Book a consultation

HiveMQ Platform: The Broker, the 2026 Suite and the Editions

HiveMQ Kubernetes: Deployment, Operator and Upgrades

HiveMQ Kubernetes explained: the Platform Operator and Helm charts, VM and Docker options, Java 21, LTS planning, monitoring and where it gets hard.

Vladimir Romanov


Where do you run a HiveMQ broker, and how do you keep it upgraded? HiveMQ runs on virtual machines or bare metal from a ZIP package, in Docker, and on Kubernetes through the HiveMQ Platform Operator and Helm charts, according to HiveMQ’s user guide. This page covers HiveMQ Kubernetes deployment in practice, then the version and Java planning that decide how often you touch it. It sits under the HiveMQ Platform page, which maps the product line, and the HiveMQ hub. For how the nodes behave once running, read clustering and high availability.

The deployment options

The user guide lists the packaging below. All are the same Java-based Enterprise MQTT Broker, so the choice is about who operates it.

OptionWhat HiveMQ documentsWho it suits (our view)
ZIP on Linux or WindowsUnpack and run as a serviceA plant with VMs and a team that already patches them
DockerContainer imagesA single site that wants repeatable installs without an orchestrator
Kubernetes with the Platform OperatorHelm charts, custom resource, managed scaling and upgradesA team that already runs a cluster and wants the broker to follow its practices
AWS or Azure guidesEC2 and EKS; ARM template and AKSTeams standardised on one hyperscaler
HiveMQ CloudManaged by HiveMQTeams that do not want to operate a broker at all; see Cloud vs self-hosted

In our view, Kubernetes is the right answer when the plant or the company already operates Kubernetes well. It is not a reason to start. A broker is stateful, it holds retained messages and persistent sessions, and a stateful workload on an orchestrator you do not yet run well adds a second thing to learn during an outage.

The Platform Operator and Helm charts

HiveMQ’s operator documentation says the operator installs, scales, configures and monitors HiveMQ nodes and reconciles configuration changes with the running nodes. One operator can manage several deployments, such as staging and production. It supports HiveMQ 4.28 (the previous LTS) and later. The legacy operator no longer receives support or updates as of April 2025.

Two charts do the work, both from https://hivemq.github.io/helm-charts:

  1. hivemq/hivemq-platform-operator installs the operator and the HiveMQ Platform custom resource definition.
  2. hivemq/hivemq-platform deploys a HiveMQ Platform cluster. You can install it more than once for more than one instance.

The quick start installs the operator first, then the platform, with --rollback-on-failure. The default platform install is a two-node cluster with minimal resources, which the documentation says is for testing only. For production it recommends separate namespaces for the operator and for each cluster, and points to the HiveMQ system requirements for sizing. The documented requirements are Kubernetes 1.24 or higher, Helm 3.10 or higher (including Helm 4) and kubectl. OpenShift, AKS and EKS have their own page.

Operator 2.3.0 (21 October 2026) changed how scaling behaves: the changelog describes one-pod-at-a-time scaling in both directions, with scale-down waiting for a stable topology. It also added a Helm persistence value that creates a persistent volume claim per pod, and the changelog says it is set at install only. Operator 2.3.1 (23 October 2026) is a security release; the changelog index lists it alongside Platform 4.56.1, 4.55.2 and 4.28.31, and we did not read the entry itself, so we do not describe its contents.

HiveMQ on Kubernetes: operator, broker pods, volumes and monitoring An operator namespace holds the Platform Operator. A separate platform namespace holds three broker pods, each with its own persistent volume claim. The operator creates and scales the pods. MQTT clients connect through a service in front of the pods. A Prometheus server scrapes each pod on port 9399 at path metrics. Operator namespace Platform Operator reconciles the cluster Platform namespace (one cluster) Broker pod 1 metrics :9399 Broker pod 2 metrics :9399 Broker pod 3 metrics :9399 volume claim 1 volume claim 2 volume claim 3 scales One pod at a time up and down (Operator 2.3.0) MQTT clients through a service Prometheus scrapes /metrics per pod
The operator and the broker live in separate namespaces, each broker pod holds its own volume, and clients and monitoring reach the pods from outside. The topology follows HiveMQ's documented charts; the drawing is ours.

Java, LTS and upgrade planning

Two version facts drive the plan. HiveMQ’s 4.55 release notes state that the requirements remain on Java 21 as the minimum and that Java 25 is supported and recommended. And 4.55 (1 September 2026) replaced 4.28 as the Long Term Support release. The 4.28.29 changelog, as we read it, says 4.28.x support continues through February 2027, six months after the new LTS. HiveMQ’s stated guidance is an LTS every 18 months, with monthly releases about twice a year. The current monthly release is 4.57 (3 November 2026).

ReleaseDateWhat matters for planning
4.28previous LTSSupported through February 2027 per the 4.28.29 changelog
4.551 September 2026Current LTS; Java 21 remains the minimum; Java 25 supported; Control Center v1 removed
4.573 November 2026Monthly release; $client-events payload fields renamed to camelCase, which the changelog marks as breaking

A practical upgrade routine, in our view:

  1. Read the release notes between your version and the target. The $client-events rename in 4.57 is the kind of change that silently breaks a consumer.
  2. Check the Java runtime on every VM or in every image before the broker.
  3. Test on a copy of your extensions, authentication and client behaviour, not just the broker.
  4. Move one node at a time. HiveMQ’s cluster documentation describes rolling shutdown one node at a time, with the last node down starting first, and notes that a temporary extra node helps for clusters running above 60% CPU. The upgrade guides cover the supported paths.
  5. Watch reconnect load afterwards, because clients that were on the restarted node reconnect all at once.

Monitoring and the Control Center

The monitoring page lists JMX, a Prometheus Monitoring Extension (default port 9399, path /metrics), an InfluxDB Monitoring Extension, an AWS CloudWatch extension, the Control Center and a Health API. Metric types are gauge, counter, histogram, meter and timer. Control Center v1 was removed in 4.55; The Control Center documentation includes a Data Hub view. Release 4.57 added a persistent metrics history, with 5-second snapshots, 72 hours in the diagnostic archive and about 0.5 GB per node at replica count 2. Pair this with a plant-side metrics practice; see edge and I/O for the surrounding design.

Angled close-up of syntax-highlighted code on a dark screen
Configuration and code ready to deploy: the broker, its extensions and its Helm values should all live in version control.

Where it gets hard

  • State on an orchestrator. Persistent sessions and retained messages are replicated (replica count 2 by default per the cluster documentation). Losing more nodes than replicas can make persistent data unavailable. Scaling or draining nodes without regard for that is how a message is lost. Operator 2.3.0’s one-pod-at-a-time scaling addresses part of this.
  • Install-time decisions. The persistence Helm value is set at install only, per the changelog. Treat storage as a design decision, not a later tuning step.
  • Cluster networking. The cluster guide says discovery extensions for common use cases such as dynamic discovery on AWS or Docker are available from HiveMQ, and that UDP transport is not suitable for production environments. See clustering and high availability.
  • A plant is not a cloud. A broker on a plant VM lives behind the OT network and its segmentation; see OT network segmentation. Patch windows, change control and Kubernetes upgrade cadence rarely line up.
  • Skills. Someone has to own the operator, the charts, the storage class and the broker. That ownership is the real cost of the Kubernetes route.

What we do here

We design MQTT and Unified Namespace architectures and integrate brokers with PLCs, SCADA, historians and enterprise systems as part of IT/OT projects. We are independent of HiveMQ: we have no commercial relationship with the company and do not resell its products. Our involvement is the architecture questions around a deployment, through IT/OT integration: where the broker sits in the network, what runs on a VM and what on a cluster, and how the upgrade plan fits plant change control.

Limits and open questions

  • We did not read the operator upgrade guide in full; the rolling upgrade mechanics beyond what is stated above are not described here.
  • The exact name of the custom resource and the licence options for the operator were not confirmed.
  • Docker image names and the 4.28 release date were not confirmed.
  • The contents of the 23 October 2026 security release were not read.

Next, the pricing and editions page covers what each edition costs and includes, as listed on 21 November 2026.

Frequently asked questions

How do you run HiveMQ on Kubernetes?

HiveMQ's documentation describes two Helm charts from the repository https://hivemq.github.io/helm-charts. The hivemq-platform-operator chart installs the operator and its custom resource definition. The hivemq-platform chart then deploys a HiveMQ Platform cluster, which the operator manages. The default install is a two-node cluster meant for testing only.

Which Java version does HiveMQ need?

HiveMQ's 4.55 release notes (1 September 2026) state that the system requirements remain on Java 21 as the minimum, and that Java 25 is supported and recommended. If you run an older HiveMQ release on an older Java, check the release notes for your version before upgrading, because the runtime requirement moves with the release.

Which HiveMQ version should a plant run: the LTS or the monthly release?

HiveMQ 4.55 (1 September 2026) is the current Long Term Support release and 4.28 was the previous one. HiveMQ's guidance is an LTS roughly every 18 months, with monthly releases about twice a year. In our view a plant that changes slowly should follow the LTS and plan the move from 4.28 before support ends.

How do you monitor a HiveMQ broker?

HiveMQ's documentation lists JMX, a Prometheus Monitoring Extension (default port 9399, path /metrics), an InfluxDB extension, an AWS CloudWatch extension, the Control Center and a Health API. The Control Center v1 was removed in 4.55, so any dashboard built on it needs the current Control Center interface or an external metrics stack.

Can I run HiveMQ without Kubernetes?

Yes. HiveMQ's user guide lists a ZIP package for Linux and Windows, Docker images, and guides for AWS (EC2, EKS) and Azure (ARM template, AKS). Kubernetes is one option, not a requirement. A small broker on one or two virtual machines is a legitimate design for a plant.

References and further reading

  1. HiveMQ Platform Operator for Kubernetes · HiveMQHelm charts, requirements, quick start and upgrade guide.
  2. HiveMQ Platform Operator 2.3.0 release · HiveMQOne-pod-at-a-time scaling and the persistence value.
  3. What's new in HiveMQ 4.55 · HiveMQThe LTS release: Java 21 minimum, Java 25 support, Control Center v1 removed.
  4. Managing HiveMQ releases and support · HiveMQThe LTS and monthly release guidance.
  5. HiveMQ user guide: monitoring · HiveMQ
  6. HiveMQ user guide: cluster · HiveMQReplica count, rolling shutdown and recovery.

← HiveMQ Platform

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.