Frequently Asked Questions
Architecture
What is the relationship between a Kubernetes Pod and an Agones GameServer?
Agones creates a backing Pod with the appropriate configuration parameters for each GameServer that is configured in the cluster. They both have the same name if you are ever looking to match one to the other.
Can I reuse a GameServer for multiple game sessions?
Yes.
Agones is inherently un-opinionated about the lifecycle of your game. When you call SDK.Allocate() you are protecting that GameServer instance from being scaled down for the duration of the Allocation. Typically, you would run one game session within a single allocation. However, you could allocate, and run N sessions on a single GameServer, and then de-allocate/shutdown at a later time.
How can I return an Allocated
GameServer to the Ready
state?
If you wish to return an Allocated
GameServer to the Ready
state, you can use the
SDK.Ready() command whenever it
makes sense for your GameServer to return to the pool of potentially Allocatable and/or scaled down GameServers.
Have a look at the integration pattern “Reusing Allocated GameServers for more than one game session” for more details.
Integration
What steps do I need to take to integrate my GameServer?
- Integrate your game server binary with the Agones SDK, calling the appropriate lifecycle event hooks.
- Containerize your game server binary with Docker
- Publish your Docker image in a container registry/repository.
- Create a gameserver.yaml file for your container image.
- Test your gameserver.yaml file.
- Consider utilizing Fleets. and Autoscalers for deploying at scale.
What are some common patterns for integrating the SDK with a Game Server Binary?
- In-Engine
- Integrate the SDK directly with the dedicated game server, such that it is part of the same codebase.
- Sidecar
- Use a Kubernetes sidecar pattern to run the SDK in a separate process that runs alongside your game server binary, and can share the disk and network namespace. This game server binary could expose its own API, or write to a shared file, that the sidecar process integrates with, and can then communicate back to Agones through the SDK.
- Wrapper
- Write a process that wraps the game server binary, and intercepts aspects such as the foreground log output, and use that information to react and communicate with Agones appropriately. This can be particularly useful for legacy game servers or game server binaries wherein you do not have access to the original source. You can see this in both the Xonotic and SuperTuxKart examples.
What if my engine / language of choice does not have a supported SDK, what can I do?
Either utilise the REST API, which can be generated from the Swagger specification, or generate your own gRPC client from the proto file.
Game Server SDKs are a thin wrapper around either REST or gRPC clients, depending on language or platform, and can be used as examples.
How can I pass data to my Game Server binary on Allocation?
A GameServerAllocation
has a spec.metadata section,
that will apply any configured Labels
and/or Annotations to a requested
GameServer at Allocation time.
The game server binary can watch for the state change to Allocated
, as well as changes to the GameServer metadata,
through SDK.WatchGameServer().
Combining these two features allows you to pass information such as map data, gameplay metadata and more to a game server binary at Allocation time, through Agones functionality.
Do note, that if you wish to have either the labels or annotations on the GameServer
that are set via a
GameServerAllocation
to be editable by the game server binary with the Agones SDK, the label key will need to
be prefixed with agones.dev/sdk-
.
See SDK.SetLabel()
and SDK.SetAnnotation() for more information.
How can I expose information from my game server binary to an external service?
The Agones game server SDK allows you to set custom Labels and Annotations through the SDK.SetLabel() and SDK.SetAnnotation() functionality respectively.
This information is then queryable via the Kubernetes API, and can be used for game specific, custom integrations.
If my game server requires more states than what Agones provides (e.g. Ready, Allocated, Shutdown, etc), can I add my own?
If you want to track custom game server states, then you can utilise the game server client SDK SDK.SetLabel() and SDK.SetAnnotation() functionality to expose these custom states to outside systems via your own labels and annotations.
This information is then queryable via the Kubernetes API, and can be used for game specific state integrations with systems like matchmakers and more.
Custom labels could also potentially be utilised with GameServerAllocation required and/or preferred label
selectors, to further refine Ready
GameServer
selection on Allocation.
Scaling
How large can an Agones cluster be? / How many GameServers can be supported in a single cluster?
The answer to this question is “it depends” 😁.
As a rule of thumb, we recommend clusters no larger than 500 nodes, based on production workloads.
That being said, this is highly dependent on Kubernetes hosting platform, control plane resources, node resources, requirements of your game server, game server session length, node spin up time, etc, and therefore you should run your own load tests against your hosting provider to determine the optimal cluster size for your game.
We recommend running multiple clusters for your production GameServer workloads, to spread the load and provide extra redundancy across your entire game server fleet.
Network
How are IP addresses allocated to GameServers?
Each GameServer
inherits the IP Address of the Node on which it resides. If it can find an ExternalIP
address on
the Node (which it should if it’s a publicly addressable Node), that it utilised, otherwise it falls back to using the
InternalIP
address.
How do I use the DNS name of the Node?
If the Kubernetes nodes have an ExternalDNS
record, then it will be utilised as the GameServer
address
preferentially over the ExternalIP
node record.
How is traffic routed from the allocated Port to the GameServer container?
Traffic is routed to the GameServer Container utilising the hostPort
field on a
Pod’s Container specification.
This opens a port on the host Node and routes traffic to the container via iptables or ipvs, depending on host provider and/or network overlay.
In worst case scenarios this routing can add an extra 0.5ms latency to UDP packets, but that is extremely rare.
Why did you use hostPort and not hostNetwork for your networking?
The decision was made not to use hostNetwork
, as the benefits of having isolated network namespaces between
game server processes give us the ability to run
sidecar containers, and provides an extra layer of
security to each game server process.
Performance
How big an image can I use for my GameServer?
We routinely see users running container images that are multiple GB in size.
The only downside to larger images, is that they can take longer to first load on a Kubernetes node, but that can be managed by your Fleet and Fleet Autoscaling configuration to ensure this load time is taken into account on a new Node’s container initial load.
How quickly can Agones spin up new GameServer instances?
When running Agones on GKE, we have verified that an Agones cluster can start up to 10,000 GameServer instances per minute (not including node creation).
This number could vary depending on the underlying scaling capabilities of your cloud provider, Kubernetes cluster configuration, and your GameServer Ready startup time, and therefore we recommend you always run your own load tests for your specific game and game server containers.
Architecture
Can’t we use a Deployment or a StatefulSet for game server workloads?
Kubernetes Deployments were built for unordered, stateless workloads. That is, workloads that are essentially homogeneous between each instance, and therefore it doesn’t matter which order they are scaled up, or scaled down.
A set of web servers behind the same load balancer are a perfect example of this. The configuration and application code between instances is the same, and as long as there are enough replicas to handle the requests coming through a load balancer, if we scale from 10 to 5, it doesn’t matter which ones are removed and in which order.
Kubernetes StatefulSets were built for ordered, stateful workloads. That is, workloads in which each instance is essentially heterogeneous, and for reliability and predictability it’s extremely important that scale up happens in order (0 ,1, 2, 3) and scaling down happens in reverse (3, 2, 1, 0).
Databases are a great use case for a StatefulSet, since (depending on the database), instance 0 may be the primary, and instances 1, 2, 3+ may be replicas. Knowing that the order of scale up and down is completely reliable to both ensure that the correct disk image is in place, but also allow for appropriate synchronisation between a primaries and/or replicas can occur, and no downtime occurs.
Dedicated, authoritative game server workloads are sometimes stateful, and while not ordered/unordered, game servers are prioritised for both scale down and allocation for player usage.
Game servers are sometimes stateful, because their state only matters if players are playing on them. If no players are playing on a game server, then it doesn’t matter if it gets shut down, or replaced, since nobody will notice. But they are stateful (most often in-memory state, for the game simulation) when players are playing on them, and therefore can’t be shutdown while that is going on.
Game Server workloads are also prioritised, in that the order of demarcating game servers for player connection and also on game server scale down impact optimal usage of the hosting infrastructure.
For example, in Cloud based workloads, you will want to pack the game servers that have players on them as tightly
as possible across as few Nodes as possible, while on scale down, will want to prioritise removing game servers from
Nodes that are the most empty to create empty Nodes that then can be deleted - thereby using the least amount of
infrastructure as possible.
So while one might be able to use Deployments and/or StatefulSets to run game server workloads, it will be extremely hard (impossible? 🤔) to run as optimally as a tailored solution such as Agones.
Ecosystem
Is there an example of Agones and other project working together?
Yes! There are several! Check out both our official and third party examples!
Feedback
Was this page helpful?
Glad to hear it! Please tell us how we can improve.
Sorry to hear that. Please tell us how we can improve.
Last modified October 3, 2024: Upgrade to Golang Version 1.22.6 and Golangci lint version v1.61.0 (#3988) (cd3bc13)