desired_capacity clashes with alarm-based scaling

[jsonmaur]

This is in reference to the ecs-cluster module.

I may be wrong on this (please correct me if I am), but doesn't setting desired_capacity on your autoscaling group conflict with the alarm-based scaling? If your asg is scaling up because of high cpu/memory reservation and you run terraform apply at the same time, it will knock your instance count back down to the number specified in your code (alarm-based scaling modifies the group's desired capacity). See this thread for reference.

desired_capacity is an optional argument for aws_autoscaling_group, so maybe it would be better to take it out (or at least not set a default).

[lukebarton]

At the risk of adding a +1 post, I too am interested in this.

From my understanding, the ASG will scale based on the CPU alert. However from my observations I think there are two issues:

• The Services themselves are limited to their predefined desired_count, meaning Services won't scale with the ASG - new instances will just be empty? Even if the ECS redistributes, this could quite easily lead to a situation where the ASG is constantly scaling up to the ecs_max_size due to "ServiceA" constantly maxing out the CPU because it's not actually scaling the service across newly available instances.
• Secondly, whenever terraform is applied, the ASG will be downscaled to the desired_capacity?

I'm concerned about both issues.