The following key points will give you more information about App Service plan autoscaling:
- You cannot assign more than one autoscale setting to one resource.
- When you have more than one instance, the threshold will be calculated according to the average of the metric across all of the instances. This is done to decide whether to scale out/in.
- If you want to investigate an autoscale operation failure, you can use the activity log.
- Make sure that you have different values for the minimum and maximum number of instances with a reasonable margin.
- If you have manually scaled out/in at the same time that you have autoscale rules, the autoscale rules will overwrite the manual scaling you have done.
- Ensure that you have scale out and scale in rules so that you can get benefits from the usage savings. For example, when the App Service plan scales out after hitting a threshold, it can be scaled in again after going below another threshold that indicates that the resource usage is quite acceptable to be scaled in.
- It is not recommended to set the scale out/in rules threshold when it goes above or below the same value. For example, do not make the threshold for the scale out when it is above 70% and scale in when it is below 70%.
- If you have added multiple rules to the same autoscale setting, it will scale out when any scaleout rule is met, but it will not scale in until all of the scale-in rules are met to maintain the performance and the availability of the apps. For example, if we have different scale in rules, such as a scale-in rule for the CPU and another one for the memory, and then one of these rules is met but the other is not. In this case, the scale in will not be triggered to maintain the performance because we have another rule that would affect the performance that not been met yet.
- When the metrics are not available, it will use the default number of instances. Therefore, make sure that you have set a reasonable number of instances so that you can get your apps up and running with no negative impact on performance.