Contribution Submission: MQTT v5 Input and Output components

[yvhaa2]

Background

We are evaluating Redpanda Connect for use in our data pipelines, but this use case requires MQTT v5 support, which currently does not exist for Redpanda Connect. Currently only MQTT v3.1.1 is supported via the mqtt input and output using the paho.mqtt.golang library.

Prior Work

There were two prior attempts to contribute MQTT v5 support to Redpanda Connect:

• PR Add module for MQTTv5 #2197: Was closed due to withdrawal of the contribution on change from Benthos to Redpanda Connect.
• PR MQTT v5 support #3731: Was closed due to unaccepted CLA.

Kafka Connect Equivalent

There is no Kafka Connect equivalent since the Kafka Connect MQTT source connector is also based on an MQTT v3.1.1 paho library.

Proposal

We would like to contribute a new mqtt5 input and output component. Per §3.1.1 of the CONTRIBUTING.md, code written by non-Redpanda engineers needs to be reviewed and scoped by Redpanda. We are opening this issue to ask for comments on two points:

1. Is an MQTT v5 input and output component still desired/wanted by Redpanda?
2. Is the following scope properly sized for a MQTT v5 input and output component?

Intended Scope

• A new, separate mqtt5 input and output component.
• Written in Go and based on paho.golang. This library is written by the same authors as the mqtt.paho.golang library that is already in use with the mqtt connector.
• Code structure will mirror the existing mqtt connector where it makes sense.
• All configuration fields shared with the existing mqtt connector are carried over unchanged
• New mqtt5 input config options:

Field	Notes
session_expiry_interval	How long the broker retains session state after disconnect.
receive_maximum	Maximum in-flight QoS 1/2 messages the broker may send to this client simultaneously.
maximum_packet_size	Maximum packet size this client will accept from the broker.
no_local	Subscription option: do not receive messages published by this client.
retain_as_published	Subscription option: preserve the retain flag from the original publisher.
retain_handling	Subscription option: controls whether retained messages are sent on subscribe.
topic_alias_maximum	Maximum number of topic aliases this client will accept from the broker. 0 disables topic aliases.
subscription_identifiers	Enables the use of subscription identifiers by the client.
will_*	Full v5 will message config.

• Metadata set by the mqtt5 input:

Metadata key	Source
mqtt5_user_properties_<key>	User Properties from the PUBLISH packet. If a key appears only once, it is stored as-is.
mqtt5_user_properties_<key>_<idx>	User Properties from the PUBLISH packet can have the same key multiple times. If this happens they receive a zero-based index as a suffix
mqtt5_content_type	Content Type
mqtt5_payload_format_indicator	Payload Format Indicator
mqtt5_response_topic	Response Topic
mqtt5_correlation_data	Correlation Data
mqtt5_subscription_identifier	Subscription identifier sent by the broker

• New mqtt5 output config options:

Field	Notes
session_expiry_interval	How long the broker retains session state after disconnect.
message_expiry_interval	Per-message TTL. Broker drops the message if undelivered before expiry.
topic_alias_maximum	Maximum number of topic aliases this client will use when publishing. 0 disables topic aliases.
will_*	Full v5 will message config.

• Metadata read by the mqtt5 output:

Metadata key	Written to
mqtt5_user_properties_<key>	User Properties on the PUBLISH packet. Plain keys (mqtt5_user_properties_<key>) are written as single entries. Indexed keys (mqtt5_user_properties_<key>_0, _1, etc.) are reassembled into multiple User Properties with the same key, preserving order.
mqtt5_content_type	Content Type
mqtt5_payload_format_indicator	Payload Format Indicator
mqtt5_response_topic	Response Topic
mqtt5_correlation_data	Correlation Data

• Other MQTT v5 Behavior:
  • Shared Subscriptions are supported by the underlying library.
  • Reason codes will be surfaced in error messages.
  • Server-initiated DISCONNECT is handled by the underlying library.