Book a consultation

HiveMQ Extensions and Data Hub: The Catalogue and How It Fails

HiveMQ Kafka Extension: Mapping MQTT Topics to Kafka and Back

The HiveMQ Kafka extension maps MQTT topic filters to Kafka topics and back. How the mapping works, Azure Event Hubs, and where it goes wrong.

Vladimir Romanov


The HiveMQ Kafka extension is the piece that lets a HiveMQ broker act as a producer to Kafka topics and a consumer from them. The question most plant teams ask is whether to use it or to run a separate connector between the broker and the streaming platform. This page covers how the topic mapping works, how Azure Event Hubs fits in, when the extension is the right place for the job, and what goes wrong in practice. It is current as of November 2026 and built from HiveMQ’s documentation.

It sits under Extensions and Data Hub in the HiveMQ technology hub. For why a plant would run both MQTT and Kafka at all, see Kafka vs MQTT; we do not repeat it here. Its sibling for access control is the Enterprise Security Extension.

How the HiveMQ Kafka extension maps topics

The extension implements the native Kafka protocol inside the broker. Among the features HiveMQ’s documentation lists are forwarding MQTT messages to Kafka topics, polling Kafka topics and publishing the records as MQTT messages, and buffering messages on the broker when a Kafka cluster is temporarily unavailable. HiveMQ lists it as a commercial Enterprise Extension, and its pricing page states that all Enterprise Extensions are included in every self-managed plan, as listed on 20 November 2026. See editions and pricing for the plan detail.

A cluster entry holds the bootstrap servers, optional TLS and optional authentication. A mapping points to exactly one cluster, and one cluster can serve many mappings.

MQTT topic filters mapped to Kafka topics and back Three MQTT topic filters on the left feed a mapping box inside the broker. In the upward direction the mapping writes to two Kafka topics, with the MQTT topic as the Kafka key. In the return direction a Kafka topic is polled and published as an MQTT topic. Messages are buffered on the broker if Kafka is unavailable. MQTT side Mapping in the broker Kafka side plant1/line1/+/speed plant1/line2/+/speed plant1/alarms/# Topic filters to one Kafka topic key = MQTT topic value = payload acks: ZERO, ONE, ALL kafka: line-speed kafka: alarms to Kafka kafka: setpoints Poll and republish plant1/setpoints/line1 Broker buffers when the Kafka cluster is unavailable. Topic names are illustrative.
One mapping routes one or more MQTT topic filters to one Kafka topic. The MQTT topic becomes the Kafka key, which is why topic design decides ordering.

MQTT to Kafka. Per HiveMQ’s documentation, the MQTT topic becomes the Kafka key and the MQTT payload becomes the Kafka value. The mqtt-topic-filters setting takes several filters with wildcards, so ger/+/speed and usa/+/velocity can both feed one Kafka topic. The kafka-acks setting takes ZERO, ONE (the default) or ALL.

Kafka to MQTT. By default the Kafka topic becomes the MQTT topic and the value becomes the payload. You can override both with mqtt-topics and mqtt-payload, using ${KAFKA_TOPIC}, ${KAFKA_KEY} and ${KAFKA_VALUE}, and one Kafka topic can fan out to several MQTT topics. The mqtt-publish-fields setting controls retained flag, QoS, content type and user properties. Offsets are committed after each batch, so consumption resumes after a restart or rebalance. The documentation does not claim exactly-once delivery.

Streams of glowing blue dots flowing across the frame like data on a fibre
Data in motion: the mapping decides which streams reach Kafka and which come back.

Azure Event Hubs through the Kafka API

We found no dedicated Azure Event Hubs extension in HiveMQ’s catalogue. HiveMQ’s own guide reaches Event Hubs through the Kafka extension, because Event Hubs offers a Kafka-compatible endpoint on the Standard tier and above. In the guide, the cluster points at <namespace>.servicebus.windows.net on port 9093 with TLS, authentication set to plain, the username $ConnectionString, and the namespace’s connection string as the password. An Event Hub plays the role of a Kafka topic, and the guide maps the filter test/blogpost/# to it.

Two cautions. The guide is a walkthrough dated 2 February 2022, not a support statement, and its file name (kafka-configuration.xml) differs from the current documentation, which says the file is config.xml. Check the current docs before copying either. And the shared access key in that guide is a namespace credential: keep it out of the config file by using the documented ${ENV:VARIABLE_NAME} substitution, which, per the documentation, cannot be hot-reloaded.

Extension or a separate connector

In our view the deciding question is failure isolation, not features. The extension runs inside the broker process, so a slow Kafka cluster, a bad mapping or an extension upgrade shares a JVM with the thing that serves your PLC gateways and SCADA. A separate connector puts that risk in another process you can restart and scale on its own.

Extension in the brokerSeparate connector
Fits whenThe job is topic-to-topic routing and the broker team owns both endsYou need transformation, many non-MQTT sources or a different owner for the streaming side
Failure domainShared with the brokerSeparate
Moving partsOne fewer system to runOne more process, config and credential store
LicensingHiveMQ states Enterprise Extensions are included in every self-managed plan, as listed on 20 November 2026Depends on the connector
Buffering during a Kafka outageBroker buffers, per the documentation; limits not stated in the pages we readDepends on the connector

HiveMQ’s Eli Lilly case study describes a customer using the Bridge and Kafka extensions to move data to the cloud, as HiveMQ tells it. We read that as the extension used for routing at a site boundary, which is the case we would also be comfortable with. For the broker-to-broker half of that picture, see bridging.

How the HiveMQ Kafka extension fails

These are our reading of the mechanics HiveMQ documents, plus general Kafka practice. Test each against your own configuration.

  1. Back-pressure lands on the broker. HiveMQ’s page describes buffering during an outage and a metric for the publish rate limit being exceeded, but it does not describe back-pressure as such. If Kafka slows down while PLC gateways keep publishing, something has to give. Decide what, then force that outage in a test.
  2. The key is the MQTT topic. Key-based placement keeps order per MQTT topic within a partition. A topic tree with a few very busy topics gives a few hot partitions. Putting a changing value (a batch ID) in the topic path gives you millions of keys. Design the topic tree and the partition count together.
  3. Uniform placement trades away order. The <uniform-based> strategy spreads load evenly, but the documentation describes it for when ordering is not important. Counter and state data usually needs order.
  4. Schema drift. The registry option is set per mapping: none, local (Avro and JSON Schema) or confluent (Avro only, per the page). With none, a PLC engineer who renames a tag silently changes the payload every Kafka consumer reads. The documentation also says the extension does not yet support deserialising the Kafka key, so keep keys as plain strings.
  5. Delivery settings are a choice. The default kafka-acks is ONE. Use ALL for data you cannot lose and accept the latency. The documentation does not claim exactly-once delivery, so consumers must tolerate duplicates.
  6. Echo loops. A mapping from Kafka back to MQTT on topics that another mapping sends to Kafka can feed itself. Keep inbound and outbound topic spaces disjoint.
  7. Misconfiguration at upgrade. HiveMQ 4.55 (1 September 2026) added XSD validation for the Kafka configuration. The documentation says validation is on by default, and a hot-reloaded file archives the previous version. Test the config against the new broker version before the maintenance window.

Where it gets hard

Ownership. A mapping spans two teams: the OT side that owns topics and payloads, and the data platform that owns Kafka topics and consumers. Without a named owner per mapping, schema and key changes arrive as incidents.

Security on both sides. The extension supports TLS and several SASL mechanisms per HiveMQ’s documentation, including OAUTHBEARER, Kerberos, Microsoft Entra ID and AWS IAM for MSK. Which one fits depends on your Kafka platform. Authorisation of who may publish to the mapped MQTT topics is a separate job, covered in the Enterprise Security Extension page.

Limits and open questions. We have not run this extension. We could not confirm buffer size, behaviour when the buffer fills, or throughput limits from the pages we read, and HiveMQ’s Event Hubs guide is a walkthrough rather than a support commitment. If you want an independent design review of a broker-to-streaming boundary, see our IT/OT integration work, and for broader event-driven design see event-driven architecture.

Frequently asked questions

Does HiveMQ have a Kafka extension?

Yes. HiveMQ lists a commercial Enterprise Extension for Kafka that is bidirectional: it forwards MQTT messages to Kafka topics and polls Kafka topics to publish the records as MQTT messages. HiveMQ's pricing page states that all Enterprise Extensions are included in every self-managed plan, as listed on 20 November 2026.

How does the HiveMQ Kafka extension build the Kafka key?

For MQTT-to-Kafka mappings, HiveMQ's documentation states that the MQTT topic becomes the Kafka key and the MQTT payload becomes the Kafka value. The kafka-partition-strategy setting chooses key-based placement, which keeps the order per MQTT topic within a partition, or uniform placement, which spreads load without an ordering guarantee.

Can HiveMQ send MQTT data to Azure Event Hubs?

HiveMQ publishes a guide that does it through the Kafka extension. We found no dedicated Event Hubs extension in the catalogue. The guide uses the namespace endpoint on port 9093 with TLS, and it needs the Event Hubs Standard tier or higher, because that tier offers the Kafka-compatible endpoint.

Should I use the Kafka extension or a separate MQTT-to-Kafka connector?

Use the extension when the broker is the natural place to map topics and you want one fewer system to run. Use a separate connector when you need processing, many non-MQTT sources, or you want broker and streaming platform to fail and be upgraded independently. We would decide on failure isolation, not on features.

What happens to messages if Kafka is unavailable?

HiveMQ's documentation lists buffering on the broker when a Kafka cluster is temporarily unavailable. It does not state the buffer size or what happens when it fills in the pages we read. Test an outage on your own configuration and measure what the broker does before you depend on it.

References and further reading

  1. HiveMQ Enterprise Extension for Kafka · HiveMQMappings, key handling, schema registry options, acks, security and configuration file.
  2. Connect HiveMQ and Azure Event Hubs · HiveMQThe Event Hubs setup through the Kafka extension.
  3. HiveMQ extensions catalogue · HiveMQThe Kafka extension listed as Enterprise and bidirectional.
  4. What's new in HiveMQ 4.55 · HiveMQRelease of 1 September 2026 with Kafka configuration XSD validation.
  5. HiveMQ pricing · HiveMQStates that Enterprise Extensions are included in every self-managed plan.
  6. Eli Lilly case study · HiveMQA customer using the Bridge and Kafka extensions to move data to the cloud, as HiveMQ describes it.

← Extensions and Data Hub

Have a project that needs technical ownership?

Send us the situation. If it is not something we should take on, we will say so.

The Plant-Side Brief

The weekly brief for the people who own a plant’s control systems.

One email every Wednesday for engineering and operations leaders responsible for controls, SCADA, MES and OT networks. Skim the top in two minutes; read the Deep Dive when you have fifteen.

  • One modernization problem, worked through to cost, risk and a schedule you can defend. The Deep Dive.
  • End-of-support dates before they reach your plant Windows, servers, controllers and HMI software. Lifecycle Watch.
  • A checklist your engineers can run that week not a framework to admire. Field Kit.
  • Every number sourced, no vendor agenda we are paid by plants, never by vendors.