ECS Service auto scaling policy conflict the one to other policy

To keep it simple I suggest you use Target tracking scaling policy for the ECSServiceAverageMemoryUtilization metric and tick the Turn off scale-in checkbox. So you rely only on CPU utilization as a criteria for scaling in.

This is the exact situation where you should use Target Tracking.

Create 2 Target Tracking policies, one for CPU, and one for Memory.

When there are multiple target tracking policies, any policy being above the threshold can trigger scale-out; but scale-in only happens when all policies want to scale-in at the same time. This prevents the oscillation you're describing, since scale-in only happens when both Memory and CPU are low at the same time

Hello,

The Issue: Your ECS service has two conflicting auto scaling policies:

1. CPU Scaling Policy (Scale Out): Adds a task when CPU utilization exceeds 50%.
2. Memory Scaling Policy (Scale In): Removes a task when memory utilization falls below 30%.

These policies can lead to unnecessary scaling activity, where a task added due to high CPU might be immediately removed due to low memory.

Here are two approaches to address this:

Prioritize Scaling Policies:

Define a priority order for the policies. The policy with higher priority will take precedence. In this case, set CPU scaling to a higher priority than memory scaling. This ensures scaling based on CPU first, and memory scaling won't remove a task immediately after CPU scaling adds one.

Steps:

• Go to the ECS service in the AWS Management Console.
• Navigate to the "Auto Scaling" section.
• Click on the "Edit" button for each policy.
• Set a higher priority value (lower number) for the CPU scaling policy.

Combine Scaling Metrics:

Create a single policy that considers both CPU and memory utilization. This policy can add or remove tasks based on a combined threshold that factors in both metrics.

Steps:

• Go to the ECS service in the AWS Management Console.
• Navigate to the "Auto Scaling" section.
• Click on "Create a new policy."
• Choose "Target Tracking Scaling" policy type.
• Define a target for combined CPU and memory utilization (e.g., average CPU above 40% and memory above 40%).
• Set scaling actions (add/remove tasks) based on deviations from the target.

ECS Auto Scaling doc: https://docs.aws.amazon.com/AmazonECS/latest/developerguide/service-auto-scaling.html.