Autoscaling

Configuring Scheduler Autoscaling

Prerequisites

• Have a game room container image that communicates with maestro through Maestro's rooms API

Learning Outcomes

After finishing this tutorial you will understand how:

• to configure autoscaling policies for your scheduler

What is

Autoscaling is an optional feature in which the user can choose and parametrize different autoscaling policies that maestro will use to automatically scale the number of rooms in the scheduler.

Maestro has an internal process that periodically keeps checking if it needs to create or delete game rooms for the given scheduler, if autoscaling is not configured or enabled, it will always try to maintain the current number of rooms equal to roomsReplicas scheduler property. If autoscaling is configured and enabled, it will use the configured autoscaling policy to decide if it needs to scale up (create more rooms), scale down (delete rooms) or do nothing.

Currently, the sync interval is configured by environment variable MAESTRO_WORKERS_HEALTHCONTROLLERINTERVAL.

By default, the scheduler does not have autoscaling configured.

How to configure and enable autoscaling

To get autoscaling working in your scheduler, firstly you need to configure an autoscaling policy and enable it, this autoscaling configuration resides in the root of the scheduler structure itself.

name: String
game: String
...
autoscaling:
  enabled: true
  min: 1
  max: 10
  cooldown: 60
  policy:
    type: roomOccupancy
    parameters:
      ...
      // Will vary according to the policy type.
        

{
  "name": "test",
  "game": "multiplayer",
  ...
  "autoscaling": {
    "enabled": true,
    "min": 10,
    "max": 300,
    "cooldown": 60,
    "policy": {
      "type": "roomOccupancy",
      "parameters": {
        ...
        // Will vary according to the policy type.
      }
    }
  }
}
        

• enabled [boolean]: A value that can be true or false, indicating if the autoscaling feature is enabled/disabled for the given scheduler. Default: false.
• min [integer]: Minimum number of rooms the scheduler should have, it must be greater than zero. For zero value, disable autoscaling and set "roomsReplicas" to 0.
• max [integer]: Maximum number of rooms the scheduler can have. It must be greater than min, or can be -1 (to have no limit).
• cooldown [integer]: The cooldown period (in seconds) between downscaling operations. This is useful to avoid scale down too fast, which can cause a lot of rooms to be deleted in a short period of time.
• policy [struct] : This field holds information regarding the autoscaling policy that will be used if the autoscaling feature is enabled:
• type [string]: Define the policy type that will be used, must be one of the policy types maestro provides.
• parameters [struct]: This field will contain arbitrary fields that will vary according to the chosen policy type.

Policy Types

Maestro has a set of predefined policy types that can be used to configure the autoscaling, each policy will implement a specific strategy for calculating the desired number of rooms and will have its configurable parameters.

Room Occupancy Policy

The basic concept of this policy is to scale the scheduler up or down based on the actual room occupancy rate, by defining a "buffer" percentage of ready rooms that Maestro must keep. The desired number of rooms will be given by the following formula:

desiredNumberOfRooms = ⌈(numberOfOccupiedRooms/ (1- readyTarget) )⌉

So basically Maestro will constantly try to maintain a certain percentage of rooms in ready state, by looking at the actual room occupancy rate (number of rooms in occupied state).

Room Occupancy Policy Parameters

• readyTarget [float]: The percentage (in decimal value) of rooms that Maestro should try to keep in ready state, must be a value greater than 0 and less than 1.
• downThreshold [float]: It adjusts how often maestro scale down Game Rooms, where 0.99 means that maestro will always scale down a Game Room when it is free (respecting the readyTarget), and a value close to 0 means that maestro will almost never scale down. Must be a value greater than 0 and less than 1.

Example

name: String
game: String
...
autoscaling:
  enabled: true
  min: 1
  max: 10
  policy:
    type: roomOccupancy
    parameters:
      roomOccupancy:
        readyTarget: 0.5
        

{
  "autoscaling": {
    "enabled": true,
    "min": 10,
    "max": 300,
    "policy": {
      "type": "roomOccupancy",
      "parameters": {
        "roomOccupancy": {
          "readyTarget": 0.5
        }
      }
    }
  }
}
        

Below are some simulated examples of how the room occupancy policy will behave:

Note that the autoscaling decision will always be limited by the min-max values! .