Queue-based autoscaling

As a service grows, scaling becomes less about raw resource usage and more about how much work is waiting to be done. A system can appear underutilized while a queue is building up in the background. CPU and memory are often lagging metrics and don’t always reflect this kind of demand. To address this, we’ve introduced queue-based autoscaling triggers that adjust capacity in response to pending work.

Queue triggers

Two new trigger types are available: RabbitMQ and Temporal. Each monitors a queue of pending tasks and scales appropriately as work arrives. These can be combined with the CPU and Memory triggers, allowing your service to respond to both queue depth and resource usage.

Triggers are available in both the UI and YAML configuration. Open Scaling in a server installation’s settings to configure them visually, or add a triggers block to your YAML configuration:

Services can also scale down to zero replicas when there’s no work to process.