Scheduled scaling as an addon

[Lou1415926]

• The solution presented below is a starting point from which users can build their own scheduled scaling addons. Users need to modify the template to work for their requirements.
• The solution is built on top of the addons feature.
• Related: Modify datapoint(Evaluation Period, Datapoints) in cloudwatch alram. #4233

---

1. Define copilot/<svc>/addons/addons.parameters.yml

Parameters:
  ServiceName: !GetAtt Service.Name

2. Define copilot/<svc>/addons/scheduled-scaling.yml

Parameters:
  # Required parameters by AWS Copilot.
  App:
    Type: String
  Env:
    Type: String
  Name:
    Type: String
  # Additional parameters defined in addons.parameters.yml
  ServiceName:
    Type: String

Type: AWS::ApplicationAutoScaling::ScalableTarget
Properties: 
  MaxCapacity: 1
  MinCapacity: 1o
  ResourceId: !Ref ServiceName
  RoleARN: ...
  ScalableDimension: ecs:service:DesiredCount
  ScheduledActions: 
    - ScalableTargetAction:
        MinCapacity: 10 # Scale to 10
        MaxCapacity: 10
      Schedule: cron(...)
      ScheduledActionName: ScaleTo10EverySaturdayFrom9to18
  ServiceNamespace: ecs

[erno]

Thanks for this example. When applying it I got this kind of error:

ValidationError: Circular dependency between resources: [AddonsStack, Service, TaskDefinition, TaskRole]

Putting in the ResourceId as a hardcoded string seemed to get over it (enough to test in a dev env). I also tried to change the param name from ServiceName to ScalingServiceName in case there was some other internal usage of that identifier, but to no avail.. any ideas?

[KollaAdithya]

Hello @erno !

can you take a look at this workaround in order to avoid circular dependencies.
#3149 (comment)

[erno]

Hi, thanks for the pointer!

I think that workaround looks too hairy for me now, and also I want to also stop the Aurora Serverless v2 cluster along with Fargate tasks, so going with a non-copilot solution for now: Github Actions scheduled workflow that uses aws ecs update-service --desired count 0 ... and aws rds stop-db-cluster in the evenings and the other way in the mornings.