Automatically scale your Amazon ECS service

Use a target metric to scale Amazon ECS services— Increase or decrease the number of tasks that your service runs based on a target value for a specific metric. This is similar to the way that your thermostat maintains the temperature of your home. You select temperature and the thermostat does the rest.

Use predefined increments based on CloudWatch alarms to scale Amazon ECS services— Increase or decrease the number of tasks that your service runs based on a set of scaling adjustments, known as step adjustments, that vary based on the size of the alarm breach.

Use scheduled actions to scale Amazon ECS services—Increase or decrease the number of tasks that your service runs based on the date and time.

Use historical patterns to scale Amazon ECS services with predictive scaling—Increase or decrease the number of tasks that your service runs based on historical load data analytics to detect daily or weekly patterns in traffic flows.

Amazon ECS sends metrics in 1-minute intervals to CloudWatch. Metrics are not available until the clusters and services send the metrics to CloudWatch, and you cannot create CloudWatch alarms for metrics that do not exist.

The scaling policies support a cooldown period. This is the number of seconds to wait for a previous scaling activity to take effect.

The service scheduler respects the desired count at all times, but as long as you have active scaling policies and alarms on a service, Service Auto Scaling could change a desired count that was manually set by you.

If a service's desired count is set below its minimum capacity value, and an alarm initiates a scale-out activity, Service Auto Scaling scales the desired count up to the minimum capacity value and then continues to scale out as required, based on the scaling policy associated with the alarm. However, a scale-in activity does not adjust the desired count, because it is already below the minimum capacity value.

If a service's desired count is set above its maximum capacity value, and an alarm initiates a scale in activity, Service Auto Scaling scales the desired count out to the maximum capacity value and then continues to scale in as required, based on the scaling policy associated with the alarm. However, a scale-out activity does not adjust the desired count, because it is already above the maximum capacity value.

During scaling activities, the actual running task count in a service is the value that Service Auto Scaling uses as its starting point, as opposed to the desired count. This is what processing capacity is supposed to be. This prevents excessive (runaway) scaling that might not be satisfied, for example, if there aren't enough container instance resources to place the additional tasks. If the container instance capacity is available later, the pending scaling activity may succeed, and then further scaling activities can continue after the cooldown period.

If you want your task count to scale to zero when there's no work to be done, set a minimum capacity of 0. With target tracking scaling policies, when actual capacity is 0 and the metric indicates that there is workload demand, Service Auto Scaling waits for one data point to be sent before scaling out. In this case, it scales out by the minimum possible amount as a starting point and then resumes scaling based on the actual running task count.

Application Auto Scaling turns off scale-in processes while Amazon ECS deployments are in progress. However, scale-out processes continue to occur, unless suspended, during a deployment. For more information, see Service auto scaling and deployments.

You have several Application Auto Scaling options for Amazon ECS tasks. Target tracking is the easiest mode to use. With it, all you need to do is set a target value for a metric, such as CPU average utilization. Then, the auto scaler automatically manages the number of tasks that are needed to attain that value. With step scaling you can more quickly react to changes in demand, because you define the specific thresholds for your scaling metrics, and how many tasks to add or remove when the thresholds are crossed. And, more importantly, you can react very quickly to changes in demand by minimizing the amount of time a threshold alarm is in breach.