Configuring AutoScale without using NetScaler¶
This feature is currently only available on the master branch and will be released in the 4.4 release.
What is AutoScaling?¶
AutoScaling allows you to scale your back-end services or application VMs up or down seamlessly and automatically according to the conditions you define. With AutoScaling enabled, you can ensure that the number of VMs you are using seamlessly scale up when demand increases, and automatically decreases when demand subsides. Thus it helps you save compute costs by terminating underused VMs automatically and launching new VMs when you need them, without the need for manual intervention.
At that time, AutoScaling without NetScaler only supports for Xenserver. We are working to support KVM also.
Before you configure an AutoScale rule, consider the following:
- Ensure that the necessary template is prepared before configuring AutoScale. Firstly you must install the PV-driver, which helps Xenserver collect performance parameters (CPU and memory) into VMs. Beside, When a VM is deployed by using a template and when it comes up, the application should be up and running.
Specify the following:
Template: A template consists of a base OS image and application. A template is used to provision the new instance of an application on a scaleup action. When a VM is deployed from a template, the VM can start taking the traffic from the load balancer without any admin intervention. For example, if the VM is deployed for a Web service, it should have the Web server running, the database connected, and so on.
Compute offering: A predefined set of virtual hardware attributes, including CPU speed, number of CPUs, and RAM size, that the user can select when creating a new virtual machine instance. Choose one of the compute offerings to be used while provisioning a VM instance as part of scaleup action.
Min Instance: The minimum number of active VM instances that is assigned to a load balancing rule. The active VM instances are the application instances that are up and serving the traffic, and are being load balanced. This parameter ensures that a load balancing rule has at least the configured number of active VM instances are available to serve the traffic.
Max Instance: Maximum number of active VM instances that should be assigned to a load balancing rule. This parameter defines the upper limit of active VM instances that can be assigned to a load balancing rule.
Specifying a large value for the maximum instance parameter might result in provisioning large number of VM instances, which in turn leads to a single load balancing rule exhausting the VM instances limit specified at the account or domain level.
Specify the following scale-up and scale-down policies:
Duration: The duration, in seconds, for which the conditions you specify must be true to trigger a scaleup action. The conditions defined should hold true for the entire duration you specify for an AutoScale action to be invoked.
Counter: The performance counters expose the state of the monitored instances. We added two new counter to work with that feature:
- Linux User CPU [native] - percentage
- Linux User RAM [native] - percentage
Remember to choose one of them. If you choose anything else, the autoscaling will not work.
Operator: The following five relational operators are supported in AutoScale feature: Greater than, Less than, Less than or equal to, Greater than or equal to, and Equal to.
Threshold: Threshold value to be used for the counter. Once the counter defined above breaches the threshold value, the AutoScale feature initiates a scaleup or scaledown action.
Add: Click Add to add the condition.
Additionally, if you want to configure the advanced settings, click Show advanced settings, and specify the following:
Polling interval: Frequency in which the conditions, combination of counter, operator and threshold, are to be evaluated before taking a scale up or down action. The default polling interval is 30 seconds.
Quiet Time: This is the cool down period after an AutoScale action is initiated. The time includes the time taken to complete provisioning a VM instance from its template and the time taken by an application to be ready to serve traffic. This quiet time allows the fleet to come up to a stable state before any action can take place. The default is 300 seconds.
Destroy VM Grace Period: The duration in seconds, after a scaledown action is initiated, to wait before the VM is destroyed as part of scaledown action. This is to ensure graceful close of any pending sessions or transactions being served by the VM marked for destroy. The default is 120 seconds.
Apply: Click Apply to create the AutoScale configuration.
Disabling and Enabling an AutoScale Configuration¶
If you want to perform any maintenance operation on the AutoScale VM instances, disable the AutoScale configuration. When the AutoScale configuration is disabled, no scaleup or scaledown action is performed. You can use this downtime for the maintenance activities. To disable the AutoScale configuration, click the Disable AutoScale button.
The button toggles between enable and disable, depending on whether AutoScale is currently enabled or not. After the maintenance operations are done, you can enable the AutoScale configuration back. To enable, open the AutoScale configuration page again, then click the Enable AutoScale button.
Updating an AutoScale Configuration¶
You can update the various parameters and add or delete the conditions in a scaleup or scaledown rule. Before you update an AutoScale configuration, ensure that you disable the AutoScale load balancer rule by clicking the Disable AutoScale button.
After you modify the required AutoScale parameters, click Apply. To apply the new AutoScale policies, open the AutoScale configuration page again, then click the Enable AutoScale button.
An administrator should not assign a VM to a load balancing rule which is configured for AutoScale.
Making API calls outside the context of AutoScale, such as destroyVM, on an autoscaled VM leaves the load balancing configuration in an inconsistent state. Though VM is destroyed from the load balancer rule, it continues be showed as a service assigned to a rule inside the context of AutoScale.