Book a consultation

HiveMQ Manufacturing: Plant Placement, Customers and Failure Modes

HiveMQ and Ignition: Cirrus Link Modules, STATE and Failure Modes

HiveMQ Ignition setup explained: MQTT Transmission and Engine with a HiveMQ broker, where the broker sits, Sparkplug STATE, buffering and how it fails.

Vladimir Romanov


HiveMQ with Ignition means an Ignition gateway running Cirrus Link’s MQTT Transmission module publishes Sparkplug B data to a HiveMQ broker, and another gateway running MQTT Engine reads it back as tags. The question this page answers is what a controls engineer must decide for that to survive a bad day: where the broker sits, how STATE and the primary host behave, where data is buffered, and which permissions are easy to miss. It is an independent read of HiveMQ’s and Inductive Automation’s documentation as of November 2026, under HiveMQ manufacturing. It makes no claim that Joltek has deployed HiveMQ.

What is documented about HiveMQ and Ignition

HiveMQ joined the Ignition Technology Ecosystem Program on 23 July 2025 according to HiveMQ’s announcement of that date. The announcement makes no Sparkplug or module claim, so it tells you about the relationship, not the engineering. The engineering evidence is HiveMQ’s guide of 20 June 2025, written by Aaron Franz. It says to upload the MQTT Engine and MQTT Transmission .modl files, create server sets with the same Primary Host ID, and enable Primary Host so that STATE works. The guide enables the default in-memory store and names no software versions beyond Ignition 8.1.43 in a module download link. Treat it as a bench recipe, not a production design.

Inductive Automation’s documentation says the broker MQTT Engine talks to can be “an Ignition Gateway with the MQTT Distributor Module installed, or a third-party server”. An old Inductive Automation forum thread claimed Distributor was the only broker that worked, and later replies disagree. We treat that as unsettled and test the exact module and broker versions together.

Where the broker should sit

Operator seats facing wall monitors and a laptop that show map-based data screens
A monitoring station: the screens only stay current if the path from the line to this room survives a broker or network failure.

In our view, put the broker at the plant, not only in a central data centre. The reasons are plant consequences. A Transmission module that cannot reach its broker must buffer, and the buffer is finite. A broker on the plant network keeps the plant’s own SCADA, historian and alarms fed even when the WAN is down. Enterprise consumers then get a bridged copy. Siemens’ Architecture Hub page for HiveMQ shows the same shape, naming Ignition among the HMI and SCADA examples that read from a local broker per factory. That page is marketing, not a measured result.

Two placements are common:

  1. One Ignition gateway per plant, broker at the plant, Engine at the plant. Plant tags stay local. A second Engine at the enterprise reads bridged topics if it needs them.
  2. Transmission at each plant, one central broker and Engine. Simpler to run, but the WAN is now part of the control data path.

The multi-site trade-off is on multi-site architecture. For Ignition itself, connecting Allen-Bradley PLCs to Ignition covers the tag side, and SCADA and historians the wider platform choice.

Ignition with a HiveMQ broker: data up, STATE down, buffering at the edge Three boxes in a row. An Ignition gateway with Cirrus Link MQTT Transmission publishes Sparkplug birth, death and data messages to a HiveMQ broker, which delivers them to a second Ignition gateway running MQTT Engine as primary host. In the other direction Engine publishes a retained STATE message that Transmission subscribes to. Below, three numbered steps describe store and forward when the broker or network is unavailable. Ignition gateway at the plant Cirrus Link MQTT Transmission HiveMQ broker retains STATE enforces topic ACLs Ignition gateway (central or plant) Cirrus Link MQTT Engine primary host NBIRTH, DBIRTH, NDATA, NDEATH (spBv1.0/...) STATE: Engine publishes online or offline, retained; Transmission waits for it If the broker or WAN is down: 1 Transmission cannot publish. With a history store enabled it switches to store and forward. 2 Buffered data lives in the Transmission history store on the Ignition gateway, not in the broker. 3 When the link returns, the buffered history is forwarded and live data resumes. If Engine goes offline: its last will flips STATE to offline and Transmitters buffer the same way. Order of events as HiveMQ's guide and Sparkplug describe them. Not measured durations.
Two flows cross the broker, data one way and STATE the other. Buffering during an outage is the Transmission history store's job, not the broker's.

Sparkplug STATE and primary host behaviour

STATE is how nodes know a host is alive. HiveMQ’s operational-behaviour post (12 April 2024) describes it generically for Sparkplug, not as a HiveMQ feature. A host application subscribes to spBv1.0/# and publishes its own STATE message. Edge nodes subscribe to STATE to track the primary host. A node with a primary host waits for it to be online before publishing NBIRTH and DBIRTH, and holds data while it is offline. If a host loses its MQTT connection, metrics from known nodes are marked STALE.

HiveMQ’s Ignition guide adds the Cirrus Link specifics: Engine has a last will and publishes a retained online status, so a Transmitter that connects later still learns the host’s state. If Engine goes offline, Transmitters with the same Primary Host ID switch to store and forward when a history store is enabled.

Two details matter on the broker side:

  • The broker must keep retained messages, which STATE depends on.
  • HiveMQ’s Sparkplug Aware extension does not touch STATE. It is open source under Apache 2.0, republishes birth messages as retained under $sparkplug/certificates/, and can replace the NDEATH timestamp. Its README does not mention STATE. It appears on the Eclipse Sparkplug compatible-software list.

Where HiveMQ and Ignition fail

  • Store and forward lives in the wrong place. It is in Transmission’s history store, not the broker. The guide uses the in-memory store and does not say whether buffered data survives a gateway restart. Find out before an outage does.
  • A mismatched Primary Host ID. In our reading, a node that never sees STATE for its configured host will hold its data.
  • STATE permissions. The January 2025 HiveMQ thread above: publishing worked only as admin until both STATE/# and spBv1.0/STATE/# were permitted. A host on the legacy STATE form is a Sparkplug 3.0 reality, so allow both on purpose.
  • Wildcard rights. The host needs to subscribe to spBv1.0/#. Granting that to every client defeats the topic-level controls described in OT network security. Give only the host the wildcard and each node only its own group.
  • Rebirth storms. When a host asks nodes to republish their births after a reconnect, a large estate sends every tag at once. Stagger large estates and watch broker and WAN load after an outage. This is our reading of how Sparkplug behaves, not a HiveMQ statement.
  • Cluster loss. A broker cluster with a replica count below 2 can lose retained and queued data on a restart, according to HiveMQ’s cluster documentation. See clustering and high availability.

Specifics worth checking

ItemWhere to lookOur caution
Module versionsCirrus Link and Inductive Automation release notesHiveMQ’s guide states only Ignition 8.1.43 in a download link
TLSYour broker listener and the server setTest the guide’s TLS options against your real listener
Store typeTransmission settingsIn-memory by default in the guide
Primary Host IDServer set on every gatewayMust match across Transmitters and Engine
RolesHiveMQ security configurationBoth STATE forms, spBv1.0/# for the host only

A bench test before you trust it

Run these on a bench with the exact module and broker versions you plan to deploy, and write the results down.

  1. Connect with TLS and a client certificate, so the real listener is tested.
  2. Stop the broker for ten minutes with data flowing and confirm the history store fills and then drains in order.
  3. Restart the Transmission gateway during the outage and see whether buffered data survives.
  4. Stop Engine and confirm STATE flips to offline and the Transmitters change behaviour.
  5. Remove the legacy STATE/# permission and confirm what fails, so the failure is known before an audit finds it.

Where it gets hard

The hard part is rarely the first connection. It is the second site, the failed WAN and the audit that asks who can read every tag. For the Sparkplug side in depth see HiveMQ and Sparkplug, and for another broker’s view Solace and Sparkplug B. Back to HiveMQ manufacturing and the HiveMQ hub. Sibling: multi-site architecture.

Frequently asked questions

Does HiveMQ work with Ignition?

Yes. HiveMQ joined the Ignition Technology Ecosystem Program on 23 July 2025, and a HiveMQ guide of 20 June 2025 connects Cirrus Link MQTT Transmission and MQTT Engine to a HiveMQ broker. The release makes no Sparkplug or module claim, so the guide is the technical evidence.

Do I need the MQTT Distributor module with HiveMQ?

Not as a rule. Inductive Automation's documentation says MQTT Engine's broker can be a gateway with the MQTT Distributor module or a third-party server, and HiveMQ's guide uses a HiveMQ broker. Forum threads disagree on third-party brokers, so test your exact versions.

Where does Ignition buffer data when the HiveMQ broker is unreachable?

In the MQTT Transmission history store on the Ignition gateway, not in the broker. HiveMQ's guide says Transmitters with the same Primary Host ID switch to store and forward when Engine goes offline and a history store is enabled. The guide uses the in-memory store and does not say whether data survives a restart.

Why can't my Sparkplug client publish through HiveMQ?

A frequent cause is permissions. In a January 2025 HiveMQ community thread, a user could publish only with the admin role until the legacy topic STATE/# was permitted as well as spBv1.0/STATE/#. Check the roles for both STATE forms, the host's subscription to spBv1.0/# and each node's publish rights.

Should the broker run at the plant or in a central data centre?

In our view, at the plant when the plant must keep logging through a WAN outage, with a bridge to a central cluster for enterprise consumers. A central broker is simpler but makes the WAN a production dependency. HiveMQ's documented multi-site tool is bridging.

References and further reading

  1. A Step-by-Step Guide to Connecting Ignition and MQTT with HiveMQ · HiveMQDated 20 June 2025: modules, server sets, Primary Host ID and store and forward.
  2. HiveMQ joins the Inductive Automation ecosystem · HiveMQDated 23 July 2025.
  3. IIoT architecture (Ignition 8.3 documentation) · Inductive AutomationSays the broker can be a third-party server.
  4. Decoding operational behavior of MQTT Sparkplug · HiveMQDated 12 April 2024: STATE, birth and death sequence.
  5. Sparkplug RBAC issues · HiveMQ CommunityJanuary 2025 thread on STATE topic permissions.
  6. HiveMQ Sparkplug Aware extension · HiveMQ on GitHubWhat the open-source extension does and does not cover.

← Manufacturing

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.