# `MqttX.Server.SessionBackend.Replicated`
[🔗](https://github.com/cignosystems/mqttx/blob/v0.13.0/lib/mqttx/server/session_backend/replicated.ex#L1)

A `MqttX.Server.SessionBackend` that keeps every saved session on every
node running the listener, in memory: when a node goes down, a client of
its resumes its session on any other node.

    MqttX.Server.start_link(MyBroker, [],
      name: MyApp.Broker,
      port: 1883,
      session_backend: MqttX.Server.SessionBackend.Replicated
    )

Each save goes to every other node of the listener's group (the listeners
with the same `:name` on connected nodes, see `MqttX.Server.Cluster`), and
a node that joins is sent what the others hold. A record carries the node
that saved it and a version; a node keeps the newest version of each
session, so a save that arrives late cannot replace a newer one. Versions
combine the wall clock with the version they replace, which orders the
saves of a session moving between nodes as long as the nodes' clocks
agree to within the time the move takes.

Costs: every node holds every persistent session, and each change to one
is sent to every node (once per `:session_flush_interval`, like any
backend). A node that dies loses at most that much of its sessions'
changes; a session saved while connected starts its expiry clock when its
node leaves.

The records live as long as one node runs the listener: a restart of the
whole cluster loses them. For that, implement the behaviour over a
database.

# `child_spec`

Returns a specification to start this module under a supervisor.

See `Supervisor`.

---

*Consult [api-reference.md](api-reference.md) for complete listing*
