A cluster is several broker hosts acting as one system. Cluster Viewer draws that system from the broker host you are already connected to, so you can see who leads, which links are up, which node owns each exclusive piece of LoT, and how partitions are placed — without opening a shell on every host.
Cluster Viewer carries a Beta badge on its window. The dock icon is labeled Cluster. It appears only while the connected broker host is publishing cluster status.
Like a seating chart that updates itself. You see who is leading, who is sitting in, and who is holding each shared task — from the one host you opened.
Confirm the cluster is up — See which members are alive and who the leader is
Check a link — See whether data-plane links between hosts are connected
Find who runs an exclusive piece of LoT — Timers, ON START actions, exclusive connectors, and scheduled connector events run on one node
See partition placement — See which host owns each partition and where the copies sit
Clustering is an Enterprise license feature. If the license does not grant it, the broker runs standalone and this icon stays off the dock. See High-availability clustering and Licensing.Open Cluster on the dock. The window is read-only. It reads the cluster status the connected broker host publishes. It does not elect a leader, move ownership, or change cluster settings.If clustering is off, the icon is not on the dock. Opening the window another way shows that the connected broker is not publishing cluster status.
A strip across the top summarizes the cluster as this broker host sees it:
Item
What it tells you
Cluster
The cluster identity
Leader
The host that currently leads
Members
How many members are alive out of the member count
Term / Commit
How far leader election and the committed log have progressed
LoT
Whether nodes are on the same committed LoT version, or the versions are mixed
Data plane
How many links between hosts are up
Listeners
Whether any server-connector listener on a member has failed
The strip also says the view is Live (or how old the status is), which host you are seeing it from, and that it is read-only.The canvas draws each member and the links between them. Select a node to open it on the Node tab. Select a link to see each direction. The leader is marked on its node. A host the status names but does not describe yet is drawn as unidentified.
The picture is the view from the broker host you are connected to. It is not a second login to every member.
The selected host: whether it is alive, its role, and its endpoint when the cluster published one.For the host you are connected to, the tab also shows the term and commit index, and whether that host holds the LoT lease.LoT shows the version this host has applied against the version the cluster has committed, whether those match, and any apply errors. Connector listeners lists the server connectors that host reported. Outgoing data-plane links lists the links that start on this host.
Node tab: the selected host, its role, LoT version, connector listeners, and outgoing data-plane links
Replication on the host you are connected to: connected peers, forward backlog, dropped forwards, quorum failures, and how long the last quorum took.Data-plane links lists every reported link, with its state (connected, connecting, disconnected, or unknown). A stale report is marked stale.
Links tab: replication on the connected host and the state of every data-plane link
Which host runs each piece of LoT that must run on exactly one node:
Kind
What it is
Actions (timers, ON START)
Scheduled actions, and actions that run once when the cluster session starts
Exclusive connectors
Connectors that hold a device, a serial port, or a server port
Scheduled connector events
A connector’s timed events
Owner is the host that runs it. Next is the host that takes it if the owner is lost. Epoch goes up each time the entity moves. Unowned means no live host can run it, so it runs nowhere until one can. An exclusive connector is defined on every node, but only its owner opens the connection. The others hold it ready.Select a node on the canvas first if you want the list limited to entities that node owns.
Ownership tab: which host owns each exclusive action, connector, and scheduled connector event
The partition map: how many partitions there are, the replication factor, how many are under-replicated, and which host owns each one.Under-replicated partitions still work. Their replica set is smaller than the factor because there are not enough live hosts to place every copy — for example while a host is down. A healthy cluster shows none.Pick a host to highlight the partitions it owns, the ones where it only keeps a copy, and the ones it does not hold.
Partitions tab: the partition map, replication factor, and owner of each partition
The MQTT clients connected to each node right now, with a total across the cluster. A client is connected to one node. Behind a load balancer, the node can change each time it reconnects.Under Show the clients of, pick All nodes or one node. Filter by client ID, user, or address, and sort by Last active, Newest connection, Most queued, or Client id. The connections the nodes open to each other are hidden unless you turn on Show the cluster’s own connections.Each card shows the client ID, user, address, MQTT version, messages in and out, and when it was last active. A queued badge counts messages waiting for that client; a number that keeps growing means the client isn’t keeping up. Expand a card for when it connected, its last packet, bytes in and out, and how long the broker keeps its session after it disconnects.Lists refresh every 7 seconds. Other nodes’ clients are relayed by those nodes, so they can be a few seconds older. An older broker reports only the clients of the node you’re connected to.
Clients tab on a three-node cluster: nine clients across acme-mqtt-1, acme-mqtt-2, and acme-mqtt-3, sorted by last active
An older broker host may publish only part of this view. The window then names what is missing and still shows the status it did receive.
