MQTT is a light weight, client to server, publish / subscribe messaging protocol. MQTT has been specifically designed to reduce transport overhead (and thus network traffic) and code footprint on client devices. For this reason MQTT is ideally suited to constrained devices such as sensors and actuators and is quickly becoming the defacto standard communication protocol for IoT.

Apache ActiveMQ Artemis supports the following MQTT versions (with links to their respective specifications):

By default there are acceptor elements configured to accept MQTT connections on ports 61616 and 1883.

See the general Protocols and Interoperability chapter for details on configuring an acceptor for MQTT.

Refer to the MQTT examples for a look at some of this functionality in action.

1. MQTT Quality of Service

MQTT offers 3 quality of service levels.

Each message (or topic subscription) can define a quality of service that is associated with it. The quality of service level defined on a topic is the maximum level a client is willing to accept. The quality of service level on a message is the desired quality of service level for this message. The broker will attempt to deliver messages to subscribers at the highest quality of service level based on what is defined on the message and topic subscription.

Each quality of service level offers a level of guarantee by which a message is sent or received:

  • QoS 0: AT MOST ONCE

    Guarantees that a particular message is only ever received by the subscriber a maximum of one time. This does mean that the message may never arrive. The sender and the receiver will attempt to deliver the message, but if something fails and the message does not reach its destination (say due to a network connection) the message may be lost. This QoS has the least network traffic overhead and the least burden on the client and the broker and is often useful for telemetry data where it doesn’t matter if some of the data is lost.

  • QoS 1: AT LEAST ONCE

    Guarantees that a message will reach its intended recipient one or more times. The sender will continue to send the message until it receives an acknowledgment from the recipient, confirming it has received the message. The result of this QoS is that the recipient may receive the message multiple times, and also increases the network overhead than QoS 0, (due to acks). In addition more burden is placed on the sender as it needs to store the message and retry should it fail to receive an ack in a reasonable time.

  • QoS 2: EXACTLY ONCE

    The most costly of the QoS (in terms of network traffic and burden on sender and receiver) this QoS will ensure that the message is received by a recipient exactly one time. This ensures that the receiver never gets any duplicate copies of the message and will eventually get it, but at the extra cost of network overhead and complexity required on the sender and receiver.

2. MQTT Retain Messages

MQTT has an interesting feature in which messages can be "retained" for a particular address. This means that once a retain message has been sent to an address, any new subscribers to that address will receive the last sent retained message before any others messages. This happens even if the retained message was sent before a client has connected or subscribed. An example of where this feature might be useful is in environments such as IoT where devices need to quickly get the current state of a system when they are on boarded into a system.

Retained messages are stored in a queue named with a special prefix according to the name of the topic where they were originally sent. For example, a retained message sent to the topic /abc/123 will be stored in a multicast queue named $sys.mqtt.retain.abc.123 with an address of the same name. The MQTT specification doesn’t define how long retained messages should be stored so the broker will hold on to this data until a client explicitly deletes the retained message or it potentially expires. However, even at that point the queue and address for the retained message will remain. These resources can be automatically deleted via the following address-setting:

<address-setting match="$sys.mqtt.retain.#">
   <auto-delete-queues>true</auto-delete-queues>
   <auto-delete-addresses>true</auto-delete-addresses>
</address-setting>

Keep in mind that it’s also possible to automatically apply an expiry-delay to retained messages as well.

3. Will Messages

A will message can be sent when a client initially connects to a broker. Clients are able to set a "will message" as part of the connect packet. If the client abnormally disconnects, say due to a device or network failure the broker will proceed to publish the will message to the specified address (as defined also in the connect packet). Other subscribers to the will topic will receive the will message and can react accordingly. This feature can be useful in an IoT style scenario to detect errors across a potentially large scale deployment of devices.

4. Debug Logging

Detailed protocol logging (e.g. packets in/out) can be activated by turning on TRACE logging for org.apache.activemq.artemis.core.protocol.mqtt. Follow these steps to configure logging appropriately.

The MQTT specification doesn’t dictate the format of the payloads which clients publish. As far as the broker is concerned a payload is just an array of bytes. However, to facilitate logging the broker will encode the payloads as UTF-8 strings and print them up to 256 characters. Payload logging is limited to avoid filling the logs with potentially hundreds of megabytes of unhelpful information.

5. Persistent Subscriptions

The subscription information for MQTT sessions is stored in an internal queue named $sys.mqtt.sessions and persisted to storage (assuming persistence is enabled). The information is durable so that MQTT subscribers can reconnect and resume their subscriptions seamlessly after a broker restart, failure, etc. When brokers are configured for high availability this information will be available on the backup so even in the case of a broker fail-over subscribers will be able to resume their subscriptions.

While persistent subscriptions can be convenient they impose a performance penalty since data must be written to storage. If you don’t need the convenience (e.g. you always use clean sessions) and you don’t want the performance penalty then you can disable it by disabling durability for the $sys.mqtt.sessions queue in broker.xml, e.g.:

<addresses>
   ...
   <address name="$sys.mqtt.sessions">
      <anycast>
         <queue name="$sys.mqtt.sessions">
            <durable>false</durable>
         </queue>
      </anycast>
   </address>
   ...
</addresses>

The setting mqtt-session-state-persistence-timeout controls how long the broker will wait for the data to be written to storage before throwing an error. It is measured in milliseconds. The default is 5000.

6. Custom Client ID Handling

The client ID used by an MQTT application is very important as it uniquely identifies the application. In some situations broker administrators may want to perform extra validation or even modify incoming client IDs to support specific use-cases. This is possible by implementing a custom security manager as demonstrated in the security-manager example.

The simplest implementation is a "wrapper" just like the security-manager example uses. In the authenticate method you can modify the client ID using setClientId() on the org.apache.activemq.artemis.spi.core.protocol.RemotingConnection that is passed in. If you perform some custom validation of the client ID you can reject the client ID by throwing a org.apache.activemq.artemis.core.protocol.mqtt.exceptions.InvalidClientIdException.

7. Wildcard subscriptions

MQTT defines a special wildcard syntax for topic filters. This definition is found in section 4.7.1 of both the 3.1.1 and 5 specs. MQTT topics are hierarchical much like a file system, and they use a special character (i.e. / by default) to separate hierarchical levels. Subscribers are able to subscribe to specific topics or to whole branches of a hierarchy.

To subscribe to branches of an address hierarchy a subscriber can use wild cards. There are 2 types of wildcards in MQTT:

  • Multi level (#)

    Adding this wildcard to an address would match all branches of the address hierarchy under a specified node. For example: /uk/ Would match /uk/cities, /uk/cities/newcastle and also /uk/rivers/tyne. Subscribing to an address would result in subscribing to all topics in the broker. This can be useful, but should be done so with care since it has significant performance implications.

  • Single level (+)

    Matches a single level in the address hierarchy. For example /uk/+/stores would match /uk/newcastle/stores but not /uk/cities/newcastle/stores.

This is close to the default wildcard syntax, but not exactly the same. Therefore, some conversion is necessary. This conversion isn’t free so if you want the best MQTT performance use broker.xml to configure the wildcard syntax to match MQTT’s, e.g.:

<core>
   ...
   <wildcard-addresses>
      <delimiter>/</delimiter>
      <any-words>#</any-words>
      <single-word>+</single-word>
   </wildcard-addresses>
   ...
</core>

Of course, changing the default syntax also means other clients on other protocols will need to follow this same syntax as well as the match values of your address-setting configuration elements.

8. Web Sockets

Apache ActiveMQ Artemis also supports MQTT over Web Sockets. Modern web browsers which support Web Sockets can send and receive MQTT messages.

MQTT over Web Sockets is supported via a normal MQTT acceptor:

<acceptor name="mqtt-ws-acceptor">tcp://host:1883?protocols=MQTT</acceptor>

With this configuration, Apache ActiveMQ Artemis will accept MQTT connections over Web Sockets on the port 1883. Web browsers can then connect to ws://<server>:1883 using a Web Socket to send and receive MQTT messages.

SSL/TLS is also available, e.g.:

<acceptor name="mqtt-wss-acceptor">tcp://host:8883?protocols=MQTT;sslEnabled=true;keyStorePath=/path/to/keystore;keyStorePassword=myPass</acceptor>

Web browsers can then connect to wss://<server>:8883 using a Web Socket to send and receive MQTT messages.

The MQTT specifications define a behavior often referred to as "link stealing." This means that whenever a new client connects with the same client ID as another existing client then the existing client’s session will be closed and its network connection will be terminated.

In certain use-cases this behavior is not desired so it is configurable. The URL parameter allowLinkStealing can be configured on the MQTT acceptor to modify this behavior. By default allowLinkStealing is true. If it is set to false then whenever a new client connects with the same client ID as another existing client then the new client’s session will be closed and its network connection will be terminated. In the case of MQTT 5 clients they will receive a disconnect reason code of 0x80 (i.e. "Unspecified error").

10. Automatic Subscription Clean-up

Sometimes MQTT 3.x clients using CleanSession=false don’t properly unsubscribe. The URL parameter defaultMqttSessionExpiryInterval can be configured on the MQTT acceptor so that abandoned sessions and subscription queues will be cleaned up automatically after the expiry interval elapses.

MQTT 5 has the same basic semantics with slightly different configuration. The CleanSession flag was replaced with CleanStart and a session expiry interval property. The broker will use the client’s session expiry interval if it is set. If it is not set then the broker will apply the defaultMqttSessionExpiryInterval.

The default defaultMqttSessionExpiryInterval is -1 which means no clean up will happen for MQTT 3.x clients or for MQTT 5 clients which do not pass their own session expiry interval. Otherwise it represents the number of seconds which must elapse after the client has disconnected before the broker will remove the session state and subscription queues.

MQTT session state is scanned every 5,000 milliseconds by default. This can be changed using the mqtt-session-scan-interval element set in the core section of broker.xml.

11. Flow Control

MQTT 5 introduced a simple form of flow control. In short, a broker can tell a client how many QoS 1 & 2 messages it can receive before being acknowledged and vice versa.

This is controlled on the broker by setting the receiveMaximum URL parameter on the MQTT acceptor in broker.xml.

The default value is 65535 (the maximum value of the 2-byte integer used by MQTT).

A value of 0 is prohibited by the MQTT 5 specification.

A value of -1 will prevent the broker from informing the client of any receive maximum which means flow-control will be disabled from clients to the broker. This is effectively the same as setting the value to 65535, but reduces the size of the CONNACK packet by a few bytes.

12. Topic Alias Maximum

MQTT 5 introduced topic aliasing. This is an optimization for the size of PUBLISH control packets as a 2-byte integer value can now be substituted for the _name of the topic which can potentially be quite long.

Both the client and the broker can inform each other about the maximum alias value they support (i.e. how many different aliases they support). This is controlled on the broker using the topicAliasMaximum URL parameter on the acceptor used by the MQTT client.

The default value is 65535 (the maximum value of the 2-byte integer used by MQTT).

A value of 0 will disable topic aliasing from clients to the broker.

A value of -1 will prevent the broker from informing the client of any topic alias maximum which means aliasing will be disabled from clients to the broker. This is effectively the same as setting the value to 0, but reduces the size of the CONNACK packet by a few bytes.

13. Maximum Packet Size

MQTT 5 introduced the maximum packet size. This is the maximum packet size the server or client is willing to accept.

This is controlled on the broker by setting the maximumPacketSize URL parameter on the MQTT acceptor in broker.xml.

The default value is 268435455 (i.e. 256MB - the maximum value of the variable byte integer used by MQTT).

A value of 0 is prohibited by the MQTT 5 specification.

A value of -1 will prevent the broker from informing the client of any maximum packet size which means no limit will be enforced on the size of incoming packets. This also reduces the size of the CONNACK packet by a few bytes.

14. Server Keep Alive

All MQTT versions support a connection keep alive value defined by the client. MQTT 5 introduced a server keep alive value so that a broker can define the value that the client should use. The primary use of the server keep alive is for the server to inform the client that it will disconnect the client for inactivity sooner than the keep alive specified by the client.

This is controlled on the broker by setting the serverKeepAlive URL parameter on the MQTT acceptor in broker.xml.

The default value is 60 and is measured in seconds.

A value of 0 completely disables keep alives no matter the client’s keep alive value. This is not recommended because disabling keep alives is generally considered dangerous since it could lead to resource exhaustion.

A value of -1 means the broker will always accept the client’s keep alive value (even if that value is 0).

Any other value means the serverKeepAlive will be applied if it is less than the client’s keep alive value unless the client’s keep alive value is 0 in which case the serverKeepAlive is applied. This is because a value of 0 would disable keep alives and disabling keep alives is generally considered dangerous since it could lead to resource exhaustion.

15. Enhanced Authentication

MQTT 5 introduced enhanced authentication which extends the existing name & password authentication to include challenge / response style authentication.

However, there are currently no challenge / response mechanisms implemented so if a client passes the "Authentication Method" property in its CONNECT packet it will receive a CONNACK with a reason code of 0x8C (i.e. bad authentication method) and the network connection will be closed.

16. Publish Authorization Failures

The MQTT 3.1.1 specification is ambiguous regarding the broker’s behavior when a PUBLISH packet fails due to a lack of authorization. In section 3.3.5 it says:

If a Server implementation does not authorize a PUBLISH to be performed by a Client; it has no way of informing that Client. It MUST either make a positive acknowledgement, according to the normal QoS rules, or close the Network Connection

By default the broker will close the network connection. However if you’d rather have the broker make a positive acknowledgement then set the URL parameter closeMqttConnectionOnPublishAuthorizationFailure to false on the relevant MQTT acceptor in broker.xml, e.g.:

<acceptor name="mqtt">tcp://0.0.0:1883?protocols=MQTT;closeMqttConnectionOnPublishAuthorizationFailure=false</acceptor>