../_images/preview.png

Scaling Applications

Scalr support two types of scaling, Auto and Manual Scaling. Autoscaling is the best option for applications that are cloud aware, while manual scaling may be the better option for a legacy application that has not been rearchitected for the cloud.

Autoscaling Mode

There are two distinct but related ways to use Autoscaling Mode:

  • With Simple Autoscaling, Scalr will ensure that you always have at least the minimum and no more than the maximum number of servers up at all times.
  • With Dynamic Autoscaling, Scalr will additionally monitor metrics on your servers and dynamically add or remove servers depending on the specified metric threshold (load, response time, etc).

Manual Scaling Mode

In Manual Scaling Mode, Scalr does not automatically launch or terminate Servers. A user must launch the Farm and then manually launch a server based on a farm role.

Auto Scaling

Definition and Scope

SCOPE_F_ROLE

Scalr has autoscaling built directly into the product, which introduces scaling for those clouds that do not have it as well maintains a consistent experience across all clouds. There are two types of autoscaling:

  • Simple Autoscaling: Ensure that you always have a minimum number of servers up at all times.
  • Dynamic Autoscaling: Using defined metrics, Scalr will scale up or down your Farm Role(s).

Simple Autoscaling

Simple Autoscaling will perform the following actions:

If your Running Server Count is lower than the minimum you requested, additional Servers will be launched. Such a situation could be caused by:

  • Terminating a server
  • Suspending a server
  • A server crashing
  • Increasing the MIN Instances by editing the Farm Role

If your Running Server Count is higher than the Maximum you requested, existing Servers will be terminated. Such a situation could be caused by:

  • Manually launching more servers in a Farm Role without updating your scaling settings.
  • Decreasing the MAX Instances by editing the Farm Role

To enable simple autoscaling in your Farm Role, find your Farm, click on the Farm Role, go to the Scaling tab and make sure it is set to Automatic:

../_images/simple_scaling.png

In the scenario above, Scalr will always have minimum of one server and a maximum of two for this Farm Role

Dynamic Autoscaling

Dynamic Autoscaling does everything Simple Autoscaling does, except it also adds and removes Servers based on utilization or activity (e.g. CPU utilization). Scalr will poll for metrics at least every minute, depending on your configuration, and uses those metrics to make the scaling decisions.

To enable dynamic autoscaling in your Farm Role, find your Farm, click on the Farm Role, go to the Scaling tab and make sure it is set to Automatic. Next, click on the Add Scaling Rule Option which will then present the metrics with which the Server can scale by:

../_images/dynamic_scaling.png
Metric Description
Load Average Load Averages can be used to scale up and down based on average system utilization. The higher the Load Average, the busier the system is. Note: Available for Linux only.
Free RAM Free RAM can be used to scale up and down based on system memory utilization. The lower the Free RAM, the busier the system is. Make sure you set thresholds that make sense given the Instance Type you chose. By default, Scalr collects only Free RAM, but you can have Scalr account for cached RAM too.
URL Response Time URL Response Time lets you scale up and down depending on how long a particular URL takes to load. The URL Response Time is collected from the Scalr Server, not from individual Servers.
SQS Queue Size SQS Queue Size can be used to scale up and down based on an AWS SQS Queue’s Size. Queue Size is checked from the Scalr Server itself.
Date and Time Using the Date and Time Metric, simply define a schedule and Scalr will ensure that an appropriate number of Servers are deployed according to that schedule. Note: The Date and Time Metric may not be used in combination with other metrics.
Bandwidth Bandwidth can be used to scale up and down based on network activity. You can decide to use either inbound or outbound network bandwidth.

Note

Dynamic Autoscaling is only enabled for Farm Roles that are using the Scalarizr agent.

Autoscaling Configuration

Along with metrics, Scalr also has many configurations that can be set for scaling decisions and termination preferences:

Scaling Decision Settings:

../_images/scaling_decisions.png

The options in the screenshot are defined as follows (in order):

  • The duration between scaling decisions can be changed, the default is 1 minute.
  • You can override the timing for a scale up decision.
  • You can override the timing for a scale down decision.
  • You can pause the scaling after launching a new server, and only resume the cycle when the previously launched server reaches a running state.
  • Scaling down will only occur if ALL of the scaling metrics that have been defined return true. For example, you may have a URL and a Load Average scaling rules, both of those metrics must hit the lower threshold before scaling down occurs.

Scaling Termination Settings:

../_images/scaling_termination.png

The options in the screenshot are defined as follows (in order):

  • Scaling Behaviour:
    • Launch / Terminate: As scaling up occurs, a brand new server will launch. As scaling down occurs, the server will be terminated.
    • Resume / Suspend : As scaling up occurs, a server that was suspended will be turned on. As scaling down occurs, the server will be suspended.
  • Consider Suspended Servers:
    • Running: If a server is suspended, that server is considered running in the min/max scaling count.
    • Terminated: If a server is suspended, that server is considered terminated in the min/max scaling count.
  • The servers which have been launched the most recently will be scaled down first.
  • By enabling safe shutdown, Scalr will run an “auth-shutdown” script on every Server before terminating it. Depending on the Script’s return code, Scalr may abort the scale-down procedure. This option is useful to prevent a Server from being terminated while performing a long-running task. Scalr will terminate an instance ONLY IF the script /usr/local/scalarizr/hooks/auth-shutdown returns 1. If this script is not found or returns any other value, Scalr WILL NOT terminate that server.

Custom Scaling

SCOPE_SCALR SCOPE_ACC SCOPE_ENV

Scalr also provides a Custom Scaling option to leave it up to a user to decide when a scaling decision should happen. In this scenario, Scalr reads a file, executes a file, or does a URL Request and based on that value it will scalr up or down depending on the calculation function.

To create a Custom Scaling Metric, click on the main Scalr menu on the top left MENU_ENV, and click on Custom Scaling Metrics. On that page, click on the New Metric button:

../_images/custom_scaling.png
Retrieve Method Description
File-Read Using File-Read, Scalr will try to read a given file on your Server, and expect to find a number there. That number will be used as the Scaling Metric’s value.
File-Execute Using File-Execute, Scalr will try to execute a given script on your Server, and expect the script to output a number to standard output. That number will be used as the Scaling Metric’s value.
URL-Request (GET) Using URL-Request, Scalr will send a HTTP GET request to the specified URL and expect a number in the returned content body. That number will be used as the Scaling Metric’s value.

The URL-Request method will return a single value from the target URL, but File-Read and File-Execute methods must run against all Farm Role instances. Each instance will return their own value and Scalr will compute a single result value from them according to your Calculation Function.

The following Calculation Functions are available for File-Read and File-Get:

Calculation Function Description
Average The metric’s value will be averaged from all servers.
Sum The metric’s value will be the sum of all results from all servers.
Maximum The metric’s value will be the largest value from any one of the servers.
Minimum The metric’s value will be the smallest value from any one of the servers.

For example, say you had 3 running instances of a Farm Role using a custom metric with File-Read method. Scalr polls for the metric and the instances return 2, 3, and 4 respectively. The table below shows the resulting metric value for each calculation type.

Calculation Function Final metric value
Average 3
Sum 9
Maximum 4
Minimum 2

Once you have created the custom metric it will then appear in the metrics list in the scaling section in each farm role. In the Scaling Metric, select your customer metric and give it rules. The scaling decisions are made based on the custom metric value going above or below a certain value:

../_images/custom_scaling_fr.png

Manual Scaling

Overview

SCOPE_F_ROLE

For those use cases where you may not have an application that has a fully automated deployment or it just doesn’t need to be autoscaled, Scalr has the option to allow users to initiate the provisioning of servers.

Using Manual Scaling

To enable manual scaling in your Farm Role, find your Farm, click on the Farm Role, go to the Scaling tab and make sure it is set to Manual:

../_images/manual_scaling.png

Manually Launching Servers

Once you have set the the Farm Role to manual scaling and launched your Farm, you will now have to manually launch a server. To do so, go to the Farms page, find your Farm, and click on the number in the Roles ROLE column (Number 3 in the example below):

../_images/dashboard_roles_column.png

That will bring you to the Farm Roles pages for that specific Farm. Within this page, click on the launch LAUNCH button for the Farm Role that you want provision a server for on the right-hand side:

../_images/farmrole_launch.png

Manually Terminating Servers

Just like you have to manually launch a server when scaling is set to manual, the same applies to terminating servers. To terminate a server, go to the servers pages, find the server you want to terminate and click on the Terminate Server button:

../_images/terminate_server.png