Vixen
dd8b0a81

Live · tutorial

Writing a realm

A dedicated server that is a shard — the one-line entry point, the four hooks it seals, and the map it opens.

Edit this page on GitHub

Documents

What it is

A realm is a normal Vixen application built as a dedicated server, plus the pieces that make it a shard. Realm is the Game you derive from, RealmApp.Run is the entry point, and RealmHost is the state machine underneath: Starting → Ready → Draining → Stopping → Stopped.

What it is for

Everything a game does on a listen server it does here, unchanged — replication, RPC, interest, prediction's server half. What the realm adds is a lifecycle somebody else can operate: it is told what shard it is, it says when it is ready, it admits only ticketed players, it reports what it is costing, and it can be asked to drain.

Using it

csharp
using Vixen.Live;using Vixen.Live.Realms;public sealed class QueensdaleRealm : Realm {    protected override void OnRealmInitialise() {        // Host.Session is doc 16's server. Replication, RPC and interest are wired here, exactly as        // they would be in a listen server — a realm is not a different kind of server.    }    protected override TransferReadiness ReadinessOf(RealmPlayer player) =>        // The hook that stops a rollout from ending a raid.        TransferReadiness.Ready;}public static class Program {    public static int Main(string[] arguments) => RealmApp.Run<QueensdaleRealm>(arguments);}

The process is launched with one argument — --realm-spec shard=…;map=…;port=… — which a placement backend writes. A process handed no spec says so on standard error and exits 2, which is distinguishable from a crash and not worth a launcher retrying.

What Realm decides for you

Realm seals four of Game's hooks and offers realm-shaped ones in their place. OnConfigure is where the shard's boot decisions are made, and they are not a realm's to undo:

VariantBuildVariant.Server — headless host, Vixen.Graphics.Null, and the content build's server profile when the realm was published with -p:VixenVariant=Server (see content in a game § A dedicated server's content)
Window, Graphics.Enablednone, off
StartupSceneRealmSpec.Key.Map
FrameRateLimitthe spec's tick rate

OnRealmConfigure runs after all of it, and everything else is fair game.

⚠ The map is AppConfig.StartupScene, not a loader of its own. The host already opens it before OnInitialise, reports its own failures and survives them. A realm that loaded the map a second way would be a second code path for content failures, tested half as often.

Examples

Is the map up?

That is the one question the host does not answer, and it is what separates Starting from Ready. MapLifetime answers it by looking for a loaded scene whose name matches the map address's leaf — which is not a workaround, it is the same identity NetworkSceneId hashes.

csharp
var map = new MapLifetime(spec.Key);map.Resolve(scenes);     // false until the host's startup scene appearsmap.Quiesce();           // draining: still simulating, no arrivalsmap.Unload(scenes);      // and down again

⚠ A realm whose map never appears never becomes ready, and that is the correct failure. It will be started, it will not be placed on, and the orchestrator will eventually stop it — a shard that quietly did nothing rather than one that admitted players into an empty world.

A realm whose map is not a startup scene — a generated map, a persistent shard rehydrating authored state — calls MapLifetime.Ready instead, which exists so that such a realm does not have to fake a scene name.

Draining

csharp
host.Drain();   // stops arrivals, quiesces the map, and moves nobody by force

The players already on the shard leave at moments ReadinessOf approved of, and the shard stops once it has been empty for RealmHostOptions.IdleGrace. The grace is not zero: a player who was moved out may have a reconnect in flight, and a shard that vanished the instant its population hit zero would turn a lost packet into a lost session.

Overriding the transport

Realm.CreateTransport is virtual and binds UDP to every interface on the port placement published. The spec's host is what a client is told — a node's external address, a relay's name — and is frequently not an address the process could bind even if it wanted to. Swapping in a relay-allocated endpoint or a composite that accepts both direct and relayed clients is a placement decision rather than an architecture change, precisely because nothing above ITransport knows the difference.

Why not VixenApp.RunRealm

Doc 27 writes the entry point as VixenApp.RunRealm<MyRealm>, and it cannot be: VixenApp lives in Tools/Vixen.App, which sits below Live/, and a static class cannot be extended from outside. So the entry point moved rather than the layering, and RealmApp mirrors the original call for call.

See also