High memory consumption (and message queue) in mqtt_node

[jo...@cloudamqp.com]

1-node, RabbitMQ 3.8.9, Erlang 22.2.7, "Internet of Things scenario" with many connections (50-60k), 50k queues, about 2-3k messages/s. After a few days or weeks of operation the node runs out of memory and restarts.

Almost all memory is used up by "mqtt_node" (see screenshot from rabbitmq-diagnostics observer). That PID also has quite a large internal queue.

We will try to replicate and submit an issue, but

a) known issue?

b) possible to restart the connection tracking without restarting RabbitMQ?

/Johan

[kjnilsson]

Looks like something is crashing. What's in the crash log?

[jo...@cloudamqp.com]

For the record (if you for instance stumble on this via a search engine at a later point) the discussion has moved to: https://github.com/rabbitmq/rabbitmq-server/discussions/2688