Skip to main content

Get started with Diskless Topics Limited availability

Diskless Topics is a deployment model for Aiven for Apache Kafka® that runs in your own cloud account. It stores Kafka topic data in cloud object storage instead of broker disks, reducing infrastructure complexity and minimizing cross-availability zone (AZ) traffic.

note

Diskless Topics is in limited availability. During this phase, Aiven creates the Kafka service and sets up the required infrastructure in your cloud account. Self-service setup will be available in a future release. To request access, contact Aiven.

Prerequisites

To use Diskless Topics, you need:

  • A BYOC project on AWS (provide the project name and cloud region to Aiven)
  • Access to Diskless Topics granted by Aiven

After access is granted, Aiven:

  • Enables the required configuration (inkless) for your project
  • Provisions Diskless Topics and Aiven-managed PostgreSQL services in your AWS account
  • Configures topic storage and metadata tracking to preserve message ordering

Service provisioning

During limited availability, Aiven provisions the service for you. The setup includes:

  • Apache Kafka 4.0 with Tiered Storage and inkless enabled
  • Configuration for object storage and metadata tracking
  • KRaft (Kafka Raft metadata mode) for managing cluster metadata
  • Access credentials and connection details

Topic creation

Once the service is ready, you can:

For detailed instructions, see Create a diskless topic.

Topic creation methods

Diskless and classic topics can coexist in the same Kafka cluster.

  • When inkless is enabled for the service, topics created using the Aiven Console, Aiven API, or Aiven Provider for Terraform are created as diskless topics by default.

  • Topics created using the Aiven CLI or tools that use the Kafka Admin API must include the topic configuration inkless.enable=true to be created as diskless topics.

    Aiven CLI example to create a diskless topic:

    avn service topic-create \
    --project demo-kafka-project \
    --service-name demo-kafka-service \
    --topic exampleTopic \
    --partitions 2 \
    --replication 2 \
    --config inkless.enable=true
note

You cannot change a topic from classic to diskless, or vice versa, after it is created.

Architecture

Diskless topics separate compute and storage. Unlike traditional Kafka, they store message data in object storage.

  • Object storage holds message data.
  • Batch Coordinator tracks message batches and offsets to preserve partition ordering.
  • KRaft mode manages cluster metadata.

To learn more, see Diskless Topics architecture.

Performance

Diskless Topics are optimized for high-throughput workloads and cost efficiency:

  • Improves throughput and reduces cost by batching. See Batching and delivery
  • Has a different latency profile due to batch writes to object storage.
  • Reduces costs by minimizing cross-AZ traffic and using object storage.

You can tune batching settings to balance throughput, latency, and cost for your workload.

Limitations

The following limitations apply to diskless topics:

  • Transactions are not supported.
  • Retention policies based on time or size are not fully implemented.
  • Auto-creation of topics is not supported.
  • Compacted topics are not supported.
  • Kafka Streams state stores are not supported.
  • Share groups and queue-type topics are not supported
  • Topic type cannot be changed after creation.

Use classic topics for workloads that depend on these features.

Related pages