I am implementing auto scaling for my ECS Fargate services on AWS, using App Auto Scaling. I am running into an issue where containers starting up cause a spike in resource usage (on start up), triggering unnecessary auto scaling.
In EC2 auto scaling groups, there is a parameter to set the estimated warmup time, preventing instances that are starting up from being included in CloudWatch metrics.
Is it possible to implement something similar for ECS?
Amazon ECS Cluster Auto Scaling supports specifying a custom instance warm-up time.
When creating or updating an ECS capacity provider, you can use managedScaling.instanceWarmupPeriod (time in seconds) using the auto-scaling-group-provider option.