🥮 mooncakes.io
    Sign in

    moon-mqtt-client

    Native asynchronous MQTT 3.1.1 client with QoS 0/1 and clean-session reconnect

    Download zip
    Version
    0.2.0
    License
    Apache-2.0
    Repository
    Last updated
    last month
    Downloads
    37

    #moon-mqtt-client

    native client

    A native asynchronous MQTT 3.1.1 client for MoonBit. Connect to an existing MQTT broker, subscribe to device or application events, and publish commands and state without writing a socket loop for each application.

    The packet codec is provided by zbhzs1/moonbit-mqtt; this package supplies the connection lifecycle, TLS transport, request tracking, heartbeat and clean-session reconnection on top of moonbitlang/async. It is an early implementation, not a certified MQTT conformance implementation.

    #Start with the state-sync demo

    The primary demonstration is a controller process and an independent simulated device process (no hardware; the device is a simulation and is labelled as one) talking to a real broker:

    ./scripts/moon.sh build --target native
.venv/bin/python examples/mqtt_demo/demo.py

    Four reproducible scenarios run in sequence: normal ON/OFF, a lost PUBACK whose command the device still executes, a broker restart, and a controller restart while the device stays ON. Every line of output keeps three things apart: desired (what hysteresis wants), result (sent/not_sent/unknown), and reported (what the device actually says, correlated by id). See docs/SCENARIOS.md for the topics and rules.

    A thin publish/subscribe CLI reuses the public API:

    ./scripts/moon.sh run examples/mqtt_demo/cli --target native -- \
  publish --host 127.0.0.1 -t lab/state -m ON --qos 1 --retain --stats
./scripts/moon.sh run examples/mqtt_demo/cli --target native -- \
  subscribe --host 127.0.0.1 -t 'lab/#' --count 1 --timeout-ms 3000

    #Scope

    • Native TCP and server-authenticated TLS (system roots or a custom PEM CA).
    • MQTT 3.1.1, QoS 0 and 1, retained messages, Last Will, username/password.
    • Subscribe and unsubscribe acknowledgements, including per-topic rejection.
    • Bounded send queue, event queue, packet size and concurrent requests.
    • CleanSession=true only: reconnect creates a new session and restores confirmed subscriptions. A Connected(generation) event follows subscription restoration.
    • Callback-scoped tasks and sockets. Normal callback return or disconnect() sends DISCONNECT; callback failure or cancellation closes the transport abruptly.

    No QoS 2, MQTT 5, persistent sessions, offline queue, cross-connection retransmit, client-certificate authentication, WebSocket, browser or microcontroller target.

    #Build from source

    Use the native MoonBit toolchain; this checkout was developed with moon 0.1.20260904 and moonc v0.10.12+1634b282e (2026-09-07).

    moon update
moon check --target native
moon test --target native
moon build --target native

    scripts/moon.sh uses a local .tools/moon installation or MOON_HOME when provided. Toolchains and build outputs are not part of the source distribution. For registry releases, install the package with:

    moon add Strangelight-Merser/moon-mqtt-client

    See GitHub Releases for available versions and their verification results.

    #API

    The following is the callback shape used by the runnable examples. Import this package as @mqtt and moonbitlang/core/encoding/utf8 as @utf8.

    async fn main {
      let config = @mqtt.Config::new("127.0.0.1", "my-moon-client")
      @mqtt.with_client(config, async fn(client) {
        let results = client.subscribe([
          { topic: "lab/temperature", qos: @mqtt.AtLeastOnce },
        ])
        if results == [@mqtt.Granted(@mqtt.AtLeastOnce)] {
          client.publish("lab/status", @utf8.encode("ready"),
            qos=@mqtt.AtLeastOnce, retain=true)
        }
        // Consume Connected, Disconnected and MessageReceived events here.
        // A long-running subscriber must continually drain next_event().
      })
    }

    with_client waits for the first successful connection before invoking the callback. An initial connection/CONNACK/TLS failure is returned to the caller. After a connection has been established, failures trigger bounded retries. wait_connected() can wait through a reconnect; publish/subscribe/unsubscribe while disconnected fail with NotConnected, rather than entering an offline queue.

    Use TlsMode::SystemRoots with port 8883 for public trust or TlsMode::CustomCA("path/to/ca.pem") for a private CA. The configured host is also the verified TLS hostname. There is no option to disable verification.

    #Delivery and failure semantics

    The normative description of timeouts, queueing, cancellation and failure classification is docs/API-CONTRACT.md. In short:

    ResultWhat it establishes
    QoS 0 publish returnsThe transport write completed; no broker acknowledgement exists.
    QoS 1 publish returnsA matching PUBACK arrived on that connection. It does not establish downstream processing or a physical action.
    NotSentA queued request failed before its write started.
    OutcomeUnknownA write began but the operation was not confirmed. Partial writes and lost ACKs are included.
    Backpressure from a requestThe send or inflight limit prevented accepting that request.
    Event/control queue overflowThe client terminates with an error instead of silently dropping messages.

    The whole operation is budgeted by operation_timeout_ms (default 5000 ms), covering queue time, the write and the wait for the acknowledgement. A single socket write is separately bounded by write_timeout_ms, and PINGRESP is timed from the completed PINGREQ write. An expired operation budget closes the connection and ends all pending requests; old identifiers never cross into the new connection. The application chooses whether to retry an uncertain operation; use idempotent state-setting commands or application command IDs where appropriate.

    Protocol control packets (PUBACK, PINGREQ, DISCONNECT) use reserved bounded slots and are never blocked behind a full business queue. If the control slots themselves run out, the client terminates loudly instead of dropping one.

    Incoming QoS 1 is acknowledged after acceptance into the bounded event queue, not after application processing. The queue is volatile. QoS 1 duplicates are possible. Reconnection with a clean session can lose messages during the gap; a broker may replay retained state on resubscription. The library does not promise exactly-once processing, durable delivery or uninterrupted subscriptions.

    Default limits: 64 queued business sends, 16 reserved control slots, 128 queued events, 32 pending operations, 65,536 bytes per packet, 5-second connect, operation, write and PINGRESP timeouts, 30-second keepalive, 10 consecutive reconnect attempts with a delay growing from 250 ms to 5 seconds plus per-client jitter. Client::set_reconnect_seed pins that jitter for tests. Confirmed subscription filters remain in memory until unsubscribed; keep their set bounded in the application.

    Client::stats() returns a read-only snapshot: generation, connection state, business/control/event queue occupancy, pending requests, reconnects, disconnects, unknown outcomes and the most recent disconnect reason. It contains no credentials and no message bodies and needs no monitoring service.

    #Runnable scenarios

    中文入口：从这里开始。

    See docs/SCENARIOS.md for complete inputs, rules, outputs and failure boundaries for three intended uses:

    1. Temperature control with hysteresis, retained state and availability.
    2. A bounded Frigate event deduplicator which alerts on completed person events.
    3. A ROS bridge JSON/primitive contract with command validation and receipts.

    The examples use fixtures. They do not claim a deployed camera, robot or Zigbee integration. The Frigate and ROS contracts remain supplementary adapter contracts; their tests reject non-finite speeds, empty command ids and quote/backslash/control-character payloads, and every emitted payload is JSON.

    #Verification

    python3 -m venv .venv
.venv/bin/pip install -r tests/integration/requirements.txt
# Install Mosquitto and OpenSSL using your OS package manager, then:
PYTHON=.venv/bin/python ./tests/integration/run.sh

    Run every local check with ./scripts/check.sh. Two heavier harnesses are run separately and are recorded in docs/VALIDATION.md:

    # Pinned official EMQX image: send/receive, subscription denial, restart recovery
./tests/emqx.sh

# 30-minute restart soak: 1 KiB QoS 1, concurrency 16, 100 disconnect cycles.
# The broker log is quiet by default and capped, so the run leaves bounded evidence.
MOONBIT_ASYNC_CHECK_FD_LEAK=1 .venv/bin/python tests/soak.py \
  --duration 1800 --cycles 100 --artifacts tests/integration/artifacts/soak

    The integration suite uses independent Mosquitto and Eclipse Paho processes, loopback-only listeners, temporary certificates and packet fault injection. See docs/VALIDATION.md for the actual executed results and remaining gaps. A workflow definition is not evidence of a completed hosted CI run.

    #License and upstream work

    Apache-2.0. See NOTICE. The dependencies remain separate packages with their own attribution and licenses; this project does not claim authorship of MQTT wire encoding, the async runtime or TLS implementation.

    ClientError

    source
    pub(all) suberror ClientError {
      InvalidConfig(String)
      ProtocolError(String)
      NotConnected
      Closed
      Backpressure(String)
      NotSent(String)
      OutcomeUnknown(String)
      ConnectionRefused(String)
      ReconnectExhausted(String)
      OperationTimeout(String)
      WriteTimeout(String)
      PingResponseTimeout(String)
    } derive(
    Debug    )

    Client

    source
    type Client

    Client::disconnect

    source
    async fn Client::disconnect(self : Client) -> Unit

    Stop reconnect, send DISCONNECT when possible, and await worker cleanup.

    Client::next_event

    source
    async fn Client::next_event(self : Client) -> Event

    Client::publish

    source
    async fn Client::publish(self : Client, topic : String, payload : Bytes, qos? : QoS, retain? : Bool) -> Unit

    Client::set_reconnect_seed

    source
    fn Client::set_reconnect_seed(self : Client, seed : Int) -> Unit

    Pin the per-client reconnect jitter stream. Intended for tests that need a reproducible backoff sequence; the value never reaches the wire. Setting it after a reconnect has started only affects subsequent attempts.

    Client::stats

    source
    fn Client::stats(self : Client) -> ClientStats

    Read-only runtime snapshot. Safe to call from any task; it never blocks.

    Client::subscribe

    source
    async fn Client::subscribe(self : Client, topics : Array[Subscription]) -> Array[SubscriptionResult]

    Client::unsubscribe

    source
    async fn Client::unsubscribe(self : Client, topics : Array[String]) -> Unit

    Client::wait_connected

    source
    async fn Client::wait_connected(self : Client) -> Unit

    ClientStats

    source
    pub(all) struct ClientStats {
      generation : Int
      connected : Bool
      business_queue_depth : Int
      control_queue_depth : Int
      event_queue_depth : Int
      pending_requests : Int
      reconnects : Int
      disconnects : Int
      unknown_outcomes : Int
      last_disconnect : String?
    }

    Read-only runtime snapshot returned by Client::stats. It contains counters and queue occupancy only: no credentials, no message bodies, and no peer address beyond the reason text the client already reports through Disconnected.

    Config

    source
    pub(all) struct Config {
      host : String
      port : Int
      client_id : String
      tls : TlsMode
      username : String?
      password : Bytes?
      will : Will?
      keep_alive_secs : Int
      connect_timeout_ms : Int
      operation_timeout_ms : Int
      write_timeout_ms : Int
      ping_response_timeout_ms : Int
      send_capacity : Int
      receive_capacity : Int
      max_inflight : Int
      max_packet_size : Int
      reconnect_delay_ms : Int
      max_reconnect_delay_ms : Int
      reconnect_attempts : Int
      control_capacity : Int
    }

    Config::new

    source
    fn Config::new(host : String, client_id : String, port? : Int, tls? : TlsMode, username? : String?, password? : Bytes?, will? : Will?, keep_alive_secs? : Int, connect_timeout_ms? : Int, operation_timeout_ms? : Int, write_timeout_ms? : Int, ping_response_timeout_ms? : Int, send_capacity? : Int, receive_capacity? : Int, max_inflight? : Int, max_packet_size? : Int, reconnect_delay_ms? : Int, max_reconnect_delay_ms? : Int, reconnect_attempts? : Int, control_capacity? : Int) -> Config

    Event

    source
    pub(all) enum Event {
      Connected(Int)
      Disconnected(Int, String)
      MessageReceived(Message)
    } derive(
    Debug    )

    Message

    source
    pub(all) struct Message {
      topic : String
      payload : Bytes
      qos : QoS
      retain : Bool
      dup : Bool
      generation : Int
    } derive(
    Debug    )

    QoS

    source
    pub(all) enum QoS {
      AtMostOnce
      AtLeastOnce
    } derive(Eq, 
    Debug    )

    Subscription

    source
    pub(all) struct Subscription {
      topic : String
      qos : QoS
    } derive(Eq, 
    Debug    )

    SubscriptionResult

    source
    pub(all) enum SubscriptionResult {
      Granted(QoS)
      Rejected
    } derive(Eq, 
    Debug    )

    TlsMode

    source
    pub(all) enum TlsMode {
      Plain
      SystemRoots
      CustomCA(String)
    } derive(
    Debug    )

    Will

    source
    pub(all) struct Will {
      topic : String
      payload : Bytes
      qos : QoS
      retain : Bool
    } derive(
    Debug    )

    with_client

    source
    async fn[T] with_client(config : Config, action : async (Client) -> T) -> T

    Own all client tasks within the callback's lifetime.

    Source Files

    Powered by MoonBit

    Site sourceReport issuePackagesBuild queueSkillsStatistics

    © 2026 mooncakes.io