Converting the targets to its original value, it says that the ai-worker has an average of 75.6 messages per replica, and video-worker 90.9 messages. The target queue length or message count per each replica is 5. This means a replica is approximately processing 15x more than the QueueLength specified in the ai-worker’s ScaledObject. This is what’s referred to as autoscaling saturation.
k get hpa
NAME REFERENCE TARGETS MINPODS MAXPODS REPLICAS AGE
keda-hpa-ai-worker Deployment/ai-worker 75600m/5 (avg) 1 10 10 47h
keda-hpa-video-worker Deployment/video-worker 90900m/5 (avg) 1 10 10 47h