multiplayer

[cursive cloud]

that keeps it synchronized

[modest crater]

You would modify the game modes spawning players or even simpler, just spawn as spectators with no input and then wait for an event (Press A to start) which then would make the players on the server possess a different actual gameplay character, done - just saved you some money

[dark parcel]

it will not on machine that doesn't get to execute the RPC.

e.g late joiners or if the actor is outside relevancy.

[thin stratus]

Yop, RPC are not meant to synchronize stuff

[grand axle]

Currently already doing something like this. players are meant to be spawned with proxy input receivers and when the "Jump" button is pressed, spawn the player at X Co-ord based on spawn spots

[modest crater]

So what's the issue then?

[cosmic trench]

the issue still exists in my game...resume game not really doing what it's expected to do and the background goes all black even though i have only set a background blur for the game menu widget. i'm also facing trouble when trying to host a session after going back to the main menu, i can only seem to host the game from the host of the previous game and others cant host, and also cant join the session that the host created after. i've also shared the output log in this video as well

[nocturne quail]

The issue is with animation montages, cause if i add the notify directly to the animation it works fine and executes once, adding the notify to the montage executes twice but the montage plays once.

Animation calls notify once, ✅
Montage calls notify twice . Engine Bug

in the original 4.26 project everything is working fine.
Recreating the montages fixed the issue in 5.4,
seems like upgrading from 4->5 , screwed the animation montages which calls the notify twice

[cosmic trench]

i've set the input action event for the game menu in my player controller as well

[nocturne quail]

spawned by server?

[sonic obsidian]

Does FFastArraySerializer have a limit on the size/amount of elements it can replicate?

[thin stratus]

Could be that they changed something and the update path didn't properly clean up the old way or so

It's not so easy to keep software updated across a lot of updates. Especially if it's projects created by such

[thin stratus]

I would say no. But you might get into a problem with large arrays on initial replication

Actors that are placed into the scene won't despawn.

You can't compare that to runtime spawned ones.

You can try setting the NetLoadOnClient or whatever it's called setting on the placed ones to the opposite of whatever it is currently

[cosmic trench]

Ig I'm gonna have to figure out on my own

[sonic obsidian]

Yea that was exactly what i was thinking about, the delta rep is nice but its useless for a join in progress case for example. Does FFastArraySerializer handle that the same as a TArray or can it for example split the data to send over multiple frames i don't mind when the data gets to the client as long as all of it does.

[thin stratus]

That idk

[sonic obsidian]

alg thanks though

[thin stratus]

I know you will hit package size limits which will potentially make the whole replication more CPU heavy but that's only once initially I guess

[cosmic trench]

I just need help with this one final thing and my project is done

[nocturne quail]

💯

[thin stratus]

The Session stuff is usually because you didn't Destroy the Session of Host and Clients when they enter the MainMenu.

It's a good idea to simply put that into your MainMenu GameMode's BeginPlay, so you always destroy a Session, even if none is there.
You can also do a fail-safe and call DestroySession before trying to host/join.

If your Clients have trouble joining afterwards, then I assume you might not be cleaning up their Session after the Host left.

@cosmic trench Also your video seems to suggest you aren't using SeamlessTravel for your ServerTravel that goes from Lobby to Gameplay map.
You don't want to perform a HardTravel for that. Clients will disconnect and reconnect, which is not very smooth, and requires the whole "login" process to run twice.
On top of that, Subsystems like Steam will not work with that.

You need to mark the Lobby and Gameplay GameModes as SeamlessTravel in their Class Defaults.

But a few words of advice for after you've changed that:

• By default you can't test this in the Editor. The Server will simply not travel.
  • You can enable that with a console variable but it's experimental and might cause a crash.
  • If the CVar doesn't work for you, you will need to start the game as actual Standalone and/or package it. Standalone is faster (easiest way to do that is to rightclick the .uproject file and click on "Launch Game". I assume the Standalone option when pressing play (not the Play as Standalone, but the one that is in the list of "New Editor Window", "Play in Viewport" etc.).
• Since SeamlessTravel keeps the Clients connected (and parked on the Transition Map you can select in the Maps&Modes in your Project Settings), some of the code-paths you usually expect to call for connecting clients won't call.
  • E.g. if you do something on "PostLogin" in the GameMode, then you need to move that to the GenericInitializeNewPlayer or so override instead, that calls for both connecting and seamless traveling.

[thin stratus]

What's the InputAction set to in terms of Triggers? The background going darker and darker looks like you are adding the menu over and over again, that's why clicking on resume doesn't do anything as you only removed one of all those widgets.
Make sure the Triggered doesn't call more than once (use a "PRESSED" Trigger for that), and maybe also ensure that the Widget won't get created if the PauseMenuRef is already valid.

[cosmic trench]

It seems to be working properly now after adding the destroy session before hosting or joining a session and also added the destroy session on BeginPlay inside the main menu GameMode

[cosmic trench]

I have set the pause menu now to show up only when the PauseMenuRef is not valid and now the darkness in the background is gone and i'm able to click the buttons (both resume and leave buttons) but once i press resume, i cant call the pause menu again when i try to trigger the pause menu input action again

I haven't set any triggers for that input action

[thin stratus]

Are you setting the PauseMenuRef back to null?

[thin stratus]

Then it's probably the HOLD Trigger by default. Try adding the Pressed one, that should ensure it's only triggering on press and not on hold.

[cosmic trench]

That's what I'm trying to do but can't quite figure out how

[thin stratus]

You just set variable without connecting something to it.

[cosmic trench]

once i do that, the darkness comes back again in the background

[cosmic trench]

this actually fixed the issue, thanks!

[thin stratus]

Yeah I guess the InputAction is constantly triggering.

Probably cause the Key isn't counted as released when you switch InputMode

Might want to check if there is a "Flush Keys" or so function in the PlayerController in Blueprints.

[cosmic trench]

now the only problem is, once i leave the session and start another session, i cant seem to control any of the players again

[thin stratus]

Or Flush pressed keys or so

[thin stratus]

That might go hand in hand with the fact that your InputAction keeps triggering. Are you properly changing the InputMode in those various situations?

[cosmic trench]

i change the input mode to ui only when the pause menu is triggered and back to game only when i press resume

[thin stratus]

Is the InputMode initially UI in the MainMenu?

Are you ever setting it to Game when you enter the Lobby/Gameplay maps?

You are probably setting it to UI in the PauseMenu, then leave the Session and join a new one, but you never undo that and are stuck in UI InputMode.

fyi, the whole InputMode stuff is best to be handled a bit differently, or maybe even directly via the CommonUI plugin.

I usually have a base-class for my PlayerController and I figure out on Tick if Menus or Dialogs are open and then change the InputMode.
That's not super clean, but it ensures that the InputMode is handled in one single place. Requirement for that is that you handle your Menus and Dialogs that needs UI interactions in your PlayerController and keep references there. As well as a proper inheritance setup so that PauseMenu and MainMenu both count as "Menus".

CommonUI makes it a bit easier I guess, there you'd use an ActivatableWidget as the base for your MainWidgets. So e.g. for MainMenu, PauseMenu and HUD.
And that class gives you the option to return a preferred input mode, so whenever such a widget become "active" it automatically changes the mode back and forth, like a stack.

Makes things a bit simpler.

But for now you should just set the InputMode in your cases to UI and Game at the right times.

[cosmic trench]

I wasn't doing that and then tried doing it in the lobby gamemode and it only worked for the host and not the clients, the clients still remained at the ui only input mode

[thin stratus]

GameMode is also not the right choice here, cause that only exists on the Server.

MainMenu GM would be fine, cause there is no Server/Client setup there.

But for the multiplayer maps you will need to use the PlayerController -> BeginPlay -> IsLocalPlayerController -> SetInputMode.

[cosmic trench]

Okay it works properly now, thank you so much!

[thin stratus]

No worries (:

[thin stratus]

No but it makes the difference to the ones placed in the scene

Those ones don't despawn if they are loaded during map load

Which is probably what you notice with the runtime spawned one happening

[fallow mango]

Need someone big brain to give me advice, my multiplayer respawn system - when the 2nd player respawns, the 1st can no longer move.

[thin stratus]

Share code of your respawn system

Probably using GetPlayerController0 or stuff like that

[thorny ore]

Hmm. I just enable server streaming to true and enableserver streaming out to true... and in editor it's teleporting the second player to a different map.

Any suggestions?

[fallow mango]

in my player character -

calls this - also in my player char - (called on server reliable)

into my controller run on server

then to my game mode

I've tested and even though player 1 cant move - the movement mode is still set walking, the input returns a 1 when pressed - the player can, jump roll attack, just not walk or run / move

I'm close to giving up at this point 😂

[thin stratus]

The delayed posses is a bit strange

Also most of what you are doing seems so standard that you could just use the built in functions for it

Let the GameMode spawn the default pawn. Let the GameMode find/choose the PlayerState and simply call RestartPlayer after destroying the old pawn.

But that aside I don't see why you wouldn't be able to control the pawn anymore.

The second RPC isn't needed unless you want to call that by the client without death.

Actually both RPCs aren't needed

Your character shouldn't die on the client.

Damage and death should be done on the server already. Only the initial input that kicked off the chain of events that leads to the death should be a server RPC

And I would remove the delay

[thin stratus]

Actually this is your problem

In theory at least

You do the whole spawn stuff and then call unposses

Order of events is all mixed up.

1. Unpossess old Pawn (keep a local ref)
2. Destroy old Pawn (via local ref)
3. Spawn new Pawn
4. Posses new Pawn (without delay)

[fallow mango]

right, so I just switched around the unposses and spawn

and now the 2nd player looses movement when the 1st respawns

without dying at all

[thin stratus]

That's not right though

The unpossess should be in the respawn function

And it should be done after keeping a reference to the pawn so you can call destroy on it

Otherwise the returned controlled pawn is null

[thin stratus]

Do you have auto posses on in the character's class default?

This is in theory really straight forward

[fallow mango]

You cant call unposses from a game mode?

[thin stratus]

Why does that matter?

[fallow mango]

so how can I put it in the respawn function?

[thin stratus]

RespawnPawn is in the Controller?

The way you have it set up atm would mean calling RespawnPawn without Death won't unposses the pawn

Not the end of the world but still strange

[fallow mango]

No, I don't

have moved the unposses to the controller but still when p1 respanws p2 is still alive but cannot move

[thin stratus]

Hm, not sure atm. Something obvious most be wrong. Where is that first screenshot from? Why is Death a ServerRPC?

[fallow mango]

my clients were not taking any damage

so i made this server damage function

and at the end of it, if 0 health remains, it calls death

so they die...

[fallow mango]

Appreciate your help, will take another look at some docs and have another run at it.

[thin stratus]

I feel like your setup is a bit wrong. You should not need a ServerRPC for death. Quite the opposite.

That means your damage is already dealt on clients.

[verbal dust]

why does this work:

but this doesnt

[thin stratus]

You know what RPCs do?

Basically, RPCs call on another instance of the game. Usually on a different Computer, but can also be the same Computer.
The Jump_RPC calls on the Server's instance of the game. It gets the MontageToPlay passed.
Then it calls JumpReplicate, which calls on the Server and all Clients. Server will be fine with the MontageToPlay being passed into the PlayMontage node like this, cause the Montage still sits in its memory.
But the Clients have no idea about the MontageToPlay. It's a different instance/process of the game, maybe even a different Computer. You need to pass it along the RPC.

Sadly Blueprints allow you to connect stuff across the EventGraph like this. In C++ this would be a lot clearer.

void AYourBlueprintActor::Jump_RPC(UAnimMontage* MontageToPlay)
{
  JumpReplicate(nullptr)
}

void AYourBlueprintActor::JumpReplicate(UAnimMontage* NewParam)
{
  PlayMontage(Mesh, ???, ...);
}

This is what you have in the lower screenshot. As you can see, the bottom function has no MontageToPlay available to it. It can only use the member properties and the inputs.

And this is the first screenshot:

void AYourBlueprintActor::Jump_RPC(UAnimMontage* MontageToPlay)
{
  JumpReplicate(MontageToPlay)
}

void AYourBlueprintActor::JumpReplicate(UAnimMontage* NewParam)
{
  PlayMontage(Mesh, NewParam, ...);
}

Hopefully that makes somewhat sense.

[verbal dust]

Yeah I think I get it. MontageToPlay is passed along to JumpRPC via the event input. Then it calls the multicast on all the clients, which is happening on a separate instance, so even if MontageToPlay is connected it isnt actually passed along.

[thin stratus]

Yop

You should generally not connect inputs of something that isn't part of the same top most exec line

That includes delays for example. You can't be sure the inputs are still valid

[verbal dust]

Am I doing things right when I plug variables into the the event input then? Whats the best approach to figuring out what to do on server and what to do on client (focused on performance and not cheat prevention). Right now my logic is to do as much as possible on the client, and to only pass along the final variable and trigger the event on server. So for example, figure out which jump AnimMontage needs to be played on cleint side, and only give the server the MontageToPlay and trigger the montage. The logic being to send/replicate as few variables as possible to the server, and to do as little work on the server as possible so theres less lag between client and server.

does that make sense?

[thin stratus]

If you are on the same machine you need to either pass along as inputs, or if it's something latent (something that doesn't call the same frame) you probably need to save it to a variable.
And if you send something to another machine you need to use the inputs.

[thin stratus]

Depends on your Game. Even with 50-100 players, the server can do quite a lot. If you only aim for a handful of players then I would suggest just sticking to doing most stuff on the server

[verbal dust]

So I should pretty much just take the action inputs and send those to the server?

[thin stratus]

In theory that's what most peeps do. Minus some prediction. But even that results in the "inputs" being sent to the server

If you fire a gun, send an RPC and do the tracing on the server, the damage on the server, the kill on the server and the respawn on the server. You can send RPCs back for one time events that aren't important, like hit events, VFX, etc. and you can set replicated variables (optionally OnRep if you need something to react to the value changing) for any state stuff.

After that you can think about prediction, but prediction has its limits and is pretty difficult.

E.g. even if you predict the weapon shot, you can't be sure the server sees the same enemies and the same location.

Which means you need to add a history and rollback to the server so it can figure out what the client saw to give more accurate hit validation. And you need to handle the rollback of false predicted shots. And ultimately you still only deal damage on the server. Cause predicting damage ends really quickly when you notice you'd need to predict death and also allow rollback of a falsely predicted death etc

So doing most stuff on the server is better. If you don't care about cheating then you can simply some things for the sake of better feeling for the client.

Such as doing the trace locally and sending the result to the server.

But that's not due to performance but just so the client feels better about shooting others

While others potentially feel worse about being shot

[verbal dust]

I'm doing something coop only, so I'd prefer stuff like damage to be applied based on what the client sees and not the server

what I'm not sure about is movement. I figured doing most of it client side would feel better, but I'm not so sure now. Movement is replicated out of the box as far as I know and I dont think theres any way to "bias" it towards the client? I think player position is corrected according to the player pawn position on the server by the character movement component.

[exotic wasp]

pretty sure theres a ticky box to make it client authoritative

ACharacter does predictive movement so it feels like its client authoritative but the server verifies what is happening

if you're using a pawn you have to completely brew your own movement code anyway

[verbal dust]

this one?

[raven bone]

Hey people, its been a while,
im trying to decouple my UI functionality from the inventory component by using event dispatchers.
are there any limitations on using dispatchers in MP?

[verbal dust]

thats rally helpful, thanks!

[nova wasp]

it's just a way to call functions in an event based way... it doesn't have much to do with multiplayer

you are on the right tracking making them a bit more decoupled though if that helps

the UI should represent information, it's not what sends it or defines what it is in most cases

[thin stratus]

The only thing EventDispatchers do is to create a newsletter like subscription setup, where things that are interested in the event bind to it, as opposed to the thing notifying everything, which also doesn't scale well.

They aren't a multiplayer concept.

[raven bone]

thats what i assumed.
Yet, im running into a issue when trying to fire a UI function, which is running only on the server

[nova wasp]

well what fires it?

of course you still need to replicate information and run stuff on the client for a UI

[blazing spruce]

What is the correct way to introduce a cooldown or delay to stop the player being able to spam inputs? Do I need a server-side timer or is there a simpler way to do it?

[nova wasp]

I would say just have it on both sides but have the server one use some forgiveness (because rpcs from the client could be fairly random in timing)

for example you could send the desire to do the input (bwantstojump or something) and wait to apply it until the timer elapses

[raven bone]

Inventory - Add Item
old iteration im using interfaces to notify the UI to add items, update and so on
As of now its a function, but its called from a multicast event on the item

[nova wasp]

"called from a multicast"

I think you misunderstand what I meant

I am asking why the code runs on the server because something ON THE SERVER ran it

does the client not have some kind of representation of the item on it?

the common inventory setup is generally to use a fast array in c++ but in bp I suppose you are a bit more limited to just regular onreps for properties

[raven bone]

not me getting confused by my own code

[nova wasp]

this stuff is difficult to get into but you kind of need to know on which machine stuff happens on

[raven bone]

right

so i might have a problem with triggering the Inventory - Add Item, because sometimes it runs only on the server, sometimes multicast (using the same action)

ill have to investigate

also

[nova wasp]

inventory managed by an rpc seems very difficult to reason about

it should be sent as a replicated property or object

rpcs can represent requests from the client to do stuff of course (you kind of have to)

but from the server their state should be persistent and not rely on rpcs showing up

[grand axle]

Anyone know why faking lag in editor settings doesn't work 5.6? Using listen subsession

[nova wasp]

show your advanced play settings and the in-game stat net

[raven bone]

which settings should be used here?
to test derived attributes in GAS i have to sometimes run it under separate processes
seems like sometimes it would affect how events are firing

[grand axle]

Also another issue I'm having when playing on laptop vs computer one player has faster physics than the other, I tried substepping and async and multiply axis input by get world delta time but nothing worked

[nova wasp]

what does "multiply axis input" have to do with physics?

[nova wasp]

it depends on what you are doing... generally it is faster to run in the same process as they can share the engine and assets

but testing that way is not as realistic as a distinct process

there is not one holy setup that you will always use... it depends on what you are testing and why

[grand axle]

It's what Google said to do and ChatGPT

all physics is handled server side and inputs are sent through to server

[nova wasp]

I am asking what you are actually doing, I don't care what chatgpt says

why you did it isn't really concerning to me... I am trying to determine why your frametimes are changing the outcome

[nova wasp]

might be easier to use a select by bool pin for the float

with this does it send every frame you hold the input?

two things...

I would first suggest marging this input into 1 rpc

making these separate does not make sense

unless they don't ever happen at the same time

[grand axle]

it's for this type of game

[nova wasp]

also you should embed wether or not the attach button is held into the rpc too I think

making it a replicated property is going to make it random whether or not it's true or false effectively

unless that's the idea, I don't know

also one thing that might help would be making thiis only apply once per frame

instead of multiple rpcs arriving in one bunch applying twice

you can accumulate or only accept one

as for the physics on the server just test it with different fps t.maxfps 30 and whatnot

you don't need a new computer test tickrate

you can even set each client to use a different tickrate in network emulation settings (this just makes them tick less than every time based on how much time has passed)

[grand axle]

thanks for the info, seems like it's not an fps issue, but the amount of packets being sent like you said, I will see how to limit that somehow

[nova wasp]

spamming rpcs for input is fine but it doens't make sense to send anything when the input axis is 0 (unless it being set back is meaningful)

[grand axle]

I think I will need these as seperate as they might be on keyboard and only be moving on X or Y

[nova wasp]

why does that force you to make two rpcs?

you can just make one that uses both inputs and sends 2 floats

[grand axle]

Oh you mean like this

[nova wasp]

that sends the rpc twice

you want to send it once when the value changes to be meaningful

it might be easier to just check on the player tick honestly

unless you can bind to either value changing in a way that always uses the second one last

[grand axle]

hmm

how would I combine the inpit axis and only send one rpc?

[nova wasp]

just get the input axis on tick and send if either one isn't 0 imo

[sonic obsidian]

Can replication be turned on at runtime on a actor. A actor that was loaded from the map package not spawned at runtime. I've tryed AActor::SetReplicates but that doesn't seam to work i have a weird type of replication scenario anyway.

[nova wasp]

Dormancy might be what you want here

[sonic obsidian]

Im not sure about that heres my scenario I have a component that can be placed on a actor that the user has no control over the replication for example take the pcg actor that holds all the ISM components created from your PCG graph you can't set that actor to replicated via the unreal ed ui is there a way in my component to force whatever the owning actor is to replicate as my component needs replication

[grand axle]

like this ?

Ok, just read the documentation on the RPC and what not, I had a-lot of variables set to replicate instead of repnotify. I changed that so there should be like 1000s of less traffic now

Hmm, client on my laptop moves even slower now

[nova wasp]

.1 is not a good value here to compare I think unless you want to ignore lower values

controller input could easily send a 10% alpha

[grand axle]

mine is 5% I think

[nova wasp]

repnotify just adds that they will trigger a notify function

it does not really change how they replicate, just makes the client able to react to changes from each incoming read packet

[grand axle]

Also the Tick thing didn't work but using Async Physics tick does kinda work. only thing the physics body that follows the physics object gets dragged behind lol

[nova wasp]

running it in Async Physics tick is a good idea, because that way it runs at the same consistent rate

can just read the incoming input there on the server and it should be more consistent

[grand axle]

It's definately more reliable

Both my laptop and computer run (almost) the same speed regardless of framerate

Only thing now is trying to fix my skeletal mesh that makes the jiggle's root bone contrain properly

I have it on kinematic and constrain and all that but nope

even tried ooveriding async but that messes everything up

[verbal ice]

Out of curiosity, what happens if a player is mid-connection while the server starts a seamless travel to a new map

[fossil spoke]

You mean they start loading the previous level then the server moves to another level before they finish loading?

[verbal ice]

Yeah pretty much

[fossil spoke]

They will finish loading the first level and then should immediately start loading the new one

The Client sends an RPC to the Server after they load a level, just to confirm to the server that they made it.

[verbal ice]

Good to know, thank you

[reef bison]

is it not possible to have different clients on different levels? i was hoping to create "worlds" on different levels like minecraft dimensions, but every time i try to travel to a different map all the clients go together

[verbal ice]

Not with a vanilla engine, would probably require some heavy modifications

[reef bison]

uh thats looks like a big limitation

how does fortnite handle different matches on the same level? is it a different concept?

[verbal ice]

wdym?

Fortnite does it?

[fossil spoke]

Unreal only supports 1 world out of the box.

[reef bison]

i mean you have different matches on the same map, so you have different instances

[fossil spoke]

Those are different matches entirely?

[nova wasp]

what does "different matches on the same level" mean in fortnite?

I think they run multiple servers on a single machine by having 1 per thread but that's multiple distinct programs afaik

[reef bison]

idk i never played it, im trying to find a relatable unreal example

so i suppose the matches are entirely separated sessions

[verbal ice]

They definitely are

[nova wasp]

the simple solution is to just have the other match occur far away from the regular area in a distinct space

[fossil spoke]

Thats highly likely, thats how we manage instances for our dedi servers

[reef bison]

that makes sense, i was thinking it would be more like a dedicated mmo server where you can have players on different maps but on the same "channel"

[nova wasp]

plenty of games with "parallel worlds" where you just move 1km to a different area

I would say unless you really benefit from the isolated world it would be easier to just have them far away from one another

I don't know if there's a simple way to have a encapsulated networked world

spawning a new world is not that hard

but making it networked I have no idea

[reef bison]

yeah that seems to be the way, i was wondering if i was missing something obvious to have multiple worlds

[nova wasp]

that's fair

[reef bison]

thanks

[thin stratus]

There is a concept of having multiple words. People have also managed to get this going without Engine Modifications. Those worlds will not really be able to communicate with each other though, at least not out of the box.

There are two to three concepts floating around:

1. Client with a local world that is cut off the networked world an is thus only available to them.
2. Client with a local world that connects to a different Server for a more seamless travel between servers.
3. Server with multiple worlds, where clients can connect to individually.

In theory, they are all relatively straight forward, as they follow the design that the Editor already follows, which is to simply spawn a second UWorld, like the Editor does when you start playing in it.

Those UWorlds are ticked in sequence though. You can probably communicate between them, as you can simply grab both, but there are obvious limits, as both have their own NetDriver, PhysicsWorld, etc.

And then there is the more obvious limitation of performance. One world is often already pretty hard to keep running smoothly. I assume Fortnite runs one match per DedicatedServer, and then potentially a few of them per cloud server instance. Exact numbers, no clue.

There is also a plugin you might come across that talks about multi worlds, but iirc it's currently limited to concept 1.

I think concept 2. is what you can find out about a bit in this video: https://www.youtube.com/watch?v=2DGB8PLnQgk

Epic is also working on a "MultiServerReplication" plugin iirc. But not sure what the features of that are and how far it is.

Did you try and check the NetRelevantFor logic in C++ to see why it even says it's not relevant anymore?

[pseudo wagon]

Your actor does not have root component so that net relevant will false.

[exotic wasp]

there's a few plugins that I literally learned about today that allow multiple simultaneous UWorlds

[fallen fossil]

For standalone, owner is the same as player controller 0 right?
and it is different with dedicated server?

[quasi tide]

Standalone - it is fine to rely on PC 0 for those things. Unless you're dealing with splitscreen

For dedicated, just don't

[fallen fossil]

ye, I just got spagetti and want to do simple switch between has UI or does not have UI

[verbal ice]

Hmmm is there a way to extend the connection flow to a dedicated server without engine modifications?

Think having the flow not be complete until additional data is received from the server

Trying to implement downloading Lua scripts from a dedicated server while connecting

I could do it post connection and just have an additional loading screen, probably the way to go with Unreal, but was curious about alternatives

[reef bison]

thank you for the video, it changed my mind on some concepts

i was initially thinking about your point 3, where i would have a single server process with multiple worlds, but i dont really need that

[thin stratus]

Hm, you could communicate that initially via Beacons.

[reef bison]

having a proccess for each world sounds way more managable and actually easier to integrate

[verbal ice]

That was my (original) plan, but I kinda want to have hot reload of lua scripts at runtime, or after a seamless travel

Which would bypass beacons

[thin stratus]

Yeah, I mean it's more or less the standard way tbh. Unless I'm misunderstanding what your idea now is.

[thin stratus]

Well, you can have both?

This was about the initial connection and the initial load after all.

[verbal ice]

Yeah that flew above my head, it would need to be after a seamless travel too

The alternative is: does the engine have a way to pretty much delay replicating anything to a player until I say they're fully connected and ready?

i.e. no other players, no gamemode, no gamestate etc

Without manually making them irrelevant to the player

[thin stratus]

Iirc Login has a Callback by now.

Don't know what you can all do during that time though.

[reef bison]

yeah now that i stop to think about it you are right. i did some research early today and this is exactly what dont starve does for the caves

i think the concept of multi proccesses didnt sound good for me initially because i have only been testing the game on PIE with a listen server so i am not sure how that would work in practice

[verbal ice]

hmmm

[thin stratus]

With a ListenServer you are probably stuck moving everyone at once.

The whole idea of multiple worlds and what not is usually a DedicatedServer topic.

[verbal ice]

What about Iris, does it have anything that would make it trivial to "skip" players? Kinda like not adding a player to any node on the replication graph

[thin stratus]

Idk if you ever played Borderlands, especially the older parts.

They all ran on UE versions iirc, and you could only travel together.

[thin stratus]

Potentially, but that could turn into a strange micromanagement

You are better off just creating a custom Socket connection at that point.

[verbal ice]

Would have to be over Steam Relay so that'd be a pain to implement

I already have to implement support for SDR over online beacons relatively soon

[thin stratus]

Never went beyond the options I listed. You could check what beacons actually do to allow a connection.

[reef bison]

my objective is to allow the player to host servers from the game, it should be as simple as "Menu > Multiplayer > Host" and then someone else can join. this might be a silly question but is this stricitly done with a ListenServer, or is it possible to spawn the extra servers for the other worlds?

[thin stratus]

I mean, in theory you do that via a ListenServer. So I player that is acting as the host.

[exotic wasp]

usually any server packaged with a game is going to be a listen server

entire extra servers probably goes beyond that

[thin stratus]

You can start a process with a Dedicated Server too, but it's not that typical to do.

[verbal ice]

Issue 2 with Beacons was that now I have to implement it over beacons & in-game if I do hot reloading support, or seamless travel lua reloading, would have to open a beacon again

[exotic wasp]

IMO this is best

[thin stratus]

Majority of the stuff can live in some shared classes though or not?

[thin stratus]

The old Unreal Tournament Repo has that.

[verbal ice]

Maybe yeah

[thin stratus]

They start a DediServer if you want to play.

[exotic wasp]

it really sounds like you just need to have several levels as opposed to entire servers

[verbal ice]

Come to think of it I also have to implement networking of the map layout since it can be procedurally generated as well as custom made in a level editor

So that'd also be a step to connecting, so beacons are probably not a good idea for this since the map can change every round (map change)

[exotic wasp]

server can load any number and tell the client what to load

[verbal ice]

agh

[exotic wasp]

listen server will crumble doing that though

[thin stratus]

The absolute majority of UE devs/games does one of 3 things:

• ListenServer, where a Client is the Host.
  • Mostly used for Coop games.
• DedicatedServer, hosted away from Players.
  • Mostly used for competitive games, where the Client should not have access to the Server at all.
• DedicatedServer, hosted by the Players.
  • Mostly used for games that have some form of persistent world where players join and leave and every set of players can have their own. E.g. survival games like ARK or even Minecraft fwiw.

UT spawned a local DedicatedServer instead of dealing with also supporting ListenServers. But that's, again, rather unsual to do.

[exotic wasp]

keeps you from having different logic for the host ig

[reef bison]

yeah probably, im still studying this networking part so i am learning the different solutions

[exotic wasp]

I had a similar issue a while back and the path of least resistance was two levels in the same world that are just far apart

[thin stratus]

Yeah

The Ascent had some inner levels that were basically just far apart.

If you entered them you were teleported away to the sublevel.

[exotic wasp]

with large world coordinationates there's nothing to really worry about

[thin stratus]

Actively having that run on a different "process" or server would have been a nightmare to manage.

As soon as you do that and you need those players to still be somewhat connected (maybe see themselves in their HUD), you notice that there is no communication between 2 servers.

[exotic wasp]

you can have a level the size of the solar system and still not have precision issues

[verbal ice]

ARK also has listen server support, but they have a "tether to host" limit where you can't be over X meters away from them for performance reasons

[thin stratus]

ARK is also shit

But that's a side note

[verbal ice]

yup

[reef bison]

the game i am trying to build is very similar to minecraft in the sense that you have a "surface" world and you can enter a portal to go to the nether. but their map is infinite so it makes sense to split into entirely separate things

[exotic wasp]

abiotic factor doesn't do that and the listen server lags if there are multiple zones loaded

[thin stratus]

The Ascent didn't have that issue. The only thing we did was divide the game into multiple huge levels, that took quite some time to go through. You had to travel together between those, but inside a given level you could go wherever you wanted.

Was a simple matter of using Level Streaming properly

[exotic wasp]

if your world is infinite maybe multiple worlds makes sense

but literally anything less than infinite (that's a lot!) and same world is fine

[thin stratus]

Keep in mind that UE is originally build to be an arena shooter engine.

A lot of limitations are still present due to that.

If you want to make an infinite world game like minecraft, with subworlds and what not, then UE might not be the right engine in the end.

[exotic wasp]

really no engine has good support for multiple concurrent worlds

afaik

[verbal ice]

no public one at least, most of them just end up making their own engines for that

[thin stratus]

Yop, which is a whole nother can of worms

Not saying there are minecraft clones done in UE.

But you might also just not see where they use smokes and mirrors or simply accepted the limitations.

[reef bison]

yeah that makes sense, my game isnt infnite so i am gonna try doing everything on the same world for simplicity

i didnt like this solution because im working on a chunk loading subsystem and it bothered me that i would have huge chunk numbers due to having the other "dimension" really far away but thats such a small thing to worry about

[exotic wasp]

well you would probably just have two separate grids

chunk coordinates would be relative to their grid

[reef bison]

yeah thats one way to deal with it

i was also thinking about stacking the dimensions on the y axis

[thin stratus]

Having two worlds for this might sound like a cool idea, as you can just connect to the other world if you step through a portal. But you gotta remember that it might make the whole server side of things more expensive in CPU time and that you might not have any proper data sharing between the two.

At least not without some extra work.

I could imagine this working though, fwiw.

[exotic wasp]

it would really only be beneficial if you had a high quantity of instanced worlds I would think

[thin stratus]

Just needs a ton of boilerplate code. In theory both worlds are in memory of the server, so as long as you can get a hold on either of them, you should be able to even move data between actors of them.

[exotic wasp]

I believe the two worlds could be async as well

[thin stratus]

Not sure, if it's just 2 dimensions, it might be fine to have your ListenServer just use 2 UWorlds.
But no one really has done that, so there might be a lot of hurdles and if unlucky some small engine changes required.

[thin stratus]

That I highly doubt

[exotic wasp]

there's a plugin that I shant name because it isn't free, but it looked interesting

[thin stratus]

You need them to tick in the FEngineLoop. There are tons of UObject related things that don't allow being done outside the GameThread.

[exotic wasp]

there's also an engine branch for server meshing I think

[thin stratus]

I looked into this shit cause I wanted to run a second world for HitValidation.

[exotic wasp]

ah

[thin stratus]

Creating a second world is in theory really easy. I never did it for this kind of stuff, but I have created fake UWorld for Functional Tests.

There are some tutorials about that at least. I'm sure if one follows what the Engine does for creating the PIE world, or generally the first one, it should be easy to wrap that into some functions to greate "the second dimension" with a different level.

And since both of them will probably sit in the same array of FWorldContext, it should also be possible to access them.

And all their actors fwiw.

Might be a fun weekend thing to test out.

Client should probably only have one world though

And only connect to either of the worlds when changing portal

Might also want to only create the other world if there are players in it or so, idk

[reef bison]

its great that it sounds possible, but it sounds a bit beyond my skills for now 😅

i found a plugin that does this but the second world is client-only so it doesnt support replication

i think thats the hardest part to get working

thanks for the thoughts, i really appreciate it, it cleared a lot of questions i had

i might come back in a few weeks when i inevitably wont be able to get my chunks loading system correctly replicating to all clients heh

[sterile plaza]

I have an array of FStrings that's larger than a bunch. How to send this to a client?

FFastArraySerializer could be an option? Although if it bunches everything that's dirtied, that's still going to result in too large of a bunch. Amortizing the data across bunches seems like the way to go, but I haven't found any tooling to help with that. Is the answer to write that myself?

[grand axle]

I have smooth sync on my character and the root being a physics object, all physics is on the server but the client is not receiving the transform updates from server. not sure if it's a smooth sync issue or something else? If I read everything correctly client shouldn't have physics running which is what happens, but smooth sync isn't passing the physics root transform to clients

[copper vale]

You'd have a better chance getting a valid response from the dev's channel instead of here.

[grand axle]

I just gave up on that idea, although I wouldn't mind knowing why it won't work

I somewhat got it to replicate but not with smoothsync and even then. reliable yes, glitching all the bones on physics object Check.

[normal fulcrum]

Hey
I need to sync some buttons + levers in car cockpits using RPC from client to server.
player tries to push a button => Server_HandleInput(ButtonId, Payload), where payload is int8
the question is how to send ButtonId. For different cockpits ids for buttons are different. I assume sending FName "CarType1.Cockpit.HazardLightsButton" is less expensive than sending similar FGameplayTag.
It seems the best way is to encode all the button/event types as integers (or enum class?). Does it make any sense?

[chrome bay]

FName will serialize as a string. GameplayTags will serialize down to a minimal-size integer if you enable Fast Replication in project settings.

It's dumb that that settings defaults to false frankly

[normal fulcrum]

so let's say I could use FGameplayTag and it will implicitly be int anyway, so no need for If(id ==1) and I could use the tags for readability keeping the speed

Thanks for the setting, will look at

[chrome bay]

yep, but it'll likely be smaller than an int because the engine will only use as many bits as needed for any tag

[normal fulcrum]

good to know, thanks.

[chrome bay]

enums are still a good option too ofc, tags may make more sense when you need some sort of hierarchy/composition/modular game features.

[normal fulcrum]

well, let's say I just need to have ~ 100 buttons synced, and in every car type they do a bit different stuff

creating an enum per car type is kinda the same as having tags per car type

[chrome bay]

In that instance, I probably wouldn't make it per-car-type

Tags may make sense, but duplicating the same tags under a different top-most hierarchy is generally bad form, tags are designed to be used compositionally

E.g, a structure like this would be very awkward to run tag queries against

Car1.Button.Hazard
Car2.Button.Hazard

[normal fulcrum]

makes sense. in that case I would prob make Car.Button.Hazard regardless of the car type. but for specific stuff it would still require a unique tag Tesla.Tablet.Button1 whatever
so do you suggest using int+enums there?

[chrome bay]

If you're doing something highly specific, tags don't really make much sense IMO - but commonly people do use them as a replacement for enums

What structure makes sense is kind of up to your project I guess, but I would just caution against highly specific tags that describe game entities

The true power of tags comes from composition IMO

[normal fulcrum]

yeah, I see
at least, some data to try with
thanks

[sonic obsidian]

Can a actor marked not replicated have replication turned on at runtime (taking about a actor placed in the world not spawned at runtime)

[fallen fossil]

I don't think so

[sonic obsidian]

damn thats a shame

[fallen fossil]

I mean, meybe there is workaround but sounds wrong ;d

without replication it won't receive any updates, so you have to send enable command through different actor 🤔 and I don't know if you can even set this in run time

[tardy fossil]

whoa i JUST realized quantized vectors don't actually quantize if they are serialized in a custom structs NetSerialize function using Ar << quantizedVector;

[normal fulcrum]

If there are more than 256 unique buttons (imagining some special duty machinery), enum would be not enough, if I want to use it in dropdowns in BP (enums shall be uint8)
Tags are limited to 65535 unique entries per project, right?

[chrome bay]

Yeah but if you get anywhere near that you've lost your mind

If it's really that extreme, just index them I guess

[exotic wasp]

pretty sure you can?

[normal fulcrum]

you mean pure integer as is instead of enum/tag?

[exotic wasp]

you could just replicate the button and use it's reference as the rpc parameter

[chrome bay]

no

[normal fulcrum]

for a struct with custom NetSerialize, is it necessary to mark all the fields as UPROPERTYs?

[exotic wasp]

everything that you want to replicate has to go through a replicated actor at some point

[dark edge]

Is the layout static or dynamic? Like, do you know the set of all buttons at editor time?

[normal fulcrum]

completely static

[dark edge]

Unless we're talking about tons of states I'd just have the states live in the cockpit as replicated values

just 8 or so payloads

[normal fulcrum]

hmm. replicating server->client is another story (for some reason my custom struct NetSerialize has been never called)
though, to RPC from client, I need some IDs which buttons are "pressed" e.g. when player presses button1, I want to send to server(Actor, "Button1", true) and dispatch this call to corresponding Actor->HandleButtonPressed("Button1", true), where Actor has logic to handle all the different buttons pressed.

[dark edge]

If you don't have that many and it's common then the ID is which RPC is called

Dunno your setup though. I'm doing modular cockpits but each control is an actor in my setup

[normal fulcrum]

Do your controls have different actions on them? Do you call different rpc or specify “action type” as param?

[dark edge]

My system uses a graph

so for controls, they just output a float value to whatever they're connected to

ThrottleControl -> Engine.Throttle for example

But say you had some effed up vehicle that steered by differential throttle (2 engines)

You could:

SteeringWheel -> RightEngine.Throttle
-> Negate -> LeftEngine.Throttle

[normal fulcrum]

Well for my situation, i have ~ 100 (for now) buttons/levers which I want to call from client to server

Since rpc could be called only from objects client own, i put an rpc call on player character, specifying id of interacted button and payload

[thin stratus]

Did you add the Struct Ops for your Struct with the WithNetSerializer stuff? Can't recall if that was needed.

[rotund sail]

FYI working with ISMs are a pain, but theres a delegate you can bind to track index changes such that its possible to track changes in an external array (usually an issue due to internal RemoveSwap in the internal indice tracker)

[normal fulcrum]

Yes

[tardy fossil]

pretty sure netserialize should always be called when going through an RPC... i believe it's for triggering OnRep's when the custom struct is a replicated variable is when you need UPROPERTY

[normal fulcrum]

I will show the code later on

[normal fulcrum]

so. My idea is
having some buttons player can press (using BP callbacks now). This calls

UFUNCTION(BlueprintCallable, Server, Reliable)
void ServerCommandInt(AMyCar* Car, int32 CommandId, int32 Payload)
{
  if (Car)
  {
    Car->ParseCommand(CommandId, Payload);
  }
}

Car->ParseCommand calculates some changes in state, and applies it to replicated struct

USTRUCT()
struct FCarReplicationData
{
    GENERATED_BODY()

public:
    ECarGearShifterState GearShifterState;
    ECarBlinkerLeverState BlinkerLeverState;

    bool NetSerialize(FArchive& Ar, UPackageMap* Map, bool& bOutSuccess);
};

bool FCarReplicationData::NetSerialize(FArchive& Ar, UPackageMap* Map, bool& bOutSuccess)
{
    uint8 StateGearShiftByte = static_cast<uint8>(GearShifterState);
    Ar << StateGearShiftByte;

    uint8 StateBlinkerLeverByte= static_cast<uint8>(BlinkerLeverState);
    Ar << StateBlinkerLeverByte;

    if (Ar.IsLoading())
    {
        GearShifterState= static_cast<ECarGearShifterState >(StateGearShiftByte );
        BlinkerLeverState= static_cast<ECarBlinkerLeverState >(StateBlinkerLeverByte);
    }

    bOutSuccess = true;
    return true;
}

ACar is smth like that:

void AMyCar::ParseCommand(int32 CommandId, int32 Payload)
{
  if (CommandId == 1)
  {
    GearShifterState = Payload;
  }
  else if (SwitchIndex == 4)
  {
    BlinkerLeverState = Payload;
  }

  FCarReplicationData NewData;
  NewData.GearShifterState = GearShifterState;
  NewData.GearShifterState = GearShifterState;
  ReplicationData = NewData;
}

UPROPERTY(Replicated, ReplicatedUsing = OnRep_ReplicationData)
FCarReplicationData ReplicationData;

AMyCar::OnRep_ReplicationData()
{
  GearShifterState = ReplicationData.GearShifterState;
  BlinkerLeverState = ReplicationData.BlinkerLeverState;
  UpdateVisuals();
}

Any ideas why OnRep_ReplicationData is never called? any suggestions?

[tardy fossil]

the 2 enums in your struct need UPROPERTY

[normal fulcrum]

well ,they are not BlueprintTypes, so I thought it was OK to have NetSerialize

[meager spade]

Hmm i would honestly just male a proper struct and only send the data you need

Chaos vehicles has an example or maybe the old one.

Where you send the data that had changed

[lament flax]

do you guys do anything special when you have a hig hcount of a smale base class of comps ?

[meager spade]

I hate generic functions that look ambiguous

[meager spade]

No?

But context matters

[lament flax]

like a replicate manager

[meager spade]

Vague question

[lament flax]

yeah i know, i dont know how to precise it without giving a full paragraph

[meager spade]

Give paragraph

[lament flax]

to late for my lazy brain 😅

[meager spade]

:/

I mean use push model for props

[lament flax]

ill try the push model, never used it before

[meager spade]

Mark actor dormant and only ipdate of it has changes

[lament flax]

any ressources on what changes and what func/macro to use ?

[meager spade]

Normal networking stuff

[lament flax]

just have to call a macro when you want to mark dirty nah ?

[meager spade]

Check the push model header

Most of the time you just mark a property dirty

Best is actually looking snd like playercontroller.cpp and actor.cpp

And how they did push model

Couple of nice mscri

[lament flax]

those are big classes but ill look

[meager spade]

Mscros

Just search like getlifteime

And MARK_

Also remember enabling push model means you must mark dirty for a prop to replicate

In our project all props are push model

[lament flax]

so for the "paragraph" here is a short version
i got a component class which holds an array of Uobject which acts like fragments

you can easily have a lot of these, maybe one per any actor containing logics (so hundreds per map i assume)
so since im replicating the UObjects as a subobject of the actor component, and each UObjects replicates there properties, i wonder how i could optimize that a bit

[meager spade]

Oof

Tons of objects

[lament flax]

yea, i debated a lot between using UObjects or instanced struct as fragments

[meager spade]

Yeah I originally did that with inventory then switched to fast array and not replicate the item instances but create them locally lol

[lament flax]

since they need functions and stuff i went for UObjects, more BP friendly to

[meager spade]

But that's specific scope

Won't work here

I mean it could work but would require a lot if thought

But 1000s of subjects per actor

Is going to be costly and if these dint change st runtime pointless

[lament flax]

not that much, maybe a few fragments per comps

(i mean, its a plugin so i need to be robust for intensive usages)

[meager spade]

Bug if they have rep properties then it will be expensive

[normal fulcrum]

So, can you a bit wrap up it for me. If I have a lot of replicated properties, I though they will make an overhead (as discussed before with some people here)

though I decided to make the atomic struct + pack enums to bits later

Are there any better methods you've mentioned?

also ,what's about Client->Server RPC, any better idea?

[meager spade]

Either way you cut it. You need to rep initially all objects

This is the cost then you have the cost of checking the objects.

Even with pushmodel

[meager spade]

More rep props don't increase it that much especially with push model. When I did vehicles the vehicle had a state component that handled all interaction snd replication. We didn't do anything super special. We dilly sent a struct on tick unreliable with the clients local state data and replicated a version out to simulated

Nothing super special

This strict held some packed values bitflags etc

So ur was quite cheap

Cmc does exactly this also

Sorry for typos on phone

And it's late

[normal fulcrum]

hmm. I'm not familiar with Push Model
But I mean you send the state data as RPC, like not 1 RPC for each button like I did? (with id + payload)

[meager spade]

We send a full struct every tick

Optimised of course

[normal fulcrum]

well probably I need to make something similar
though, such struct shall have uproperties + replicated flags on them?

[meager spade]

With what yhe player wants to do

Yes ofc

Let me see if I can find what we did

Give me a few

[normal fulcrum]

Appreciate. Because this drives me nuts, especially with amounts of buttons I plan to have (and not only cars in consideration)

[lament flax]

is there anything i can use if i rep only once ? (read only frags), i only know about the COND_RepInitial thing

[lament flax]

i wish there was an update rate per replicated subobject

[meager spade]

oops

didnt mean to dm @normal fulcrum

[lament flax]

how does replication happen when collecting props ?

does the world/engine gets all replicates actors on tick, and for each actor they get the replicated comps, and for each replicated comp they get their replicated subobjects

(while checking for dormancy and rep rate)

and @meager spade, would having a giant fast array replicated in one replicated actor singleton be better ?
one one replicated comp, which sends the updated data to all comps with their fragments

so like, each entry of that fast array would be

• the comp we refer to
• fragments

[fossil spoke]

UNetDriver::ServerReplicateActors_BuildConsiderList Is a good place to start.

Though, Im not sure how much that has changed in UE5

With Iris and all the rest

RepGraph etc

[lament flax]

im not using either

[fossil spoke]

I have a free plugin I wrote that uses InstancedStructs to do an Itemization system. Take a look in my bio for the link.

It might give you some ideas. Its fully replicated using FastArrays.

[meager spade]

still expensive

cause you have to gather all replicated subobjects, then check if they need to be replicated

and this has to be polled

cant be push model (even with iris)

so its pretty expensive if you have a bunch of Subobjects

like 100s per actors

[fossil spoke]

I was only responding to his question about what process the engine goes through.

Not what is optimal

[meager spade]

yeah you mention what has changed etc

but nothing really still same as in UE4

[fossil spoke]

Good to know

Does Iris change all of that path?

[meager spade]

Iris is a bit faster

and UE5 has a smarter way (where you register to a list the objects)

but you still need poll those objects, you just skip a few steps

[fossil spoke]

Right

I really wish I had the time to catch up on UE5

[meager spade]

        {
            ReplicatedOwner->AddReplicatedSubObject(Child);
        }```

like you can push and pull the objects

without needing the for loop or w/e

[lament flax]

the thing is w/e solution i would find with instanced structs, i will not change to that over UObjects because instanced struct cant have things a UObject has.

unless i do an array of uScripts ... and accept both with custom property view

[meager spade]

in ReplicateSubobjects virtual

[fossil spoke]

What do you mean?

What things are you talking about?

If you mean like Blueprints that can execute code, you can easily associate static Blueprint Objects with Structs.

The Struct is the Data, then the associated Blueprint is the code that can operate on that Data.

You only replicate the Data.

[meager spade]

thinking about it, you could do it smartly and only replicate the object if the object has changed, but i think removing it will destroy the object on the client side, so this might not actually work well. GAS abilities uobjects are not replicated unless you mark them replicated, so you could design the system to have a specific "bReplicate" to only replicate that subobject that needs replication, and not all of them.

some might not ever need replicated properties

and just be built from the static data

[lament flax]

virtuals and graphs yeah

and just, functions ...

[meager spade]

thats exactly what GAS does

[fossil spoke]

Take a look at my plugin, you might learn some new things.

[lament flax]

it my case they can be pure data frags or actually hold logic

[fossil spoke]

Or look at GAS

[meager spade]

if its not replicated, create local instance of the objects, if it should be replicated, only replicate that

instancedstructs are the new cool

[fossil spoke]

https://github.com/mattyman174/GenericItemization

[meager spade]

you dont blindly want to replicate all subobjects

client can check if the subobject should rep or not and create a local copy

server does the same but produces a rep copy if it should replicate

this leaves it in the designers hand

[fossil spoke]

@meager spade Do you know of a reasonable way to selectively replicate properties to specific connections? Is Iris capable of this?

[lament flax]

is that what ur plugin does for items logic ?

[fossil spoke]

Yes

[meager spade]

hmm, right cause GetLifetimeProps only has a bunch of defined options but not per connection

[lament flax]

i need to look into these

[meager spade]

iris uses the standard GetLifetimeProps, for props

[fossil spoke]

Look at FItemInstance and FItemDefinition

[meager spade]

so i don't think so, and i havent seen "property filters"

[lament flax]

its almost been 6 months without MP, i didnt dig into what unetdrivers, unetconnections and unetactorchannel really are and managed

[meager spade]

only class filters

[fossil spoke]

One issue I ran into with FastArrays and InstancedStructs for itemization was that, there wasnt a reliable way to not replicate everything to everyone, outside of just using OwnerOnly.

Thought maybe UE5 had a solution

[lament flax]

feels like doing a giant FA is the best way for a generic system

(living on a replicated actor singleton)

its just that testing the few paths and profile them takes a lot of time, time that i dont really have

[fossil spoke]

Sometimes its useful to architect a solution you think will be performant enough, then profile that.

If it doesnt meet your requirements, you can either optimize it, or use the knowledge you gained to explore a different approach.

You will never write a perfect system the first time.

[lament flax]

thing is im doing a plugin so my requirements are "the most i can do"

i guess supporting 500 replicated actors with each 10 fragments would be the maximum ill optimize

[fossil spoke]

You have to start somewhere, having no plan is worse than having a bad plan.

[lament flax]

my infamous inventory rewritten 5 times 😆

[fossil spoke]

Experience and knowledge come with time.

[hollow saffron]

any of yall using server streaming? or know of any writeups on the issues it comes with?

[fossil spoke]

If you write an inventory system enough times you will become an expert.

[lament flax]

im maybe reading to fast or to tired to find it, but i cant find where your items handle logic in your readme

[fossil spoke]

Depends on the logic.

An ItemDefinition is the static data of what an Item is.

This is where you would store objects that execute things for that ttype of item.

An ItemInstance is an instance of an ItemDefinition

It has access to its ItemDefinition and would then use it to execute whatever logic it needs.

https://github.com/mattyman174/GenericItemization?tab=readme-ov-file#5216-user-data

The UserData is the most generic spot to add Objects that would execute Logic for an Item

https://github.com/mattyman174/GenericItemization?tab=readme-ov-file#5217-item-instancing-function

The InstancingFunction is more specific to when an Item is created, you can override it to setup dynamic elements or introduce new functionality for when an item is instanced.

The Affixes are also something that could be used to store "logic"

For example you could theoretically extend them to apply a GAS ability to the Character.

But that isnt covered in the sample project

[lament flax]

so your item instance is an UObject defined from a FInstancedStruct

which holds an array of user data which are UObjects

[fossil spoke]

No

An ItemInstance is not a UObject

An ItemDefinition is not a UObject

They are both InstancedStructs

They can hold onto references to UObjects

[lament flax]

NetDriver.h and PushModel.h are suprinsingly well documented

[snow trail]

hello, im having network issues in my game (i think), some stuff appears to desync or fail to sync and sometimes even reliable events fail to fire it looks like
it also starts to have an impact on the game thread (i think? unless its caused by something else, most of the load appears to be in the EndPhysics tickgroup although i dont believe i have anything taxing that operates in that group)

what i want to ask is, would enabling iris replication somewhat help with this? or is everything in my hands with optimization?
i dont expect it to be a perfect fix that i just turn on and everything magically works, but if it does help make things faster before i try to start optimizing, it would be good

[nova wasp]

profile it to see what is actually slow

if net broadcast is slow you might see some gain from iris but be warned it is experimental and may not work for what you are doing immediately

[snow trail]

thing is, it doesnt happen in the editor and especially not on a clean save of my game

[nova wasp]

it might help to just reduce distinct network objects or catch something capping out bandwidth from sending redundant data (network insights)

[snow trail]

hmm

[nova wasp]

the editor is not the only place you can profile in

the editor IS YOUR GAME

just with some extra buttons on it

Insights can attach to remote and local builds that have tracing, a test build would emit events

[snow trail]

oh huh

i had no idea insights could do that

[nova wasp]

you may need to start it with a console command I guess

trace.file MyEpicFile default

then trace.stop

network insights is more involved but you can just read the docs for that

[snow trail]

hmm alright, thank you

i need a friend to wake up before i can start profiling since i dont have the issue in any of my save files, ill come back here if i find anything interesting

[nova wasp]

It would be a good idea to have them send it in for you to mess with locally in a test build

if possible

sometimes stuff like this is hard to reproduce though so hopefully that is enough

[snow trail]

yeah, but i need them to be awake for that lol

[nova wasp]

as for network bandwidth profiling network insights is the fanciest (shows every single packet in detail) but you need to launch the game with some args https://dev.epicgames.com/documentation/en-us/unreal-engine/using-the-network-profiler-in-unreal-engine -NetTrace=3 etc (I think 3 is the max as of 5.6)

[snow trail]

ive tried getting the savefile and running but i wasnt getting any issues, i also tried joining my friend's hosting world and i wasnt getting any framerate issues anymore unlike yseterday
he was getting terrible framerate though as the host so i got him to profile for me

i posted about it in the profiling channel
#profiling message

[lament flax]

any useful commands to see net stats ?

like how much time the engine took to iterate replicates actors, processes bunchs, etc

[nova wasp]

that stuff is easy enough to see on a regular profile

[lament flax]

yeah im just wondering if their is any cool net profiling commands, so i dont have to trace each time to check how much time it takes

[nova wasp]

I can't think of one but that seems like something that would be a decent stat command

I just use regular profiling every time

Stat net is mostly bandwidth related

[lament flax]

is there a "authority" defined for a not replicated actor/component ?

because, if its placed in the level, it exists on both server and client, but if its spawned/added at runtime, it will only exist locally

[nova wasp]

remote role might be useful here

also role isn't really set until it arrives afaik

unless there's another flag I missed

[lament flax]

okay

i assume not replicated actor/comps cant send/receive PRCs ?

thats what i got for now

main issue for me now is this will save some overhead in MP context, but will add some in SP context.
i guess i'll have to make 2 PSC classes, one for SP and one with networking

so it would look like that

[woeful ferry]

Actor components that are not replicated can send/receive RPCs if the owning actor replicated

[meager spade]

correct

this is a useful detail

[exotic wasp]

Really? I had no idea that was a thing

I guess they're stably named as long as they aren't added at runtime

[woeful ferry]

Yea because they use the owning actors callspace or whatever its called

[quasi tide]

Even if they get added at runtime, this works.

It relies on stable naming though. And there is a function IsStableyNamedForNetworking or something like that.

[exotic wasp]

Well that's what I was going for; adding a component at runtime won't give it a stable name by default

How much of a cost is even associated with marking a component replicated any way? Esperanza with push model?

[quasi tide]

I believe it is a pretty tiny cost. It sends the identifier for the component (which is an additional int I think?). Other than the normal cost of having something replicated. Which is mitigated with the push model.

[chrome bay]

If it's not replicated, the local instance has authority over it

Also true if it is replicated but the actor is torn off

[exotic wasp]

thanks autocorrect for turning especially into esperanza

[normal fulcrum]

Is it possible to mix Ar << and Ar.SerializeBits in one NetSerialize function? I assume yes, but it's a bit tricky how to read those with Ar.IsLoading(), I have some garbage values, do I need to unpack it somehow specially?
Appreciate any reference

[wise ermine]

https://www.udemy.com/course/multiplayer-in-unreal-with-gas-and-aws-dedicated-servers/

I'm sorry if I shouldn't post the link here, just let me know and I'll delete it.

Is there anyone who ever used it? Do you recommend? Or do you have other recommendations to learn multiplayer (globally, not LAN)?

[lament flax]

i dont have for AWS, but for general MP: https://notes.hzfishy.fr/Unreal-Engine/Networking/_Resources-(Networking)

the docs are a good entry point to https://dev.epicgames.com/documentation/en-us/unreal-engine/networking-and-multiplayer-in-unreal-engine

[wise ermine]

thanks

[robust linden]

generally speaking, how is aiming replicated and predicted?

say I want to cast a ray to the direction the player is facing

[exotic wasp]

depends on the game

[robust linden]

I don't think using GetCameraComponent()->GetFacingDirection() is safe to use for multiplayer and locally predicted stuff

It doesn't seem to be correctly replicated if replicated at all

[exotic wasp]

many games have the player do the raytrace and determine if they hit something. then the server just verifies if it's generally possible that it could have occurred

[robust linden]

hmm okay

so generally speaking should I just associate to each "shot" a facing direction that will be sent to the server?

[exotic wasp]

some make snapshots of player locations in time. then when a player submits a ray to the server, it checks where players were at that point in time and sees what that ray hit

[exotic wasp]

that's safest but it'll feel like shit

[robust linden]

it's weird cause I've never had issues using GetOwner()->GetFacingDirection() in CMC

GetOwner()->GetActorForwardVector().GetSafeNormal2D()```

That I mean

it's always been in sync

[exotic wasp]

easy way to get some security is do that first method and see if the player hit the targets bounding box.

if that isn't enough use rollback

[robust linden]

hmm okay

okay I just thought for a second

maybe you could update the facing direction when needed in the CMC

send it over with the compressed flags and stuff

and use it as a safe variable on the client and server?

that sounds pretty good, predicted and will be corrected

[exotic wasp]

much better off having facing being a cosmetic thing unless you have game mechanics that heavily depend on it

in a perfect world you trust the client with nothing and the server performs/checks every single thing they do

[robust linden]

yes but I'm pretty sure CMC already handles facing prediction

[exotic wasp]

but also, trust the client as much as feasible for your game

[exotic wasp]

i am telling you it wont work as is 🙂

[robust linden]

ooh it actually handles rotation

how do I tell if, in the case of a listen server, am a client playing on the server?

because Authority will be true for every character

but how do I tell if this one specifically is controlled by the server? Can I use GetOwnerRole() & ENetRole::Autonomous_Proxy?

basically a IsLocallyControlled() function, that exists with GAS but idk for the rest

[exotic wasp]

GetNetMode()

also if you're doing listen servers dont even worry about preventing cheating

host can always cheat

[steel anchor]

Hi, I have a question about Launch Character.
I’m not sure why, but on the server, the character’s mesh sinks into the ground after launching, while on the clients everything looks fine — the mesh doesn’t sink.
Does anyone know what might be causing this? (Multiplayer problem)

[hollow frigate]

sounds like a desync, are you multicasting the launch?

[lament flax]

when using DOREPLIFETIME_WITH_PARAMS_FAST i gives me this error for my fast array

other macros (not fast) works fine

(my goal was to use push model with FA)

[fossil spoke]

PushModel doesnt work with Arrays

[lament flax]

it does

there is macro for that, you need to give index tho

[fossil spoke]

That must be new in UE5 then

[lament flax]

or maybe only StaticArray

yeah

[fossil spoke]

Did you forget an include?

[lament flax]

the type is in header

[fossil spoke]

Did you forget the PushModel header as an include?

#include "Net/Core/PushModel/PushModel.h"

#include "Engine/Public/Net/UnrealNetwork.h"

Typically need both of these

[lament flax]

neither

okay, i was missing a replicated attribute

[lament flax]

when adding fast array entries, i should be able to keep a ptr to the struct entry fine until its removed right ?

[verbal ice]

Is it a static array?

[lament flax]

no

[steel anchor]

No I launch character only on the server

[verbal ice]

Unsafe then

[lament flax]

this

FPSPrefabPropertiesObjectEntry& APSPrefabNetManager::AddEntryFromObject(UPSNetPrefabSourceComponent* NetPSC, UPSActorPrefabPropertiesBase* Object)
{
    return PrefabPropertiesObjectContainer.Entries.Emplace_GetRef(NetPSC, Object);
}

[verbal ice]

Other elements being added/removed can screw with it

I mean if you dont touch the array for the duration of that reference then w/e that's fine

Just don't try to store it

[lament flax]

so, no way to store that

sad

i assume c++ is super fast at iterating arrays ?
i maybe doing to much micro optimization just because ill have 1k-5k entries

[fossil spoke]

You will be fine iterating it to find an entry.

[snow trail]

does this need optimization 🥀

[chrome bay]

ItDepends(TM)

[snow trail]

lmao

tbf that only happens when joining, after that it settles down

but it still hangs around 4000 as soon as a player starts doing anything so idk

no idea why its so high though

there shouldnt be anything actively replicating really and 90% of replicating actors have the Min Net Update Frequency set to 0

[chrome bay]

Only way to find out what it is would be to use insights and see what's replicating

That way you'll get per actor/per-property breakdowns

[snow trail]

network tracing works only in packaged right?

https://tenor.com/view/steins-gate-mayuri-mayuri-shiina-mayushii-anime-burger-gif-7177124858534691268

[nova wasp]

it works even in editor builds

[snow trail]

i tried it in the editor earlier (i think?) and i didnt get any trace

[nova wasp]

It can be fickle but I find stopping and starting a trace before pie can help

also show startup args

[snow trail]

on the editor? i set none

lol

but rn my friend is online so i can test in a real game scenario

does this look ok?

[nova wasp]

the editor is also a game

[nova wasp]

yep, I think that's good

[snow trail]

do i need to trace.start and .stop still?

[nova wasp]

yes

[snow trail]

i havent analyzed the trace yet but

w h a t

[steel anchor]

I think is because I change the Capsul Collision Half Height

[robust linden]

is gravity scale predicted?

I'm assuming it's not

it's not even replicated 😔

so how would I change gravity scale predictively?

[tardy fossil]

how often does it change? personally i'd just replicate it as a variable in gamestate or something and have clients update it with an onrep and just let the slight rubberbanding happen when it changes

[robust linden]

it changes like once every 10 seconds

that would be bad to rubber band

i think i found a way I'll put it after

[robust linden]

Hey, despite having all these classes:

my safe variable (which is not a compressed flag) is not being passed to the server

the reason is that PrepMoveFor is NEVER CALLED

and I don't know why 😔

[robust linden]

Does FNetworkPredictionData_Server also needs to be overridden?

[snow trail]

im having an odd issue, some actors are just... not existing? my friend is spawning in a replicated actor and it doesnt sapwn in for me (connected client)
also some actors in the map havent loaded on my client either, very odd

all i could find in the log is this and some other similar lines but idk if its related

[2025.07.24-15.50.19:877][  0]LogNetPackageMap: Warning: InternalLoadObject: Unable to resolve object. FullNetGUIDPath: [3601]NOT_IN_CACHE
[2025.07.24-15.50.21:607][ 74]LogNetPackageMap: Warning: InternalLoadObject: Unable to resolve object. FullNetGUIDPath: [1941]NOT_IN_CACHE
[2025.07.24-15.50.21:607][ 74]LogNetPackageMap: Error: UPackageMapClient::SerializeNewActor. Unresolved Archetype GUID. Guid not registered! NetGUID: 1941.
[2025.07.24-15.50.21:607][ 74]LogNetPackageMap: Error: UPackageMapClient::SerializeNewActor Unable to read Archetype for NetGUID 1496 / 1941
[2025.07.24-15.50.21:607][ 74]LogNet: Warning: UActorChannel::ProcessBunch: SerializeNewActor failed to find/spawn actor. Actor: None, Channel: 239

[frank birch]

Can the HUD exist before the PlayerState? 🤔

[lofty mica]

Where do I mod the engine code for P2P? Do I need source code?

I want to optimize for 300 players and remove lag by basing the updates off the highest ping dude in game

[lament flax]

I think yeah, go where the HUD is made (i think PC or GM on Login) and where the PS is made (PC iirc) and see what gets logically always called before

[frank birch]

gestures vaguely at the entire codebase
You will need to modify the engine source to mod the engine, yes.

[lofty mica]

Has anyone tried locking packet updates to the highest ping player to remove ping advantage? So all players get updates at the same time

[sterile plaza]

You want to change the networking model from server/client to P2P and you're not sure if you need source code for that?
That work is incredibly complicated and is based on decades of work in the area.

[lofty mica]

Don't care just doing it no time to beat the bush

basically gonna P2P update across eachother and compare data(antihack) and lock to highest ping player, unless their ping is like, 300, in which case auto kick

[graceful flame]

You are describing deterministic lockstep which isn't something unreal engine supports by default

[lofty mica]

Thanks for the proper terms

[sterile plaza]

It's likely several months of work, if not more.

[graceful flame]

mostly used with RTS games

[lofty mica]

I'll just cut the fap from the schedule

[lament flax]

What type of game are you working on ?

[lofty mica]

should be done in a few weeks

an FPS with revshare

[graceful flame]

they talked about it in 2015 here ---> https://forums.unrealengine.com/t/please-add-deterministic-lockstep-networking-for-rts-style-games/19255/10

[lament flax]

Why p2p

[frank birch]

Lockstep fps? 🙃

the what?

[sterile plaza]

rubber banding is scary okay?

[frank birch]

why don't we throw ggpo with rollback too? 👀

[lament flax]

Yeah you cant mix lockstep and realtime shit like fps

[graceful flame]

Tim even replied on that thread

[lofty mica]

2015 and not been added? epic needs to cuthe fap

[sterile plaza]

you're telling me that if you add a 300ms delay to input in an FPS it won't feel good?

[lament flax]

If only you knew ...

[lofty mica]

P2P cuz I dont need a million dollar server

[lament flax]

UE misses a ton of shit

[graceful flame]

This is a question that has been raised in every generation of the Unreal Engine, and it is definitely not in the works. Lock-step execution can only work if every engine system on every platform is designed for absolutely deterministic execution. That’s not the case with UE4, as a big engine with a time-dependent game update model and reliance on a wide range of middleware and a wide range of platform and runtime OS-version and API-dependent optimizations. An engine could be designed this way, but UE wasn’t, and it’s not practicable to retrofit it for a special usage case.

Instead of lock-step determinism, I recommend working with the engine’s network replication framework. This code has fairly good performance and typical internet connections have plenty of bandwidth, so I expect it to be practical for RTS usage.

Also keep in mind that you can customize the engine’s relevancy checks (which determine whether a given object is replicated to a given client) based on the client’s view of the world, which will usually change fairly slowly in an RTS.

• Tim Sweeney April 2015

[lofty mica]

yeah I just started and i discovered that water is almost impossible

I made water in doom 2 in like a day

[lament flax]

People can cheat in listen server and p2p

[lofty mica]

not if they all compare data and send a special packet

it becomes a blockchain style

[sterile plaza]

and if you use quantum computers, then you won't have to worry about lag

[frank birch]

oh, I've been this naive before 🥹

[lament flax]

If it was that "easy" we would see robust p2p anti cheat already

[lofty mica]

naaahhhh internet moves at 2gbs and 144 frames/packets a second is a mere few KB of data a second. They just suck.

[frank birch]

you not only want p2p network, you want meshing with distributed authority

p2p is listen server, one person is the authority (and can cheat)

[sterile plaza]

isn't it too early in the morning for trolls?

[frank birch]

you want distributed authority

[lofty mica]

Why do uneducated hipsters always think real code is troll?

get lost

and it wont add 300ms

just means you can't play on slow internet

[sterile plaza]

that's not what I meant. RTS games add delay to input to hide the lag from using lock-step.

[lofty mica]

dude the last RTS game was made 100 years ago its starcraft

this is 2025 and DIY code>corporate garbage code

[graceful flame]

It isn't so much about the amount of data you can send over the network its more like how much of it can you processes during a single frame while remaining within ms budget.

[lofty mica]

I don't think that a few KB of data will effect that

[graceful flame]

So you ideally want to keep the amount of data traversing the network low because you still have to process all the other stuff too

[lofty mica]

I'll make sure to keep that in mind cuz just upgraded my PC to 400mb ram

[graceful flame]

so you said fps with rev share on p2p, so its a PVE coop?

because if its pvp you'll have lots of cheaters using p2p

[lofty mica]

no its a highly competitive fps but right now im new so I'm trying to get water to work, everything else is easy

I know how it works and how to do variable compares you avoid injects

right now I need to get water working

[graceful flame]

what are you comparing variables against?

[lofty mica]

themselves thats how you doit

[graceful flame]

then they just alter their memory and bypass

thats how cheats work

[lofty mica]

example: HP compared against: LastHP

sets flags for cheaters

[graceful flame]

without a server for the trusted authority of the entire simulation a peer can just alter their memory and do whatever

[lofty mica]

k

[graceful flame]

if u make the code check if var X is the same as it was last frame or whatever that doesn't matter because a cheater can just have their program alter the same part of memory every frame after your check runs

but with a dedicated server it simply doesn't alter the memory like that and so if a cheater changes the variable in their memory, the server corrects it after

Say for instance a cheater teleports right behind another player and then tires to shoot them, on their screen they'll see themselves teleported but then on the next frame (or close to it depending on lag) the server teleports them back because their location is invalid

Does that make sense?

Also just to be clear I'm not saying using a dedicated server as the authority will prevent all cheats but it will at least make it more difficult for them.

[exotic wasp]

2 day old account

[gritty mason]

Hello guys! I have an serious issue. Me and my brother test our multiplayer listen server project. When my brother the host, on my Pc the character animations jittering but when I am the host, on my brother computer the character animations are good. FPS is good, stabile 60-80 fps . What could be the problem?

[dry pebble]

Trying to understand FUniqueNetIdRepl. I have a custom UUID for each player in my database (I use Cognito for auth). Can I set that as the user's replicated unique net id in PreLogin from the Options passed in or what is the best practice here? I'm trying to use OSSv2 (the auth interface) as well

[dark edge]

Water to work how?

waves + physics, just volumes that switch the CMC to swimming, just visuals, what

[thin stratus]

Animations on the Server for Characters are ticked through the CMC, which is a bit painful for ListenServers. Iirc one can play a it with the settings for that but I haven't dealt with that in a while.

[ionic sapphire]

Is there an event that fires on the server when a client connects to it? I know there's a callback function for OnJoinSessionCompleted, but that seems to fire on the client system. I attempted to use OnSessionParticipantJoined, but that didn't seem to fire at all.

Originally this is all due to the server not running OnPossess after the client connects to it, as, by my understanding, it's supposed to.

In fact, it seems like, despite using ClientTravel and explicitly stating not to use Seamless travel and having Seamless travel turned off in the Game Mode, that the behavior I'm experiencing is similar to travelling Seamlessly. I'm not sure what I'm doing wrong. Any ideas are welcome!

[lofty mica]

@dark edge Flat plane water that collides at peak waves instead of a generic box

painting surface and realistic sim physics irrelevant

[grand axle]

Finally solved framerrate dependant physics sync. and changed from a physics asset to a controlrig character which now sticks to the characters root. but now for some reason the jump button and movement seems to be more laggy now and jump sometimes doesn't even work. any ideas? Jump is a simple sphere cast downward that sets a bool on repnotify and if true, set impulse Z

[dark edge]

Why does jumping involve a repnotify?

That seems silly

[grand axle]

My character does things in the air and the server needs to know if the character is jumping and not just not touching the ground

I wish they exposed what data I send in BP, I bought smooth sync to do this but I can't get it to work

[dark edge]

are you using Character or did you roll your own?

it kinda sounds like you have 3 ways of syncing here

[dark edge]

define "is jumping"

[grand axle]

My own

Ignore the Branch that does nothing. I am planning to put different surface types in there for ice/sticky etc

Ow, I had the jump release on not reliable lol

My bad!

Although I do have one other question

It's hard to explain. but at the begining of matches or join the prediction is off, and over time it becomes more reliable, is that just how it has to be or is there some setting I need to do to improve it?

This is at the beginning of session. you can see the like magnetic replusion (this is with 12Ping)

And this is a little bit more into the match where it resolves. I guess it's not really an issue pre se, I just want to understand it. I also increased the lag and everything that predicts still works as if 100% synced no issues

Jumping is now fine btw

I wonder if I have to kinda freeze everything, wait for all the players to join then start simulation on server. I think that's the cause?

So far with my 2 weeks of reading and learning all these new things I think I have finally managed to get stable physics online brawler working with my custom player controller. however just that first related issue which is prediction related. which I think will be solved when I freeze everyone for a split second when they join

[grand axle]

Yep, that was it. also can't turn off or on simulate physics anywhere in the player BP good to know! Game now has no bugs. that I am aware of

[dry pebble]

Trying to set up Online Services V2, with Common User/Common Session Subsystem (plugins from Lyra) but keep getting this error: Assertion failed: OnlineServices [File:C:\Development\Games\GameLiftUnrealApp\Plugins\CommonUser\Source\CommonUser\Private\CommonSessionSubsystem.cpp] [Line: 382]

I have a default service set like this in DefaultEngine.ini:

[OnlineServices]
GameDefined_0=OSBackend
DefaultServices=Null

And also setup IOnlineServicesFactory, FOnlineServicesCommon, IOnlineAccountIdRegistry.

Any ideas?

[queen cave]

so the inventory and crafting is working "okay" but should i be handling it like this is it okay, ps im not worried about cheaters

[fossil spoke]

You may want to try #online-subsystems they might have better directed knowledge.

[dry pebble]

Ah thank you!

[dry pebble]

Nevermind I fixed it, needed this plugin too

[quasi tide]

Been awhile since I've seen you here Meags

[dry pebble]

Oh hi! Yes I was busy working on the website and game launcher hehe

[mystic estuary]

Hello, is there any way to make certain sub-object component properties not to be replicated to certain clients? Something like IsRelevantFor, but per-property basis in an actor component. I see that there's COND_Dynamic and GetReplicatedCustomConditionState(), but I don't see how I would identify what client it's evaluating it for, so that I would discard certain clients from replication of that property

[lament flax]

(idk but im interested for the answer)

[latent heart]

Is there a way to replicate an object to a specific team only? Assuming the team mechanic is already implemented. Something like a function callback for the rep condition?

[tardy fossil]

i know theres a COND_NetGroup and you can control a players net group through their PC.. but i've never used it before so im not 100% sure how it works

[latent heart]

Oh that's handy. Thanks man.

[craggy tusk]

If anyone experienced in replication, I will be grateful if you help me out why the client to server to multicast is not working for the player rotation ?

[tardy fossil]

no problem

[frosty harbor]

Is there any known issues with 5.6 or is it safe to use compared to 5.4/5.5

[lost inlet]

at least a couple of things here. player controllers are only relevant to the owning player, secondly GetPlayerCharacter(0) is just going to refer to the first character available in the world, not a specific player's character

GetPlayerController(0) is pretty redundant when you're in a player controller BP

[craggy tusk]

so what should i use instead?

[tardy fossil]

if you leave it unconnected it will use the player controller it's running the code on

since its in the player controller blueprint

also im pretty sure "Get Hit Result Under Cursor" wont work on the server with a remote player so the first branch will always fail

[velvet current]

why might servertravel work in editor but not in a packaged build?

my guess is that a player controller isn't getting assigned after travel and the level loading because its stuck in the default camera. via a widget that shows players connected, the server sees both players connected but the client does not.

Both players can quit and i can host a new session and the client joins just fine, but when trying to travel to a new level once a session has already begun, player possession only seems to be successful in editor

[cursive mirage]

Q:
In a multiplayer game (assuming 4 players), is it realistically possible to create a moderately large, photorealistic map that performs well (i.e., is lightweight) on a listen server?

For example, something like an urban alleyway with a dense concentration of meshes...

Additional Info:
The GPU is a GTX 1660, and the CPU can be anything as long as it's not a bottleneck.

This image represents one grid, and the idea is to have about 20 of these grids

[dark parcel]

Why are you running client RPC?

the Hotbar inventory is stateful, you don't need to tell client to update via one time event like RPC, you gonna have bugs fairly quickly, especially with people that reconnect.

simply set the value on server and it will replicate to client, if you need client to respond to the changes, do it on the OnRep

[lament flax]

as for 5.5, is {,,UnrealEditor-Engine.dll}::GPlayInEditorContextString still working ?

i keep getting Not In Play World

[lost inlet]

well the context of where you're observing it is important

but it worked last I checked

[lament flax]

well here inside OnRegister/WSS Initialize im no getting it

[queen cave]

i gotta say thank you too you as well man, i was fighting trying to learn with ai and i didn't believe what it was saying about the client RPCs and when i was deleting some of them it wouldn't fix anything well i saw that you were saying the same thing as ai and i figured finally i was definitely doing something wrong, was about to just butcher all the progress i made cause nothing would fix turns out i was just stupid all along and just had to delete ALL my Client RPCs and now it finally works all through the server. Literally spent all my weekend trying to figure out why. so thank you!!!

[exotic wasp]

might be worth looking at replication graph

[latent heart]

With Iris?

[exotic wasp]

oh idk

[finite kernel]

thats exactly what iris filtering - group filtering specifically is for

granted ive only read the docs and never implemented it, since the API changed over versions and dev wiki example is outdated

But maybe someone has working example for 5.5/5.6

[latent heart]

Yeah.

[finite kernel]

On that note im wondering if the filtering can be done on actor or component/subobject level aswell. I thought of some cases where i would like to replicate subobject data holder only to specific connections as part of a globally replicated actor

[latent heart]

Right now, without Iris, I fear the only way to do it is to have a unique object per player and repeat the data for each team on each player.

Which is absolutely awful.

[finite kernel]

Im absolutely rooting for Iris, the legacy system is attrocious for scaling and rep graph is not easy..

[quasi tide]

Root away - it is the future

UE6 and FN depends on it

And FN already switched to it in 5.5 last I read

[lament flax]

im trying to debug why a replicated actor which is spawned on server in a world subsystem isnt spawned on clients once they join

[latent heart]

Is it set to replicate?

Are the components, including the root component, set to replicate?

[lament flax]

yeah

[lament flax]

its a empty actor (AInfo) so i assume no

[quasi tide]

Does it have a root component?

[lament flax]

not on server

oh, hidden actors are not repped yeah ?

[quasi tide]

They are not.

[lament flax]

AInfo hides the actor so maybe thats why

but i tried using AActor and had the same issue

[latent heart]

Yeah. Probably a relevancy issue.

[quasi tide]

That's the other thing I was thinking of. Otherwise, show your setup.

[latent heart]

If it has no root component that may affect relevancy.

[quasi tide]

Oh, didn't see you say that it doesn't have a root component.

[lament flax]

yeah

[quasi tide]

You should always have a root component.

[lament flax]

well i think unhidding the AInfo actor worked

because now i have a crash when the fast array is serializing the received bunch

which means it received the replication from the server

i think its because i used replicated on the container member var

idk, im testing now, getting a crash with not much info

(nvm its not the case)

[fair halo]

Does anyone have any suggestions for a “most correct” way to implement a multiplayer knockback that can be locally predicted?

I’d been hoping to just locally predict a simple Impulse on a Character Movement Controller, but haven’t had much luck so far.

[lament flax]

nope, still no replicated actor

[exotic wasp]

it's getting added to the persistent level right?

[lament flax]

on server its there

i tried using bAlwaysRelevant but still not here

[lament flax]

ive done that (RootComponent = ....), didnt fix it

i compared my actor ctor with classes like gamestate

i dont know why its not working

is it because i spawn the actor inside the world SS ?

[gilded vapor]

You probably want to implement it as a custom movement mode in the CMC or implemented in an FSavedMove so its properly replayed on the server

[exotic wasp]

whats your constructor for the ainfo?

also when is your ss doing the spawn?

[lament flax]

PrimaryActorTick.bCanEverTick = true;
bReplicates = true;
bReplicateUsingRegisteredSubObjectList = true;
bAlwaysRelevant = true;

im debugging with AActor now, not AInfo

[exotic wasp]

same difference

[nocturne vault]

Hello. I'm trying to replicate throwing an item like a bottle. When the input is pressed, it is run on server, spawns the actor and calls the throw event where it adds the impulse. From what I can tell, this should all be on the server, but the client doesn't see the throw

[exotic wasp]

is the throwable actor replicated? is its position replicated?

[nocturne vault]

The throwable actor is set to replicated, but I didn't set replicate movement

[exotic wasp]

if you dont replicate the movement, then the impulse is getting applied only on the server, and the client is none the wiser that its moving

[nocturne vault]

Alright, I'll set it to that. didn't think the movement needed to be replicated if it was run on server. Thought clients would still be able to see it. I also thought I heard in the past setting replicate movement can make movement choppy if there is some latency

[exotic wasp]

itll look choppy even without latency unless you smooth the position

otherwise it will only move every time the network updates its position which might not be frequent enough to look good

you could also multicast the add impulse and not replicate movement at risk of somewhat desyncing the projectile

for clients its basically cosmetic though

[lament flax]

Initialize

[nocturne vault]

I see. that makes sense. Thanks

I got the throw to aim for the same spot for clients. Now for clients, the thrown actor sometimes bounces off a wall instead of running Event Hit, and then on the second bounce runs Event Hit. How is Event Hit handled for the server? Should event hit call a run on server?

I'm trying to make sure the bottle breaks at the same location for everyone

I'm also confused why it sometimes bounces of a wall for the client but not the host if movement is replicated

[exotic wasp]

theres a few ways you can handle it. i would probably have the client hit handle VFX and the server handle gameplay effects like damage or something

[nocturne vault]

I'm guessing I should run on server the hit location and spawn an overlap sphere to see if there is an enemy in range, then multicast the breaking effects/sounds

[exotic wasp]

i dont see why collision should work differently on client and server. if you're replicating movement it might be possible that the position update tick hits right before and right after an impact occurs

[exotic wasp]

you can do that, or have the client handle the effects

[lament flax]

still stuck on this godam actor