Scaling Modifiers to intervene from CRON trigger

[paragbhingre]

Hi All,

I’m currently using CRON trigger to scale up and down my resources like given below. And I would like to use scaling modifiers such that if we are under CRON time schedule and cpu utilization is less than 75% then I want to the new desiredReplicas. Would that be possible? If yes, what formula I would need?

triggers: - metadata: Value: 75 metric type: Utilization name: metric_one type: cpu - metadata: desiredReplicas: 5 start: 10 10 * * * end: 10 12 * * * time zone: America/New_York name: metric_two type: CRON

I’m new to building formulas in scalingmodifiers so tried couple of ways but didn’t work.

Thank you for your help in advanced.

[SpiritZhou]

It is not possible to combine CPU scaler and cron scaler as the scaling modifiers do not support CPU/memory scaling. By the way, why do you want to scale when the CPU usage is less than 75%? Normally, scaling is required when CPU utilization exceeds a certain percentage.

[dogzzdogzz]

I have similar requirement, here is my use case, I want to enable the autoscale based on rps during peak hours (8-22) only, and scale down to 0 during off-peak hours for saving cost. Is it feasible?

  minReplicaCount: 0
  maxReplicaCount: 3    
  triggers:
  - metadata:
      desiredReplicas: "1"
      start: 0 8 * * *
      end: 0 22 * * *
      timezone: Asia/Singapore
    name: cron
    type: cron
  - metadata:
      query: rps
      serverAddress: http://prometheus
      threshold: "10"
    name: rps
    metricType: AverageValue
    type: prometheus
  advanced:
    scalingModifiers:
      formula: "cron > 0 ? rps : 0"
      target: "10"
      activationTarget: "10"
      metricType: "AverageValue"

[paragbhingre]

I have similar requirement, here is my use case, I want to enable the autoscale based on rps during peak hours (8-22) only, and scale down to 0 during off-peak hours for saving cost. Is it feasible?

minReplicaCount: 0
maxReplicaCount: 3
triggers:

• metadata:
  desiredReplicas: "1"
  start: 0 8 * * *
  end: 0 22 * * *
  timezone: Asia/Singapore
  name: cron
  type: cron
• metadata:
  query: rps
  serverAddress: http://prometheus
  threshold: "10"
  name: rps
  metricType: AverageValue
  type: prometheus
  advanced:
  scalingModifiers:
  formula: "cron > 0 ? rps : 0"
  target: "10"
  activationTarget: "10"
  metricType: "AverageValue"

I’ve the similar case too. I scale up during peak hours 9-16 but if CPU/Memory utilization is not above 75% during peak hours then I would like to scale down by half. How is that possible?

[zroubalik]

@dogzzdogzz or you could use something like http scaler, to allow you scale to zero.

@paragbhingre unfortunately cpu/mem scalers can't be used in scaling modifiers

[dogzzdogzz]

@zroubalik My workaround would be to send a PEAK_HOUR custom metric to Prometheus and leverage a scaling modifier as shown below. However, either this approach or http scaler requires a custom logic to generate and send the metric, and could be complex if we need different time window for different SO. I'd like to explore the possibility of achieving a simpler solution using KEDA's built-in support for custom scaling logic based on a specific time window.

  minReplicaCount: 0
  maxReplicaCount: 3    
  triggers:
  - metadata:
      query: peak_hour
      serverAddress: http://prometheus
      threshold: "10"
    name: peak_hour
    metricType: AverageValue
    type: prometheus
  - metadata:
      query: rps
      serverAddress: http://prometheus
      threshold: "10"
    name: rps
    metricType: AverageValue
    type: prometheus
  advanced:
    scalingModifiers:
      formula: "peak_hour > 0 ? rps : 0"
      target: "10"
      activationTarget: "10"