With Auto Scaling (AS), you can increase or decrease CVMs in a scaling group based on monitoring metrics. You only need to define an alarm-triggered policy, specifying the status of the monitoring metrics that trigger scaling and the related scaling activity.
You need to specify the conditions and actions when creating an alarm policy, as shown in the figure below:
We recommend you create two policies for each scaling group, one for scale-out and one for scale-in. Once the traffic to your web application reaches the threshold of the alarm policy, AS executes the associated policy to scale your group in (by terminating instances) or out (by launching instances).
For example, assume you have an e-commerce web application that currently runs on five instances. You plan to carry out a promotional activity and are concerned that the access traffic might be much greater than you expect. In this case, you can configure a scaling group to add two new instances when the load on the current instances reaches 70%, and terminate extra the instances when the load decreases to 40%. This is shown in the figure below:
For the proper running of your existing business, if the CVMs in the cluster are used for the following purposes, you need to prevent them from being removed by the scale-in policy:
Multiple purposes: apart from the tasks specified by the cluster, a CVM in the cluster is also used for other purposes, for example the CVM is used as both a cache server and a file server.
Data storage: the CVM is stateful or stores data that other CVMs do not have. For example, the CVM stores the incremental data of other running CVMs in a cluster.
Image/Snapshot updates: the CVM is used to regularly update images and snapshots.