How does Load Based Auto Scaling Work?

Load based auto scaling is defined on the role level. Scalarium computes the average CPU, memory, and load consumption over all online instances in this role and compares those values against your defined thresholds.

If one of the upscaling thresholds is met, Scalarium will start a number of instances (so at least one metric needs to be defined). How many instance will be started depends on the batch size you specify. Scalarium will start any non-running instances out of the pool of defined load based auto scaling instances of this role. So only specific load based instances will be started, not normal '24x7' instances or time based auto scaling instances.

After Scalarium started the instances, it will ignore any metrics of the role for the specified duration. Once this timespan is over, Scalarium will again compare the metrics to the thresholds and maybe start additional instance.

If you have no non-running load based instances left, Scalarium will not do anything.

If the upscaling thresholds are not met, Scalarium will compare the metrics against the downscaling metrics. If any of the averages is below a defined threshold, Scalarium will stop the defined number of instances. Only running load based auto scaling instances will be stopped. Not normal '24x7' instances or time based auto scaling instances.

After stopping the instances, Scalarium will again ignore any metrics for the specified duration. Afterwards we begin a new cycle again where upscaling or downscaling can happen.

If neither the upscaling nor downscaling thresholds are breached, no action will be taken.