using the rabbitmq scaler, why does it scale up to max if there are no messages?

[Loki-Afro]

hi there,

i'm trying to figure out why after all messages in a queue have been processed without reaching the maxReplicaCount it still goes up to maxReplicaCount instead of scaling down

what i want:

• scale to 0 if there are no messages in certain queue
• scale maximum up to 40, if there are messages in the queue (it doesn't need to be abruptly 40)
• scale down of there are no messages in the queue

some extra info:

• usually there are no messages in the queue, they get created via another job, like every few hours a few thousands up to many thousands
• each messages takes way under a second to process
• prefetch count is set to 25
• in my test scenario there are ~2700 messages, so really not much

this is my scaled object:

apiVersion: keda.sh/v1alpha1
kind: ScaledObject
metadata:
  name: api-ldap-worker-scaledobject
  namespace: bc-6559
  labels:
    app: api-worker
spec:
  scaleTargetRef:
    name: api-worker-deployment
  idleReplicaCount: 0
  minReplicaCount: 1
  maxReplicaCount: 40
  triggers:
  - type: rabbitmq
    metadata:
      protocol: amqp
      queueName: sync_ldap
      mode: QueueLength
      value: "1"
    authenticationRef:
      name: rabbitmq-trigger-auth

here is what i observed:

• scale up to 1
• after 30 seconds scale up to 4
• all messages are processed already (also no non-acknowledged are left)
• 30 seconds after it has been scaled up to 4 it scales it up to 8
• 30 seconds later 16
• 30 seconds later 32
• 30 seconds later 40
• 30 seconds later 0

now my question: why is it not scaling back to 0, if there are no messages left?