Setting up autoscaling

Now that your packaged game server has been successfully deployed and tested on the cluster, you can set up a fleet of game server containers on Agones and configure Agones to autoscale them based on demand.

Opening the firewall on the Kubernetes control node

To date, we've created all of the Agones/Kubernetes resources by using the SSH terminal on the Kubernetes control node.

For obvious reasons, this doesn't work long term, especially when you're deploying your game servers from a build server. Instead, we'll open the Kubernetes API port on the control node so that we can access it from our local machine and build servers.

Identifying your IP address

You'll need to know your IP address to allow your computer direct access to the Kubernetes API. You can use Google to find your IP address.

Open the firewall permit connections from your local machine

On the Kubernetes control node, run the following command, replacing 1.1.1.1 with the IP address Google gave you in the last step. Replace eth1 with the public interface name as identified in Securing your network.

ufw allow in on eth1 from 1.1.1.1 proto tcp to any port 6443

Copy the kubeconfig file from the control node

caution

The k3s.yaml file grants full administrative permissions to your Kubernetes cluster. If you're a company with multiple employees, you might want to issue certificates for each employee and utilize RBAC to lock down permissions. However, issuing user certificates is outside the scope of this guide.

On the Kubernetes control node, print the contents of the k3s.yaml file so you can copy it's contents to a file on your local machine:

cat /etc/rancher/k3s/k3s.yaml

You'll get a file that looks like this:

apiVersion: v1
clusters:
- cluster:
    certificate-authority-data: ...
    server: https://127.0.0.1:6443
  name: default
contexts:
- context:
    cluster: default
    user: default
  name: default
current-context: default
kind: Config
preferences: {}
users:
- name: default
  user:
    client-certificate-data: ...
    client-key-data: ...

You'll need to change https://127.0.0.1:6443 to the public IP address of the control node. For example, if the public IP was 1.1.1.1, you would change it to:

clusters:
- cluster:
    certificate-authority-data: ...
    server: https://1.1.1.1:6443
  name: default

Then to utilize this file, you have a few options:

• If you don't need to connect to any other Kubernetes cluster, you can just save this file as ~/.kube/config (%userprofile%\.kube\config on Windows).
• If you already have a Kubernetes configuration file, you can rename the default values to something more identifying (like west-europe-01) and then merge the cluster, context and user values into your existing Kubernetes file.
• Or for all of the commands below, you can pass --kubeconfig=path/to/file as an additional parameter right after kubectl.

Test your connection to the Kubernetes control node

Make sure you can successfully see the Kubernetes pods from your local machine by running:

kubectl get pods

You should be able to see the Agones-related pods, even if you haven't deployed any game servers.

Defining the game server fleet

On your local machine, create a game server fleet using the following YAML as a template:

apiVersion: "agones.dev/v1"
kind: Fleet
metadata:
  # Change this as appropriate. If you roll out a fleet with the same name, it will
  # upgrade the existing fleet. If you want to run multiple versions side-by-side, use
  # a unique name for each version.
  name: eos-dedicated-server
spec:
  # The number of replicas by default. This isn't super important, as we'll control
  # this with the autoscaler in a moment anyway.
  replicas: 2
  template:
    spec:
      # Set up your game ports as appropriate.
      ports:
      - name: game
        containerPort: 7777
      - name: beacon
        containerPort: 12345
      template:
        spec:
          # Don't forget to use your GitLab pull secret.
          imagePullSecrets:
          - name: gitlab
          containers:
          - name: eos-dedicated-server
            # Update the image to the URL you pushed your packaged game server to.
            image: registry.gitlab.com/redpointgames/internal/eos-game-server:latest
            resources:
              requests:
                # Adjust these values for your game server.
                # 1GB of RAM
                memory: "1Gi"
                # 1.5 vCPU
                cpu: "1500m"
              limits:
                # These should be the same settings as requests.
                memory: "1Gi"
                cpu: "1500m"

Deploy the fleet by running the following command:

kubectl apply -f C:\Path\To\FleetFile.yaml

After the fleet deploys, you should be able to see the pods with kubectl get pods:

NAME                                    READY   STATUS      RESTARTS   AGE
eos-dedicated-server-47tpv-cgf6q        2/2     Running     0          2m
eos-dedicated-server-47tpv-vpwpq        2/2     Running     0          2m

You should also be able to see them with kubectl get gameservers:

NAME                               STATE   ADDRESS     PORT   NODE             AGE
eos-dedicated-server-47tpv-cgf6q   Ready   10.0.0.1    7181   west-europe-01   2m
eos-dedicated-server-47tpv-vpwpq   Ready   10.0.0.2    7072   west-europe-02   2m

Defining the game server fleet autoscaler

With the game server fleet deployed, it's now time to deploy an autoscaler. The Agones autoscaler will automatically create new game server containers as existing ones become allocated (by the game server calling the /allocate endpoint).

These are fairly simple to define and deploy, using the following YAML as a template:

apiVersion: "autoscaling.agones.dev/v1"
kind: FleetAutoscaler
metadata:
  name: eos-dedicated-server-autoscaler
spec:
  fleetName: eos-dedicated-server
  policy:
    type: Buffer
    buffer:
      # The number of game servers to keep available at all times. The autoscaler will
      # create or remove game servers as appropriate to keep this buffer.
      bufferSize: 2
      # The minimum number of game servers to run at any given time.
      minReplicas: 2
      # The maximum number of game servers to run at any given time. You can usually set
      # this to a significantly large value, as capacity will be limited by the number
      # of Kubernetes machines you have. However, if you're bursting into cloud, you
      # might want to limit this to prevent cost overruns.
      maxReplicas: 2000

Deploy the fleet autoscaler by running the following command:

kubectl apply -f C:\Path\To\FleetAutoscalerFile.yaml

You can test that the autoscaler works by manually creating a GameServerAllocation. Typically allocations will happen when the game server calls /allocate, but to save us some time in spinning up a client and connecting to a server, we're just going to do this manually. Create a file with the following contents:

apiVersion: "allocation.agones.dev/v1"
kind: GameServerAllocation

Then create the GameServerAllocation with:

kubectl create -f C:\Path\To\GameServerAllocationFile.yaml

If you use kubectl get gameservers, you should see a game server container allocated:

NAME                               STATE       ADDRESS     PORT   NODE             AGE
eos-dedicated-server-47tpv-cgf6q   Ready       10.0.0.1    7181   west-europe-01   12m
eos-dedicated-server-47tpv-vpwpq   Allocated   10.0.0.2    7072   west-europe-02   12m

After a few moments, you should see the autoscaler spin up a new game server container to keep the buffer of ready game server containers:

NAME                               STATE       ADDRESS     PORT   NODE             AGE
eos-dedicated-server-47tpv-cgf6q   Ready       10.0.0.1    7181   west-europe-01   12m
eos-dedicated-server-47tpv-vpwpq   Allocated   10.0.0.2    7072   west-europe-02   12m
eos-dedicated-server-47tpv-f74jg   Scheduled   10.0.0.1    7083   west-europe-01   2m

After a few moments, the new game server container should become Ready:

NAME                               STATE       ADDRESS     PORT   NODE             AGE
eos-dedicated-server-47tpv-cgf6q   Ready       10.0.0.1    7181   west-europe-01   12m
eos-dedicated-server-47tpv-vpwpq   Allocated   10.0.0.2    7072   west-europe-02   12m
eos-dedicated-server-47tpv-f74jg   Ready       10.0.0.1    7083   west-europe-01   2m

If you see this behaviour, congratulations! You have successfully deployed your game server containers onto your cluster with autoscaling enabled.

Before moving onto next steps, clean up the game server we manually allocated:

kubectl delete gameserver eos-dedicated-server-47tpv-vpwpq