Skip to main content
Version: Develop

a9s Messaging Resources Considerations

This document describes concerns and limitations to be considered when allocating resources for your a9s Messaging plans.

Disk and Memory Usage

RabbitMQ uses both disk and memory to store queues, and some configuration parameters might affect the performance and usability. a9s Messaging configures the process with vm_memory_high_watermark.relative = 0.42 and vm_memory_high_watermark_paging_ratio = 0.5, when the process starts using more than half of 42% of the total memory (21% of the total memory), RabbitMQ starts to move messages from memory to the disk.

E.g., in a 2GB of memory machine, with 1GB of persistent disk, your total memory usage for RabbitMQ would be 860MB. If this limit hits, users connection will be blocked until there is memory available again. Bear in mind that the garbage collector and other internal RabbitMQ processes can take up to double the current memory usage. Memory for a9s Logstash, a9s Backup Agent, and a9s Consul must also be reserved.

Once usage hits 430MB, messages start to get moved to the disk, giving a maximum of 1.43GB of available space (shared across connections, messages, etc.) for the RabbitMQ process, right before the writers connections start to get blocked. Still, before this reaches this critical point, a9s-parachute should kick in (when storage hit 80%), and the process will be stopped.

Keep in mind that the disk space is also shared with durable and lazy queues.

Furthermore, RabbitMQ requires (disk_free_limit_mem_relative * total memory) GB as reserved disk space in order to ensure proper function. The default value for the disk_free_limit_mem_relative parameter is 1.5, the minimum recommended value is 1.0. E.g. when disk_free_limit_mem_relative is set to 1.0 in a 10GB of memory machine, with 15GB of disk space the actual usable disk space is around 5GB. When the available disk space is under this threshold the messaging broker will start blocking clients.

For further information, consult the RabbitMQ [runtime](https://www.rabbitmq.com/runtime.html documentation).