Multiplayer Scripting – Arma Reforger

From Bohemia Interactive Community
Revision as of 16:13, 26 August 2022 by Lou Montana (talk | contribs) (Add codec information)
Jump to navigation Jump to search

Multiplayer Scripting in Enfusion is based on Replication - it determines which machine does what in a situation.

When a Prefab entity is created on the server (and has a replication node set to broadcast), it is replicated on other network machines for them to see a representation of this object. This created representation is called a proxy – more information below.


Network

The Arma Reforger network architecture is a classical one with one server to which clients connect. Clients do not communicate between each others, they communicate with the server which redistributes data.

In single player, the local machine is considered a player-hosted server. A properly designed MP code should run flawlessly in single player without the need for specifics!

Server

The Server is the core network machine: it is the central system that receives and redistributes network information. There is only one server per multiplayer game, and the multiplayer game is destroyed if the server loses connection (server role does not get transferred).

A server can be
  • dedicated, as in it is a game instance running by itself, without a player behind
  • player-hosted (also known as listen server), meaning that a player's game instance is also hosting a game.
Arma Reforger does not allow for Player-Hosted Server as of now.

Client

A Client is a machine that is connected to the server. A client is everything else that is not the server.

A client can be
  • a player client, the most common understanding of a client: a machine connecting to the server with a player behind it to play the game and control a character
Headless client / distributed server is not a thing for now.

Join in Progress

A client can join after a game has started (even leave and join again) depending on the scenario. Such client is known as JIP, standing for Join in Progress.

By extension, JIP also covers "join in progress" network codes and systems.


Replication

Replication (shortened as RPL) is the name of the system used to replicate an entity or effect on all the concerned machines. For example, when a grenade explodes the damage is calculated on the server, whereas the visual effect is broadcast (replicated) to all the clients.

Remote Procedure Call (shortened as RPC) is the system allowing to queue network messages on other machines.

An entity that is to be replicated should always have an RplComponent script component.

Preamble

The network structure is based on server-client relationship; Scripting and RPC Replication methods, on the other hand, should be oriented towards Roles - code should be written on authority role principle, server-client principle should be avoided:

  • is the entity present on this machine the Authority?
  • if not, the local entity is a Proxy
    • if so, is it an Owner entity?

See the next chapters for further explanations.

Entity Role

When an entity gets created on a network machine, this entity is the Authority and this role is set in stone; it can never get changed or transferred.

If the network machine is the server, the entity's existence is broadcast to other network machines; the authority (server's entity) is represented on each client by locally-created entities named Proxies (local representation of the authority's reference).

  • For an entity created on the server:
    • the server hosts the authority entity
    • the entity is replicated as proxy on other machines
    • ownership can be given and taken by the authority
  • For an entity created on a client:
    • the client hosts the authority entity
    • the server being unaware of it, the entity is not replicated on any network machine
An entity that has been created on a client is not known by the server and its methods cannot therefore be broadcast to everyone.

Authority

The Authority is the reference entity.

All mission objects belong to the server; the server has almost if not all the authority entities.

A client can host an authority on its machine by having it created locally; by doing so the entity will be local to this client, no other machine including the server will be aware of it and therefore no replication can ever happen on it.

Proxy

A Proxy is an entity that represents an authority on another machine; it only receives entity updates from the authority and cannot send network updates about it - unless the machine has been designated owner (see below).

Owner

An Owner is a special entity role that comes in addition to an authority or a proxy. An owner entity is the entity that has minor rights on the entity such as owner-specific methods. Only one machine can own an entity. The server can be an owner, and ownership can be transferred by the authority.

A client owner is only a machine with an "elevated" proxy access, being able to use RPC methods about this entity towards the authority (and only towards the authority).

Ownership behaviour example:

  • a machinegun car exists on the server; the server is hosting the authority and is the owner of it
  • a player enters the car as the driver; the host (the server) gives car's ownership to the player's machine, which becomes the owner of the car
  • another player enters the car's machinegun; the host (still and always the server) gives machinegun's ownership to this player's machine, which becomes the owner of the car's machinegun entity
  • the server keeps authority of this car and its machinegun - it is the one that overrides all clients' positions/states
  • when said players leave their respective entities, the server takes ownership back
Giving ownership is not giving authority and is unrelated to it!
Ownership can be changed, authority location is set in stone.

RPC Call

An RPC call is an entity method's call through the network, e.g from the authority to its proxies (on other machines). Specific targets can be defined, such as "all proxies" or "owner only".

Join In Progress

Join In Progress is the fact of having a client that connects to the server while a multiplayer game is already running.

Streaming

An entity is not always present on all machines. Streaming is the way of saving network and CPU process by only sending relevant entities' information.

The streaming system is also what is at work during Join In Progress.

Relevance

Relevance is a value decided authority-side by the engine that sets whether or not a proxy should be streamed to a network machine or not, in order to save network messages and CPU cycles.

A simple example would be a car that is five kilometres away from a player is not relevant to that player; so the proxy on this player's machine is deleted by the replication system automatically, and re-created/synchronised again when it becomes relevant (e.g close-by) again.

Special Cases

The authority always exists, as it is the reference entity for replication.

The owner entity is always present on its machine; it is never unloaded/reloaded depending on relevance unless ownership is taken away.

Broadcast

An RPC call with broadcast as a target will send the message on machines that have said entity present on their system; as seen above, a proxy owner entity is guaranteed to exist on

This is a very important aspect of Replication: if the proxy is not streamed on a machine, the broadcast message will not reach this machine, as there is no entity target!

Operations Order

As of v0.9.5: loading order between RplLoad and member variable updates is not guaranteed and may be subject to change.

Current loading order:

Order Operation
1 entity hierarchy creation
2 RplLoad operations
3 member variable values in unguaranteed order

triggering their respective onRplName methods


Code

Entity methods get executed with the Rpc method using their RplRpc attribute to determine the targets.

Rpc Method

The Rpc method (present in both GenericEntity and GenericComponent) is the starting point to trigger a network-friendly code call.

Rpc(Rpc_MyMethod, parameter1, parameter2);

The engine will make it so that this method will call the provided one where it needs to be, depending on the RplRpc target method attribute.

By convention, methods that can be called through replication are prefixed with the Rpc_ prefix: e.g Rpc_MyMethod().

the RpcAsk_ prefix is used by the owner to ask the authority to do an action (sender = owner, receiver = authority).

the RpcDo_ prefix is used by the authority to do an action (sender = authority, receiver = owner proxy or all proxies).

Usage

void Rpc(func method, void p0 = NULL, void p1 = NULL, void p2 = NULL, void p3 = NULL, void p4 = NULL, void p5 = NULL, void p6 = NULL, void p7 = NULL);

Network-wise, a method RPC is added to a list of "to do tasks"; it means that they might be sent separately, or all in one network packet. The order of the calls between two machines is however guaranteed.

Calling an RPC method from EOnInit is not supported for initialisation order reasons.

RplRpc Attribute

The RplRpc attribute is to decorate an entity's method, allowing the engine to run said method on targeted machines through the Rpc method.

[RplRpc(RplChannel.Reliable, RplRcver.Owner)]

Usage

Here is RplRpc attribute's signature:

void RplRpc(RplChannel channel, RplRcver rcver, RplCondition condition = RplCondition.None, string customConditionName = "")

Parameter Usage
channel RplChannel enum, can be one of:
  • Reliable: the packet is guaranteed to reach its destination
  • Unreliable: the packet is sent but may be overridden depending on the order of arrival
Using Reliable is more expensive and should be used carefully. Unreliable should be used for non-important or very frequent calls (e.g updating positions).
rcver (receiver) RplRcver enum, can be one of:
  • Server: run the method on the authority entity
  • Owner: run the method on the owner entity
  • Broadcast: run the method on proxies only (not on the authority)
    • only machines that have the proxy streamed on their machine will receive the broadcast, not all the machines.
    • an immediate broadcast may not reach the owner entity right after transferring ownership; however an owner RPC call is guaranteed to reach it.
Permissions Table

Checked = RPL call is processed over the network
Unchecked = RPL call is abandoned

Action Authority Proxy
Server Client (local entity) Owner Client
RplRcver.Server

Send to the authority

Checked

(code is run locally)

Checked

(code is run locally)

Checked
Unchecked
RplRcver.Owner Send to the owner Checked

(code is run locally if authority is owner)

Checked

(code is run locally)

Checked

(code is run locally)

Unchecked
RplRcver.Broadcast Send to all but authority Checked

(code is run on all proxies and not locally)

Unchecked

(other machines have no proxy)

Unchecked

(code is run locally)

Unchecked

(code is run locally)

condition RplCondition enum, can be one of:
  • None: run the method in all cases
  • OwnerOnly: run the method on the owner machine only
  • NoOwner: run the method on all the other machines that are not the owner
  • Custom: uses the customConditionName field to use the named function as custom execution condition
customConditionName if condition has been set to Custom, customConditionName must be set to the name of the method to be checked. This method must return a bool.

Example:

class RplClass { [RplProp(onRplName: "OnBroadcastValueUpdated", condition: RplCondition.Custom, customConditionName: "RpcConditionMethod")] protected bool m_iBroadcastValue; void OnBroadcastValueUpdated() { // this method will only run on proxies if authority's RpcConditionMethod returns true } [RplRpc(RplChannel.Reliable, RplRcver.Server)] void RpcAsk_Method() { m_iBroadcastValue = 33; Replication.BumpMe(); } bool RpcConditionMethod() { return m_iBroadcastValue < 50; } }

RplProp Attribute

The RplProp attribute is to decorate an object's property, allowing the engine to synchronise values on network machines.

Only a change on the authority and notifying the replication system with Replication.BumpMe() method will trigger a broadcast to other machines (see the BumpMe section).

RplProp-decorated properties are automatically synchronised by the Streaming system.

Usage

Here is RplProp attribute's signature:

void RplProp(RplGroup group = RplGroup.Mandatory, string onRplName = "", ScriptCtx ctx = NULL, RplCondition condition = RplCondition.None, string customConditionName = "")

Parameter Usage
group RplGroup enum, can be one of:
  • Mandatory
  • Group_1
  • Group_2
  • Group_3
  • Group_4
  • Group_5
  • Group_6
These groups are meant to update parts of said entity in the event of a big data structure - more details later
onRplName The method name to be called on proxies when the property's value is changed on the authority (and synchronised with Replication.BumpMe() method).
  • This method will not be called on the authority
  • This method will be called on proxies' streaming (synchronisation)
  • only machines that have the proxy streamed on their machine will receive the broadcast, not all the machines.
ctx Script context - engine only, unused script-side
condition RplCondition, can be one of:
  • None: update the property in all cases
  • OwnerOnly: update the property on the owner machine only
  • NoOwner: update the property on all the machines that are not the owner
  • Custom: uses the customConditionName field as custom execution condition
customConditionName if condition has been set to Custom, customConditionName must be set to the name of the method to be checked. This method must return a bool.

Example:

class RplClass { [RplProp(onRplName: "OnBroadcastValueUpdated", condition: RplCondition.Custom, customConditionName: "RpcConditionMethod")] protected bool m_iBroadcastValue; void OnBroadcastValueUpdated() { // this method will only run on proxies if authority's RpcConditionMethod returns true } [RplRpc(RplChannel.Reliable, RplRcver.Server)] void RpcAsk_Method() { m_iBroadcastValue = 33; Replication.BumpMe(); } bool RpcConditionMethod() { return m_iBroadcastValue < 50; } }

RplSave/RplLoad

RplSave

RplSave is the method called on the authority's side to write custom synchronisation data.

RplSave writes data to a provided ScriptBitWriter instance. Different data types have different sizes, see Values - Primitive Types for their default size.

Helper methods such as WriteBool or WriteVector exist to save some trouble as syntactical sugar.

Note that using the default size is not always the best/economical option; see below example.

RplLoad

RplLoad is the method called on the streamed entity after the entity, its components and the whole entities hierarchy to which it belongs has been loaded, if any.

It is called immediately after initialisation but before anything else (EOnFrame, etc).

RplLoad reads data from a provided ScriptBitReader instance. Reading from it must be done in the same order and with the same number of bits than the writing that happened in RplSave, otherwise data discrepancy may happen and behaviour is not guaranteed.

Replication Class

The Replication class is a very useful Replication tool, holding the following methods (only the most used ones are listed here).

FindId

Return the provided entity's id. should be an RplId

If the entity does not have one, it returns Replication.INVALID_ID.

This method uses a table lookup and therefore should not be used in a tight loop - once found, the id should instead be stored in a variable.

FindItem

Return the provided id's entity.

If no entities have the provided id, it returns null.

FindOwner

Return the owner machine's network id.

If the entity has no owner id, it returns Replication.INVALID_IDENTITY.

BumpMe

This method tells the Replication system that at least one property has been updated and that the change(s) should be broadcast.

This method should be used whenever an RplProp property has been changed and is not to be randomly called hoping a property has been changed.
class RplClass { [RplProp()] protected int m_iValue; [RplRpc(RplChannel.Reliable, RplRcver.Server)] protected void RpcAsk_Authority_Method(int newValue) { if (m_iValue != newValue) // not necessary for small changes { m_iValue = newValue; Replication.BumpMe(); } } }


Network Modding

Definitions

Snapshot
An "image" of an object's replicated properties - used by the engine to compare with the current status' snapshot and detect a change
Packet
(naming may change) An object's transferrable data - unrelated to network packet

Methods

EnNetwork.c provides an example class with declarable methods overriding the default behaviour. These methods can be written/overridden in any Enforce Script classes.

See e.g SCR_PlayerRespawnInfo.c for a real implementation.

class CustomClass { //! From snapshot to packet. static void Encode(SSnapSerializerBase snapshot, ScriptCtx ctx, ScriptBitSerializer packet); //! From packet to snapshot. static bool Decode(ScriptBitSerializer packet, ScriptCtx ctx, SSnapSerializerBase snapshot); //! Snapshot to snapshot comparison. static bool SnapCompare(SSnapSerializerBase lhs, SSnapSerializerBase rhs, ScriptCtx ctx); //! Property mem to snapshot comparison. static bool PropCompare(RplTestPropType prop, SSnapSerializerBase snapshot, ScriptCtx ctx); //! Property mem to snapshot extraction. static bool Extract(RplTestPropType prop, ScriptCtx ctx, SSnapSerializerBase snapshot); //! Snapshot to property memory injection. static bool Inject(SSnapSerializerBase snapshot, ScriptCtx ctx, RplTestPropType prop); }

Codec Steps

On Server On Client
  • Replication.BumpMe() is used to signal that properties of an item have changed and they need to be replicated to clients
  • Replication compares replicated properties against the most recent snapshot it has using the PropCompare() codec function. If the codec says snapshot is the same as the current state, the process ends
  • Replication creates a new snapshot and uses the Extract() codec function to copy values from instance to snapshot
  • That created snapshot is transmitted to clients as needed.
This process tracks multiple snapshots per item per connected client, join-in-progress, streaming, relevancy, etc.

It often uses the SnapCompare() codec function to determine whether two snapshots are the same.

When a snapshot is finally being prepared for transmission over network, the Encode() codec function will be used to convert snapshot into compressed form (using as few bits as possible for each value) suitable for network packet.
  • When a new packet with the compressed snapshot arrives, it is decompressed using codec function Decode()
  • Snapshots are compared using the codec function SnapCompare() to determine whether or not changes have occurred
  • Replication updates properties of the replicated item using the codec function Inject().


Examples

For the sake of the examples, let's assume that:

  • the server hosts the ComputerEntity authority entity
  • a client hosts the owner entity

RplRpc

In the following code:

  • the protected RpcAsk_Authority_Method method is set to be RPC-called on the authority
  • the protected RpcDo_Owner_Method method is set to be RPC-called on the owner
  • the protected RpcDo_Broadcast_Method method is set to be RPC-called on every proxy

We assume the public TurnOn method has been triggered owner-side, making the following events happen:

  • the TurnOn method calls the RpcAsk_Authority_Method through RPC; as stated earlier, this method executes on the authority's machine - the server in this case
  • the RpcAsk_Authority_Method method (running on the authority):
    • prints "authority-side code" in the authority machine's logs
    • calls the PlayMusic method on the authority
    • calls the RpcDo_Broadcast_Method method on all the proxies (see below)
    • calls the RpcDo_Owner_Method method on the owner's proxy (see below)
  • the RpcDo_Broadcast_Method method (running on every proxy, including owner's):
    • prints "proxy-side code" in proxy machines' logs
    • calls the PlayMusic method on the proxy
  • the RpcDo_Owner_Method method (running on the owner's proxy):
    • prints "owner-side code"

The end result:

  • all the machines' instances:
    • called the PlayMusic method and played the "OSWelcome" music
  • the authority's machine has
    • "authority-side code" in the logs - had the server been the owner too, "owner-side code" would be present as well
  • the owner's machine has
    • "proxy-side code" then "owner-side code" in the logs (the order is guaranteed)
  • all other machines have
    • "proxy-side code" in the logs

class ComputerEntity : IEntity { protected bool m_bIsTurnedOn; // this value is edited only on authority's side [RplRpc(RplChannel.Reliable, RplRcver.Server)] protected void RpcAsk_Authority_Method(bool turningOn) { Print("authority-side code"); if (turningOn == m_bIsTurnedOn) // the authority has authority { return; // prevent useless network messages } m_bIsTurnedOn = turningOn; PlayMusic(turnOn); // play music on authority Rpc(RpcDo_Broadcast_Method, turningOn); // send the music broadcast request Rpc(RpcDo_Owner_Method); // run specific code on the owner's entity (that may or may not be the authority) } [RplRpc(RplChannel.Reliable, RplRcver.Owner)] protected void RpcDo_Owner_Method() { Print("owner-side code"); } [RplRpc(RplChannel.Reliable, RplRcver.Broadcast)] protected void RpcDo_Broadcast_Method(bool turningOn) { Print("proxy-side code"); PlayMusic(turningOn); } protected void PlayMusic(bool turningOn) { if (turningOn) { SomeSoundClass.PlayMusic(this, "OSWelcome"); } else { SomeSoundClass.PlayMusic(this, "OSGoodbye"); } } // public methods void TurnOn() { Rpc(RpcAsk_Authority_Method, true); } void TurnOff() { Rpc(RpcAsk_Authority_Method, false); } }

RplProp

In the following code:

  • the m_bIsTurnedOn property is set to update to all the proxies on authority value's modification
  • the m_bIsTurnedOn property is set to execute the OnTurnedOnUpdated method on proxies when its value is updated by Replication (JIP included)

We assume the public TurnOn method has been triggered owner-side, making the following events happen:

  • the TurnOn method calls the RpcAsk_Authority_Method through RPC; as stated earlier, this method executes on the authority's machine - the server in this case
  • the RpcAsk_Authority_Method method (running on the server):
    • sets LED colour on the server
    • sets the m_bIsTurnedOn property to "true" on the authority; as the m_bIsTurnedOn property is set to broadcast its changes thanks to the RplProp attribute, the change is broadcast to every proxy machine
  • the OnTurnedOn method is triggered on proxies and "The proxy has been updated" is printed

The end result:

  • all the machines' instances:
    • have the m_bIsTurnedOn property set to "true"
    • have their light set to green
  • the authority:
    • printed "authority-side code"
  • all the proxies:
    • printed "proxy-side code"

class ComputerEntity : IEntity { [RplProp(onRplName: "OnTurnedOnUpdated")] protected bool m_bIsTurnedOn = false; // this value is to be updated by the authority, not set locally by proxies (even owner) // if it is set locally, the change will not broadcast and there will be a difference between the proxy and the authority // this state discrepancy will last until authority's next update broadcast [RplRpc(RplChannel.Reliable, RplRcver.Server)] protected void RpcAsk_Authority_Method(bool turningOn) { if (turningOn == m_bIsTurnedOn) { return; // prevent useless network messages } Print("authority-side code"); m_bIsTurnedOn = turningOn; // m_bIsTurnedOn is changed only in an authority-targeting method // it will broadcast over the network automatically due to Replication setting (line 3) SetLedLightColour(); // SetLedLightColour is not automatically called on the authority Replication.BumpMe(); // tell the Replication system this entity has changes to be broadcast // the Replication system will update the member variable AFTER RpcAsk_Authority_Method is done } protected void OnTurnedOnUpdated() { Print("proxy-side code"); // this method is called on proxies when m_bIsTurnedOn is updated through Replication // it is NOT called on the authority SetLedLightColour(); } protected void SetLedLightColour() { if (m_bIsTurnedOn) { SomeLightClass.SetLedLightColour(this, Color.Green); } else { SomeLightClass.SetLedLightColour(this, Color.Red); } } // public methods void TurnOn() { if (m_bIsTurnedOn) // m_bIsTurnedOn can be read from any entity { return; } Rpc(RpcAsk_Authority_Method, true); } void TurnOff() { if (!m_bIsTurnedOn) { return; } Rpc(RpcAsk_Authority_Method, false); } }

RplSave/RplLoad

Without RplSave/RplLoad override, the following entity would be created with m_iSoldierId, m_iH ealth and m_bHadLunch set to their default value, as these are not decorated with RplProp.

class SCR_RplTestSoldier : IEntity { [RplProp()] protected string m_sName = "Player 1"; // joining in progress automatically synchronises RplProp variables protected int m_iSoldierId = 12345678; protected int m_iHealth = 100; // range 0..100 protected bool m_bHadLunch = false; protected string m_sSoldierDogTag = "PID 1234"; // let's assume length can change // Called on the authority when an entity gets streamed override bool RplSave(ScriptBitWriter writer) { // m_sName is automatically synchronised, no need to do it manually writer.Write(m_iSoldierId, 32); // write 32 bits of soldier ID - int is 32 bits in size writer.Write(m_iHealth, 7); // write 7 bits of health - 7 bits are enough if the value cannot go over 100 (7 bits range is 0..127) writer.WriteBool(m_bHadLunch); // write 1 bit only writer.WriteInt(m_sSoldierDogTag.Length()); // write following string's length as it can change writer.Write(m_sSoldierDogTag, 8 * m_sSoldierDogTag.Length()); // one character is 1 byte (8 bits) return true; } // Called on the streamed proxy override bool RplLoad(ScriptBitReader reader) { // m_sName is automatically synchronised, no need to do it manually if (!reader.Read(m_iSoldierId, 32)) // read 32 bits of data - the authority wrote soldier ID first, so it needs to be read first { return false; } if (!reader.Read(m_iHealth, 7)) // read 7 bits of data - the authority wrote health second, so it needs to be read second { return false; } if (!reader.ReadBool(m_bHadLunch)) // read 1 bit only { return false; } int dogTagLength; if (!reader.ReadInt(dogTagLength)) // read string's length { return false; } m_sSoldierDogTag = ""; // string should be emptied before being filled otherwise it could be only partially updated, e.g // old string = "PID 1234" // new string = "a b c" // result = "a b c234" if (!reader.Read(m_sSoldierDogTag, dogTagLength)) // read the whole string { return false; } return true; } }