We are upgrading from rabbitmq 3.9 to RabbitMQ 4.0.5 and Erlang 26.2.5.7 on Ubuntu 22 running on an AWS EC2 i3 instance with 1 x 950 NVMe SSD. The mnesia directory is on the NVMe SSD.
hard and soft ulimit: 65000
Our setup includes 400 classic queues, and we are observing unexpected behaviour when handling persistent messages:
- Persistent messages are quickly cleared from disk when publishing to a single queue.
- However, when publishing to multiple queues, the persistent messages remain on disk for a significantly longer time before being cleared.
This slow release of disk space only occurs when publishing to multiple queues simultaneously.
Any help would be appreciated.
