namespaces, or topics. It is recommended to modify the configuration in the clients, for example, setting send timeout to infinity.

How message deduplication works​

You can enable or disable message deduplication at broker, namespace, or topic level. By default, it is disabled on all brokers, namespaces, or topics. You can enable it in the following ways:

  • Enable deduplication for all namespaces/topics at the broker level.
  • Enable deduplication for a specific namespace with the pulsar-admin namespaces interface.
  • Enable deduplication for a specific topic with the pulsar-admin topics interface.

Configure message deduplication​

You can configure message deduplication in Pulsar using the broker.conf configuration file. The following deduplication-related parameters are available.

ParameterDescriptionDefault
brokerDeduplicationEnabledSets the default behavior for message deduplication in the Pulsar broker. If it is set to true, message deduplication is enabled on all namespaces/topics. If it is set to false, you have to enable or disable deduplication at the namespace level or the topic level.false
brokerDeduplicationMaxNumberOfProducersThe maximum number of producers for which information is stored for deduplication purposes.10000
brokerDeduplicationEntriesIntervalThe number of entries after which a deduplication informational snapshot is taken. A larger interval leads to fewer snapshots being taken, though this lengthens the topic recovery time (the time required for entries published after the snapshot to be replayed).1000
brokerDeduplicationSnapshotIntervalSecondsThe time period after which a deduplication informational snapshot is taken. It runs simultaneously with brokerDeduplicationEntriesInterval.120
brokerDeduplicationProducerInactivityTimeoutMinutesThe time of inactivity (in minutes) after which the broker discards deduplication information related to a disconnected producer.360 (6 hours)

Set default value at the broker-level​

By default, message deduplication is disabled on all Pulsar namespaces/topics. To enable it on all namespaces/topics, set the brokerDeduplicationEnabled parameter to true and restart the broker.

Even if you set the value for brokerDeduplicationEnabled, enabling or disabling via Pulsar admin CLI overrides the default settings at the broker level.

Enable message deduplication at namespace or topic level​

Though message deduplication is disabled by default at the broker level, you can enable message deduplication for a specific namespace or topic using the pulsar-admin namespaces set-deduplication or the pulsar-admin topics set-deduplication command. You can use the --enable/-e flag and specify the namespace/topic.

The following example shows how to enable message deduplication at the namespace level.

bin/pulsar-admin namespaces set-deduplication \
public/default \
--enable # or just -e

Disable message deduplication at namespace or topic level​

Even if you enable message deduplication at the broker level, you can disable message deduplication for a specific namespace or topic using the pulsar-admin namespace set-deduplication or the pulsar-admin topics set-deduplication command. Use the --disable/-d flag and specify the namespace/topic.

The following example shows how to disable message deduplication at the namespace level.

bin/pulsar-admin namespaces set-deduplication \
public/default \
--disable # or just -d

Pulsar clients​

If you enable message deduplication in Pulsar brokers, namespaces, or topics, it is recommended to make the client retry infinitely the messages until it succeeds, otherwise it is possible to break the ordering guarantee as some requests may time out and the application does not know whether the request is successfully added to the topic or not.

So you need to complete the following tasks for your client producers:

  1. Specify a name for the producer (this is a requirement, Pulsar will use the producer name to filter duplicated messages).
  2. Set the message timeout to 0 (namely, no timeout).

The instructions for Java, Python, and C++ clients are different.

To ensure the guarantee order on a Java producer sending to a topic with message deduplication enabled, set the producer name using the producerName setter, and set the timeout to 0 using the sendTimeout setter.

import org.apache.pulsar.client.api.Producer;
import org.apache.pulsar.client.api.PulsarClient;
import java.util.concurrent.TimeUnit;

PulsarClient pulsarClient = PulsarClient.builder()
.serviceUrl("pulsar://localhost:6650")
.build();
Producer producer = pulsarClient.newProducer()
.producerName("producer-1")
.topic("persistent://public/default/topic-1")
.sendTimeout(0, TimeUnit.SECONDS)
.create();