multiplayer

[remote tusk]

So there are no instances, 1 Player, playing as client.. Prints two times in Server.. Strange.

If I play it in Server only, wouldnt it supposed to print 1 time?

So simply, When doing in server, it prints 1 time.
When doing in client, it prints 2 times.

Both on server.

As you see, I printed the Owner actor's name, It both prints for same client. And number of players is 1. Playing as clien.

[sinful tree]

Just out of curiosity.... What happens if you put a HasAuthority node (the one for gameplay abilities, not the one with the bool return) and place it before the playmontageandwait node?

Try it with False first and activate the ability from the client.

[wanton sleet]

shouldn't server be responsible of overlapping ?

[remote tusk]

If I start it from client, and not check authority inside the Anim Notify. I only get once.

There is literally no reason to print 2 times on the server, even if i am playing as 1 client... IDK. I already checked for duplicate methods, I only call the method one time for the server. But maybe this Play Montage and Wait is called twice on the server idk...

Here, Ability is called once. So Play Montage and Wait is also called once.. But its Notify is called twice, when triggered in client, and printed double in server.
I also tried 3-4 clients at the same time.. It just prints 2 times..

😦
Where is this SECOND Server log comes from?!! From dedicated server when played in client?! But why 2 times. There are no other players.. And it comes from same client.. Gosh.. I think there is something i dont know about anim notifies or replication..

[sinful tree]

There is definitely something else going on... This is in a Test 4.27 project, but I only get a single print for server and client from my animation notify regardless of who is triggering it. (first print log is server activation, second print log is client2 activation)

And run as client with only 1 player:

Have you restarted the editor recently? 😛

[remote tusk]

Yes. :/(

This is how I use my ability. Already checked that its called once on the server.

[sinful tree]

Why are you activating it like that?

[remote tusk]

Testing Things out. Any recommends? I dont know if this is the problem. It prints one time on the server.

😦

Something really strange going on.. Ah.. I will delete and restart from start i think..

So if i create 4 players. As you see everything works fine, EXCEPT Server has 2 times.. 😄 WTF man.

[left marsh]

I managed to find the problem. The ability in question was being granted only on Possess, which is never called on the client.

[remote tusk]

Btw, I think this is not related to GAS.
As you see, If I play montage on the server. It will still print 2 times on the Notify. Also Validated that Play Montage is called only once.

Would it be related to Anim Graph or smth? Since my Anim State variable is already Replicated

[sinful tree]

Possibly. If you're entering some state that could cause that montage to play that's a possibility.

[remote tusk]

This is my anim montage.

[remote tusk]

My State is only for Upper Body/Lower body, to set idle anim etc. Nothing relevant to that montage actually. Something is really bugged out i think..

[remote tusk]

Does something look wrong in my Anim Notify?

[sinful tree]

Not that I can tell... Animation stuff isn't usually my forte to be honest, but there's nothing that sticks out to me.

[remote tusk]

I think this is related to this stuff.

[thin stratus]

That depends a bit. Here it is better if one handles it everywhere, or at least on Server and OwningClient.

Because due to the Movement Prediction, Server and OwningClient will perform the same movement, and they will both step into the overlap at the same "time".

Means both will apply the LaunchCharacter stuff at the same time and everything stays nicely in sync.

[errant charm]

Is setting replicated variables on event tick expensive? Too expensive?

[dark edge]

no

what is this variable?

[errant charm]

Hit Result from a line trace that runs on tick

[dark edge]

I'm guessing you're maintaining an AimPoint or something of the sort?

You need to remember that the client won't see that thing updating butter smooth though, they'll get it at some varying rate. You'll want to do some smoothing clientside if it's meant to be shown

[errant charm]

Yeah

[dark edge]

If it's meant to be seen by the owning client it'll be lagging behind their view. What actual mechanic are you making?

[errant charm]

Wondering if there's a better approach to it tho, it's just that I cannot find a way to get linetraces running properly on server RPCs for on clients

[dark edge]

What are you trying to do?

[errant charm]

I want the item to smoothly fall to the floor using a timeline. Before the timeline starts, I do a line trace is to the hit location. This trace is called from a Server RPC, and if it hits, the hit location is saved in an EndLocation variable which is replicated. Then I call the "PlayDropEffects" multicast which lerps the actor to that EndLocation using a timeline for a smooth drop effect

The issue is, when the server tries to perform the line trace on the client's character (character is passed into the interaction logic), the trace fails completely. Wondering if it's because I'm running the trace on the camera but I don't see why that would be

this logic is not being ran in the player, but is ran in the interact parent class, which i call in the children of it. i do it this way for a more modular system

[dark edge]

are all dropped items dropped by some player?

[errant charm]

yeah

[dark edge]

You can do it like:
Client:
DropButton -> trace -> call DropItemtoLocation(Item, Location) //run on server event
Server:
Check that Location isn't like miles away from the character's location -> get item, call MulticastDrop(Location)
Everyone:
Item plays timeline, lerps location towards the target location.

the server doesn't need to do any traces

although you'll have the be careful to gracefully handle the item state, does it not have replicate movement turned on? What would a late joiner see?

[errant charm]

yeah that makes more sense, thanks for the help!
replicated movement makes it very jittery, doesn't seem to work well with timelines, so i turned it off.

[dark edge]

you probably want to turn it on when the timeline ends

that way a late joiner sees the replicated position of the thing, that is, it lying on the ground.

[errant charm]

ohhh yeah, damn, didnt even consider that at all

thanks a bunch

[sweet ore]

can someone tell me why i still can use cheat engine and change the value of x?
i used rpc run on server to multi cast and made the server change the value

[left marsh]

Does BeginPlay only get run on the Server? I am trying to figure out why the client is not running a necessary function that's under begin play.

[sweet ore]

btw i used the client net mode with running a dedicated server on so i'm not playing as a server and i still can change the value on client

[severe widget]

Just because the server is the source of the initial value, that doesn’t mean the client isn’t storing that value in memory.

Also, if you’re testing PIE, you may have both the server and clients in the same process

Detection of CheatEngine or similar things is pretty complex

[sweet ore]

stand alone game - i have tested changing value of only server values variables in the game mode and i could not, just to make sure i don't have any auth to server

[severe widget]

You might’ve locked the value of the float variable that is the input property of the event?

So even if the value changes from replication the client could just use its own value

[sinful tree]

Just because you're using server authority to send the value, your server does not ultimately get full control over client machines. You can tell them what to do, but they don't have to do it.

[dark parcel]

Server RPC = client telling server to run function in the actor in server's version.

Essentially you just ask the client for a value without any check.

So ofc a client can just run a cheat engine and just tell server hey, run this function with X that i passed which is 999999

General rule in multiplayer is to trust the client as least as possible.

And as for movement, you normally only grab the input from the player.

You dont ask. Hey what's your position in your world? Instead you just ask which input direction is the player currently holding. Then server can resolve the direction and location the player is going simply from the direction input the client send.

[sturdy sand]

Should I worry about this? It used to not be a problem

[nova wasp]

from reading the code this might actually cause some weird issues because it guesses the remote role early
doesn't seem that hard to fix... just do what it says?

[sturdy sand]

Well the actor doesn't have the option to set replicate in the properties though?

And yes right now I'm having some weird replication bugs like the server seeing different values from the client

[nova wasp]

Read the error message... this is not a property window problem

this is about something calling SetReplicates before the actor is initialized instead of setting it directly

[sturdy sand]

ok

[nova wasp]

just breakpoint the error log, no need to guess

Not sure if fixing this will help with your recent issues but warning logs are not sensible to ignore here

[sturdy sand]

I'll figure it out. I was going to ignore it at first because warnings are not usually a problem for me.

Is this related to the issue? I tried to cook the content

Got many of these

[nova wasp]

what would an input action have to do with replication issues?

I believe in most games input is local only

[sturdy sand]

no it's not only that as I said. There's multiple of those warnings for every asset

Well I guess that is true though. It's not related with the replication errors.

[remote tusk]

What might be the reason an Anim Notify called twice on the Server, when playing as Client as 1 player.
Here the result is;
Playing Montage prints 1 time.
But Owner Name printed 2 times in the server.

"Play Montage" prints as 1 time, validating that input is called only once, and the montage.

[dark parcel]

@remote tusk are you aware triggered runs every frame?

use started if you want to fire something on Press

[tardy fossil]

if you hook up a print to the Notify Begin exec pin, does that also print twice?

[nova wasp]

only UAnimNotifyState has begin

this is probably a single UAnimNotify (where it's just a single event, not a range of time that begins and ends)

I would try to see what the callstack looks like for each invocation in the actual code

afaik montages can get weird with dead instances calling again depending on ticking rules on the server

[remote tusk]

My trigger is set to "Pressed" only in the input, so as you see "Play Montage" log prints only once.

[remote tusk]

Yes. Its really weird behaviour, I couldnt solve it since the last night..

I validated that this Play Montage nodes work only once. But Anim notify gets twice.
And this is only happening on "Client." Server is printing only once when played on server instance.

If i start game as client,and only one player.
It prints 2 on the server.. Regardless of "Client" count..

Am I doing smth. wrong by using Anim Notify inside "Montage" and not in Sequence?

[dark parcel]

I use anim notify and anim notify state all the time inside montage with no issue.

[remote tusk]

Now I will try something. I will use simple Anim Montage Notify, and subscribe to On Notify Begin see how it looks like.

[dark parcel]

OnNotifyBegin is for AnimNotifyState

just use the regular AnimNotify

it's working as intended

[remote tusk]

Yes, I am already using Anim Notify on the problem.

😦

[dark parcel]

the problem doesn't lies from using the AnimNotify vs the AnimNotify you create in the montage imo.

[remote tusk]

I meant on the Play Montage node

Let me try this now, lets see whats going on. 😦

It only prints one, when used Anim Notify, and use On Notify Begin node.

Well there is no custom notify for that. I am testing this if smth. wrong with my GAS setup, but nah. It also didnt work for Play Montage node with "Custom Anim Notify that i previous earlier.

I use upper body/lower body to seperate.
I also tried to log "Mesh Component" name if there is multiple meshes that playing the anim.. 😦

Its weird that it only happens on client.

I can solve that with "Trigger Once" in Gas when waiting for event.. But still, Anim Notify triggers twice makes me angry and trying to solve it..

As you see its a simple Anim Notify, called on a Montage..

[tardy fossil]

try changing your Montage Tick Type to branching point as a quick test

[remote tusk]

Expected behaviour was Client and Server only, when used GAS. But server triggers twice IDK.. I think smth. wrong with my animations.

I thought problem was with GAS. But I also tried Play Montage node on server only.

[remote tusk]

Alright

LMAO

this worked?

But how? Why... MY WHOLE 6 hours was just for a montage tick type?! 😄

[remote tusk]

Thank you very much man.. Thank you. But How this is worked, that was wrong

[tardy fossil]

maybe some weirdness with the fact a dedicated server ticks slower by default or something

branching point is more precise

[remote tusk]

Hmm IDK. This also happened on P2P btw. 2 players, one is listen, one is client

Thanks man anyway.. I will search into on that!

[tardy fossil]

no prob

[remote tusk]

Are there any chance that my Montage is in Loop or smth? Its not looped on the settings though.. Anyway.. This was the problem.

Should i be "best practicising" to make branch point for every montage i will replicate?!

[dark parcel]

There's no p2p but aside from that mine doesn't get called twice on Queued

not sure if there would be a problem in the future though

[remote tusk]

Maybe because its some kind of animation replication stuff on going behind IDK. 😦

I'm not totally relieved that it works with only changing Tick type. 😂

Should i be using "Anim Notify" to check Traces for harvestables? Or are there any better ways

I simply Play Montage --> On Notify Hit --> I trace and send data to server to attack.

I cant even now if this is true or not.. 😄

[glossy wigeon]

I'm currently implementing some Server functions, now I wonder, since for example I have a EquipItem function. What are common patterns approaches here?

Currently I'm creating a ServerEquipItem function (which is annotated with Server), and that function then calls the EquipItem function which is just blueprint BlueprintAuthorityOnly).
I wonder, does it make sense to just have the EquipItem function be a Server function in the first place?
Also wonder should I maybe add a Authority check in the EquipItemFunction and then call the server function from there (and remove the BlueprintAuthorityOnly annotation), and then on the server the EquipItem function is called via the server call again.

Is there some common patterns/approaches that make sense?

[remote tusk]

F... you ChatGPT, some guy called Omnicide is better than you!

[remote tusk]

In Server -> Equip Item (i assume you will spawn the item) -> Change your Equipped State enum which is replicated (create it) to sync for animation.
Then you can attach your item, which resulting others will see item

Here is how i use hotbar, and spawn an equippable item.

The picture called on server.

[glossy wigeon]

It already works and is implement, quite decent I think, it's just a conceptual question.
What are the best practices when it comes to things like that.

[glossy wigeon]

Also I'm in C++, but currently I see a few options:

Option 1
Implement the EquipItem function, so that it checks for authority and if not automatically call the Server function.

Option 2
Just have EquipItem and ServerEquipItem separate and ServerEquipItem calls EquipItem (like in option 1), but EquipItem does not call the Server automatically.

Option 3
Make the EquipItem function be a Server function in the first place.

In option 1 and option 3 you basically can call EquipItem everywhere, possibly avoiding mistakes, where you oversee authority checks, having the EquipItem function deal with when to call on server when not itself. While option 3 might be the simplest/cleanest, you won't be able to have custom logic that is run on the client immediately, when equipping an item. Option 3 would require to call the appropriate server function dependent on the context, but might avoid a additional a additional HasAuthority check.
I feel like option 2 is just the thing one gravitates to while implementing first, while option 1 and option 3 might be better to use implementation wise, where option 1 might be the the best?

Besides that, the question is, do I maybe extract the logic for equipping the item into it's own function, separating it from the client/server calls completely, but that might be a bit extreme and only if one really needs it.

So basically I'm just asking, if there is some common patterns, of how to implement this calls, or if you just take it how it comes, if someone knows any pitfalls here which I may regret later on.

[torpid crest]

Hey guys, I’m having a weird issue with server travel via Steam.

I have both the listen server and a client in the lobby. When the server initiates a server travel, the client joins the session but doesn't have an assigned player controller.

However, if the client closes the game and rejoins the session, they do get a controller 😄 Any idea what might be causing this?

[twin vessel]

While using server side rewind to confirm hits, how do you choose which actors to rewind?

Only rewinding the target the client suggested, feels like a bad idea, because a teammate might have moved in the way blocking the shot, but when client took the shot, the teammate was not there

[pine flicker]

Anyone know how to replicate FloatingPawnMovement? I'm using it in my vehicle, when the Server enters the vehicle and increases the Throttle (Float), that Throttle variable is being sent into a AddMovementInput and it looks fine for both the client and the Server. However when the Client enters the vehicle (either pilot or passenger seat) the Vehicle jitters for the Client but not the server. I assume it's just an incorrect replication setup but everything I've tried so far has had the same result (still jittering). Not sure if the FloatingPawnMovement requires additional setup. I've replcated all related variables and components.

[nocturne quail]

what this log want to be fixed?

Warning: (navmesh: RecastNavMesh_UAID_544BFE8D4231BC5B08-Default) Added an oversized dirty area | Tiles marked:  2 | Source object = SphereComponent /Memory/1DYT2RL8VAWC8N7U98M2Q9ACC.Playmap:PersistentLevel.BP_Bullet_C_281.BulletSphere | Potential comp owner = BP_Bullet_C /Memory/1DYT2RL8VAWC8N7U98M2Q9ACC.Playmap:PersistentLevel.BP_Bullet_C_281 | Bounds size = X=190.400 Y=190.400 | Threshold: -1

[nocturne quail]

setting bAllowJoinInProgress = false will actually not allow any new player to join?
i call this function when max player limit is reached.

void AARMAGameStateBase::CloseJoiningAbility()
{
    //Restrict player join if match is already started or max player join limit reach
    if (const UARMAGameInstanceBase* GameInstance = Cast<UARMAGameInstanceBase>(GetGameInstance()))
    {
        IOnlineSessionPtr SessionInterface = GameInstance->GetSessionInterface();
        if (SessionInterface.IsValid())
        {
            FNamedOnlineSession* Session = SessionInterface->GetNamedSession(NAME_GameSession);
            if (Session)
            {
                if (Session->SessionSettings.bAllowJoinInProgress == true)
                {
                    Session->SessionSettings.bAllowJoinInProgress = false;
                    SessionInterface->UpdateSession(NAME_GameSession, Session->SessionSettings, true);
                }
            }
        }
    }
}

[limber gyro]

never tried this, but if you want to stop players from joining you could probably destroy their controller on "initNewPlayer()" in the game mode(i havent tested this), theres probably something better that you can do on "Login()" or "PreLogin()" in the GameMode.

[nocturne quail]

Thanks i will try them

[compact flame]

I have an event on game mode. I also have a custom notifacation function that creates notifacation for the player controller I input I want the notifacation to show up after the event on every client not just the server how can i go about doing this?

[nocturne quail]

this works perfectly

void AARMAGameModeBase::PostLogin(APlayerController* NewPlayer)
{
    Super::PostLogin(NewPlayer);

    if (!NewPlayer)
        return;

    if (CurrentJoinedPlayers >= 2)
    {
        UE_LOG(LogTemp, Warning, TEXT("Logout :: Reason :: Steam Session is Full on a private app ID, try later to join!"));
        KickPlayer(NewPlayer, TEXT("Session Full"));
        return;
    }

    AARMAGameStateBase* CurrentGameState = Cast<AARMAGameStateBase>(GameState);

    if (!CurrentGameState || CurrentGameState->GetIsMatchStarted())
    {
        UE_LOG(LogTemp, Warning, TEXT("Logout :: Reason :: You can't join an ongoing match. Try later!"));
        KickPlayer(NewPlayer, TEXT("Match Already Started"));
        return;
    }

    CurrentJoinedPlayers++;
    CurrentGameState->Execute_NotifyPlayerJoined(CurrentGameState, NewPlayer);
}

void AARMAGameModeBase::KickPlayer(APlayerController* PlayerController, const FString& Reason)
{
    if (!PlayerController) return;

    FString KickURL = FString::Printf(TEXT("?KickReason=%s"), *Reason);
    PlayerController->ClientTravel(KickURL, ETravelType::TRAVEL_Absolute);
}

[limber gyro]

how did u figured this line "PlayerController->ClientTravel(KickURL, ETravelType::TRAVEL_Absolute);" would kick the player?

[nocturne quail]

yes i tried it and it works 😄

it travels me to the default editor map

[limber gyro]

also, you might wanna move this var "CurrentJoinedPlayers++;" to the game state, i did the same in my game and eventualy needed to move it

[nocturne quail]

in the gamestate i have one which is replicated, but in game mode i want a non replicated

replicated one to use for updating player count on UI

[limber gyro]

but why would you have 2?

you can just use the one from the gameState

just fetch it in the game mode

[nocturne quail]

yeah but i don't want to access gamestate for this property check

[limber gyro]

im sure you have your logic for not acessing it, but regardless its super easy if you ever wanna change it for simplicity GetGameState<AArenaGameState>()->hasPlayersToStart = true;

thats how i do it in my game mode

[nocturne quail]

Thanks for sharing the snippet 😄

[sinful tree]

You'd probably want to multicast it on the GameState then have whatever notification is received on clients from there.

[compact flame]

Ohhhh yeah huh I forgot the game state existed thanks

[dark edge]

I'd have ServerEquipItem which actually sets the state, then use a rep notify to do whatever the fact that the item is equipped should do

[wanton sleet]

void ACannonBall::Launch()
{
    MultiCast_Launch_Implementation();
}

void ACannonBall::MultiCast_Launch_Implementation()
{
    UE_LOG(LogTemp, Warning, TEXT("Launch(%d), NetRole(%d)"), HasAuthority(), GetLocalRole())

    ProjectileMovementComponent->Velocity = -GetActorForwardVector() * ProjectileSpeed;
    ProjectileMovementComponent->bSimulationEnabled = true;
}

I call ACannonBall::Launch() from another class on Server. This code works on server side, however on client bSimulationEnabled would not be updated. How can I solve the issue ?

[wanton sleet]

thank you for your answer, it is solved

[thin stratus]

Are you sure the Actor is replicated?

[wanton sleet]

yes, actor is replicated but projectile movement component is not. Should I set the component to replicate ?

It always log Launch(1), NetRole(3), so it is not called on any non-auth actor

[twin vessel]

The function is only called on authority cause you are calling MultiCast_Launch_Implementation, so you are just running the function, not calling the rpc.
Just call MultiCast_Launch

[wanton sleet]

ohhh, thats valid point. I am checking right now. Thank you

that solve the issue. thank you for your time

[final bluff]

Hey! Could anybody point me in the right direction on how to create/join lobby WHILE remaining on the same level .

I can create a session and then WITHOUT changing levels , I can turn the current level into a listen server, which then enables players to join me and that works perfectly fine. BUT when a client uses the join session node , it changes the level to whatever the server level is. If my client is already on that level prior to joining , how can I get a client to join the session without reloading the level?

I’m using blueprints but can add any cpp if the functionality is exclusive to cpp

[thin stratus]

I'm a bit surprised you can start listening without re-opening the level. In theory, both, listening and connecting (as well as disconnecting) are HardTravels and will load the level. I don't think connecting can be done without the client loading the level

In standard UE that is

There are c++ beacons but they aren't meant for gameplay

[final bluff]

Thanks for the reply!

Took me a lil bit to figure out lol , but I’ve been using this UGameInstance::EnableListenServer . And this allows me to change to listen server during runtime . And so far it’s been working fine without needing to load to another level . I’m not sure if there’s something I’m missing but its working how I figure it would and how I want it to lol

But yah I’ve been trying to find any documentation or forums on how to join without level change but coming up empty like your suggesting

I have seen/played some indie games (unity) where they create and join lobbies while still remaining on the same level ( or atleast as a player it looks like no new level was loaded ) and I’m curious how this is done

I’m ultimately trying to create a player lobby screen where users can host and join a lobby while still having that lobby screen stay Persistent (akin to Fortnite , COD) . Maybe this is done via level streaming?

[nocturne quail]

using the latest tool chain, i'm getting this error

error : non-portable path to file '"WeaponActor.h"'; specified path differs in case from file name on disk [-Werror,-Wnonportable-include-path]

this only happen if i build for linux

[lost inlet]

it tells you what's going on

and I remember this being an error with cross-compiling even in UE4

[nocturne quail]

I used UE4 for this before and don't have any errors.
I know it is telling me the file name not matches which is incorrect

[lost inlet]

it tells you the casing doesn't match

[nocturne quail]

because the file name is correct and for windows it build correctly

[lost inlet]

windows is not case sensitive, linux is

[nocturne quail]

the file name WeaponActor.h in the folder directory

same as in the header file

the folder name is Weapon/WeaponActor.h

this is the issue?

[thin stratus]

I would be worried that some actors don't have the right net roles and similar if you don't reload the map

[thin stratus]

That's done via Beacons

Lightweight connection that supports replication but doesn't actively load the map

Check #online-subsystems pins. There should be some info. Needs C++ or course.

[final bluff]

Awesome , will look into beacons and this thread . Ty 🤙

[pale yew]

Hello all,

I need to test whether a host of a listen server is connected to the internet during play. We have a potential issue where the host could unplug their internet during play and get a bunch of kills and a rank up once everyone else disconnects, so we want to terminate the game session.

The problem is, I'm unsure how to get whether the listen server host is able to reach the internet. Is there something I can poll to test this?

[nocturne quail]

solved:
changing the disk drive name from Local Disk to Localdisk

no idea why this is case sensitive issue

[winter pollen]

im going nuts. My boolean returns the proper value when playing as client only. if i set it to replicated then it stops working on client aswell. So right now, it only works when playing as client, not standalone

[dark parcel]

Hmm I can't figure out how to stop correction for this case.
So the server disable the capsule collision to a pawn channel. But the client won't do the same instantly as ther is delay.
There will be a buffer where the collision is disabled on server but enabled on client.

This will cause correction when the client tries to walk over the dead body.
Any idea on how to tackle this?

[dire burrow]

why isnt this rep notify working in the animation bp?

[dark parcel]

anim blueprint is not replicated

Anim blueprint job is to read from it's owner.

never do any game logic in anim blueprint.

[dire burrow]

yeh this is why i am switch over to using a interface to update the animbp

[ancient adder]

Does the server request the client to disable it or the client does it predictively?

[dire burrow]

https://youtu.be/R3mPtalY4o4?feature=shared&t=251 ( i was following this tuto)

[dark parcel]

Death happend on server, so the server sets the collision first on it's machine.
Then tell client to do the same via repnotify.

[dark parcel]

interface or not has nothing to do with the role of Anim bp and how it is not replicated.

[ancient adder]

The repnotify is for death and collision disabling?

[dark parcel]

just for the collision

[dire burrow]

Well i am not doing any game logic in the animBP .. part of the reason i switched to a interface is also to give the code something to replicate because it was at the end of the graph... it was orginally just reading it in from a refference to bp that held it .. but this would give you rep notify node that could do nothing except set the value again which becomes a loop ..

[dark parcel]

// Allow for being pushed for a moment.
    if (HasAuthority())
    {
        Mobility = EAGMobility::CanBePushed;

        FTimerDelegate TimerCallback;
        TimerCallback.BindWeakLambda(this, [=]()->void
        {
            // Remove collision with other pawn after a short while.
            Mobility = EAGMobility::IgnorePawn;
            GetCapsuleComponent()->SetCollisionResponseToChannel(ECollisionChannel::ECC_Pawn, ECollisionResponse::ECR_Ignore);
        });

        FTimerHandle Handle;
        GetWorld()->GetTimerManager().SetTimer(Handle, TimerCallback, FMath::RandRange(0.2f,1.5f), false);
    }
    

Then the rep notify for client

void AAGCharacterBase::OnRep_Mobility(EAGMobility OldMobility)
{
    if (Mobility != OldMobility)
    {
        switch (Mobility)
        {
        case EAGMobility::NotMoveAble:
            break;
        case EAGMobility::CanBePushed:
            break;
        case EAGMobility::IgnorePawn:
            GetCapsuleComponent()->SetCollisionResponseToChannel(ECollisionChannel::ECC_Pawn, ECollisionResponse::ECR_Ignore);
        }
    }
}

[ancient adder]

Assuming death is also replicated, Why not disable the collision on client aswell when the actor dies?

[dire burrow]

but in any event the kneck rotation (modify bone transfor) isnt and wasnt replicated on clients in either scenario .. just puzzled as to why this is then if it im not doing any thing wrong

[dark parcel]

death is only called on server

since damage is server only

if damage fatal -> Death is called

then server do what it has to do. Then server will replicate w/e client need (e.g capsule collision)

actually with 50 ms, seems allright

corrections are more evident in 100 ms environment

[dark parcel]

Nvm still getting correction, especially when the player sprint.

[blazing bear]

Does the server still have to replicate the dead character? If not then you can just disable replication so it doesn't keep correcting

[dark parcel]

the correction happend because in the server, the player can walk over the dead boy where
the client have some window before it disable the collision in the machine.

[blazing bear]

yeah but the server shouldn't correct the client if replication is disabled no?

[dark parcel]

so correction will happend, lets not focus on the death part but the fact that there will be disagreement in a certain window on the capsule collition.

[dark parcel]

The one that being corrected is the player that is still alive.

[blazing bear]

right my bad

[dark parcel]

nah, all suggestions are welcomed.

thanks

[blazing bear]

I thought it was correcting the dead body

all good

[dark parcel]

@blazing bear actually gave me an idea with disabling the replication

if I set it to false, will the client assert Authority?

[grand kestrel]

I think the video is grossly mistitled, had a quick look

[dark parcel]

Yeah, I saw the video. The title doesn't do justice.

he is setting the variable in the character

abp just read that replicatied variable.

deffinitly a very bad title.

[grand kestrel]

Yeah lol

[dark parcel]

Vaei, got a question. If I turn off replication, does the client will assert authority over the not replicated anymore actor?

[grand kestrel]

If you tear off the actor then it won't get any updates from the authority (server)
But I haven't checked if it then becomes authority and returns true for auth checks

[blazing bear]

What about using bIgnoreClientMovementErrorChecksAndCorrection

/**
* If true, we should ignore server location difference checks for client error on this movement component.
* This can be useful when character is moving at extreme speeds for a duration and you need it to look
* smooth on clients without the server correcting the client. Make sure to disable when done, as this would
* break this character's server-client movement correction.
* @see bServerAcceptClientAuthoritativePosition, ServerCheckClientError()
*/

[dire burrow]

i think what your kind of missing is if you SET a variable at the end of the graph there is nothing else you can do... this is why i brought up that i switched the system so that it could have something to replicate, eg the final step now is set the value inside the animation BP using the interface and thats the bit of code to replicate

any way i figured out what was causing the problem I hadnt ticked replicates 🫣

[dark parcel]

Yeah but I can't toogle that on the player that is alive as I need authorative movement.
Trying to see if I can push things simply just checking overlap on tick now as opposed to Capsule Hit event.

cheers guys.

[grand kestrel]

What is the original issue you're trying to solve? I read up but there's a lot of discussion @dark parcel

[blazing bear]

it only prevents the server from checking the clients for errors, have you tried it?

[grand kestrel]

It prevents the server sending any information whatsoever as well, not just positional either; everything

[blazing bear]

well basically ServerCheckClientError doesn't run if the property is true

maybe have the client predict the enemy's death

otherwise

[dark parcel]

Im pushing the capsule component for a few moment before disabling the collision.

^ Above is single player listen server, I haven't got time to take a video of the problem in multiplayer.
But hopefully it can be understood.

right now, just gonna try to push without the Hit event and just check for overlap on tick on server.

that may do the trick 🤷‍♂️

[blazing bear]

having NetworkMaxSmoothUpdateDistance set to a higher value may make the corrections seem less bad if they're actually tping in your case

[dark parcel]

does that apply to Autonomous or just proxies?

i think any sort of smoothing only apply to Proxies

actor controlled by player got snap correction.

[blazing bear]

should apply to autonomous as well

the locally controlled player can slowly but smoothly interp to the correct pos

ARK for example does it in some cases, the cost is that the local player feels sluggish and slow during that moment

[dark parcel]

no doubt, but I am not sure if there's something in built for autonomous to be smoothly interp.

with my skill I can't fight CMC yet.

[grand kestrel]

Damage and death are server events
Whatever they apply should be updating along with the sim proxy, i.e. without any delay
For the local player themselves, when they get damaged/killed, because its from a server event any knockback, movement speed/etc. modification, stun, disabling collision, etc. will result in a de-sync
I still don't quite clearly understand the specifics of your issue but does it fall into that area?

[dark parcel]

yea, to be specific the part to disable the collision.

server disable collision then tell the client to do the same.

[grand kestrel]

Presumably you also stop movement?

[dark parcel]

but that will create a window where server already have the collision disabled where as the client still have the collision enabled.

I stop the movement on the dead character.

[grand kestrel]

Right, yeah, and it's also the same problem with walking into another character in MP, whether its AI or player controlled, you de-sync like crazy because they're in different places on the server, even when the local clients don't see themselves touching

[dark parcel]

Yup 😦

[grand kestrel]

Coz we don't predict other characters than our own

Which is why I made PushPawn 😛

[dark parcel]

So right now, just disabling the collision without pushing is fine.
With that in mind, I can just apply the push without depending on the capsule collision but an overlap event on tick in server only.

So in the end, I think this could work since server alone will dictate where the dead body get pushed.

im not good at explaining so this may sound confusing.

[grand kestrel]

Yeah I'm not 100% following with your specific use-case

I do have a solution for granting partial client authority solely for location, its useful for applying snare, knockback, stun, etc. without any de-sync (up to a point)

But if its affecting the local client colliding with simulated proxies when the server has already disabled their collision causing a correction it might not help you, because generally you wouldn't grant client auth just for killing someone, but you could

[dark parcel]

@grand kestrel would love to go through some of your github stuff one day. Trying to accumulate the basic to understand this monsterous CMC atm.

[grand kestrel]

Its in the PredictedMovement repo, you can search the CMC on there for ClientAuth and copy those parts out

[thin stratus]

Not to sidetrack your question, but why is there a concern for the host ranking up quickly? If you use listen servers, then cheating should be expected anyway.

Your host can just kick everyone or kill them without pulling the Internet after all.

[pale yew]

I just have a job to do and the question was asked of me, unfortunately.

[thin stratus]

Right

The person who asked should probably be informed that using a listen server will allow the host to do whatever they want anyway.

[pale yew]

They have been informed

[thin stratus]

Pulling the cord is the smallest issue. The project should use dedicated servers. I will see, once I'm awake and walked the dog, if there is a callback or sorts for the disconnect.

[pale yew]

Thank you. And yes I understand we should be using dedicated servers

[thin stratus]

Must be something on platform level as there is nothing internet wise that would check any connection

[thin stratus]

@pale yew Haven't found anything yet, but you could add this functionality yourself, as long as you have some webserver that you could create a SocketConnection to fwiw.

#include "Sockets.h"
#include "SocketSubsystem.h"
#include "Interfaces/IPv4/IPv4Address.h"

void USomeTickableWorldSubsystem::Initialize(FSubsystemCollectionBase& Collection)
{
    ISocketSubsystem* const SocketSubsystem = ISocketSubsystem::Get(PLATFORM_SOCKETSUBSYSTEM);
    if (SocketSubsystem == nullptr)
    {
        /// Might want to actively handle this cacse.
        return;
    }
    
    /// You can define this in some Developer Settings for example.
    const FString Address = TEXT("123.456.789.012");
    constexpr int32 Port = 8912;

    /// Create new FSocket.
    Socket = SocketSubsystem->CreateSocket(/** SocketType = */ NAME_Stream, /** Description = */ TEXT("Internet Connection Tester"));
    
    /// Parse the Address, this can be useful to ensure the IP isn't malformed.
    FIPv4Address IP;
    if (!FIPv4Address::Parse(Address, IP))
    {
        /// Might want to actively handle this cacse.
        return;
    }
    
    /// Create a "proper" InternetAddress, needed for the FSocket::Connect function.
    const TSharedRef<FInternetAddr> InternetAddress = SocketSubsystem->CreateInternetAddr();
    /// Fill in IP and Port.
    InternetAddress->SetIp(IP.Value);
    InternetAddress->SetPort(Port);
        
    /// Try to connect. Return value tells you if it worked or not.
    const bool bConnected = Socket->Connect(*InternetAddress);
}

void USomeTickableWorldSubsystem::Tick(float DeltaTime)
{
    /// Check if the Socket is valid and the connection state is correct.
    /// Not sure if you want to count the Socket being invalid as not being connect.
    if (Socket != nullptr && Socket->GetConnectionState() != SCS_Connected)
    {
        /// Handle not being connected.
    }
}

The FSocket can provide you with information if the connection is active or not iirc.

[winter pollen]

do macros work differently when networked / not? i just pulled some logic out of my animation handler and it started working. But it wouldnt when i called the whole macro. Weird asf

[thin stratus]

Blueprint Macros?

[winter pollen]

Yes

[thin stratus]

Pretty sure Blueprint Macros are just wrappers for code to make it re-useable of sorts. Bit more fancy than Functions and Events, due to being able to have Latent nodes and multiple Exec pins in them, but lacking features like being able to be marked as an RPC.

So in other words, they shouldn't have anything to do with networking.

[winter pollen]

hmm..good to know. it must have failed somewhere then which i wil have to debug..

which will be intresting as you cannot put breakpoints inside macros afak 💀

[thin stratus]

Yeah I never really use Macros, like at all, despite the built in ones.

[pale yew]

Cheers, thanks

[thin stratus]

I just had a look at UNetDriver, which would be the closest to what handles all the connections.
The biggest thing to take away is that the Server not being able to "connect" to the Internet isn't really UE's problem I guess.

Do you know what happens if you play as a ListenServer and you disconnect the internet? Does it put the Server back into the MainMenu after a while?

[pale yew]

No the server can continue. It's the client that gets punted back to the main menu.

At least as far as I could tell.

[pale yew]

Might there be a way to poll EOS in some way?

[queen mortar]

Hey I'm currently making stamina logic in blueprint. With my current setup, when player presses sprint, server rpc is called that runs a timeline and from timeline update, sets Stamina(float RepNotify) variable. Inside OnRep function, I set the progress bar percent inside player widget. I know this setup may not be ideal but is it very bad? If yes, then what solution should I use?

[blazing bear]

is your sprinting done in bp as well? Does it lag if you enable network emulation?

[queen mortar]

No it's in c++

[blazing bear]

Ideally you better update the UI by having the HUD class or the widget itself bind to an "Update Stamina Bar" delegate that is executed whenever stamina is changed

[blazing bear]

You should then calculate stamina inside the PerformMovement function in the CMC

[queen mortar]

Well I'm using timeline only because I want the progress bar to progress smoothly.

[blazing bear]

Oh I thought you were using it to update the stamina itself

Stamina is best calculated in the CMC, while the UI is like I said, using a delegate

[queen mortar]

So with what you suggest will the progress bar UI able to move smoothly?

[blazing bear]

having stamina done in the cmc will allow for client side prediction and server reconciliation in case of lag

[blazing bear]

Yes, given that stamina is set on tick when sprinting

[queen mortar]

I also tested it with network emulation and it does not cause any lag. I mean it would be ideal if it would be possible to do it just in blueprints because I don't have much experience in c++

[blazing bear]

Basically PerformMovement is a ticking function, you calculate stamina inside it, and if stamina has changed, then send a delegate

[blazing bear]

you're talking about updating the actual stamina or the progress bar?

[queen mortar]

Both would be ideal because player could only for example sprint when stamina is greater than 0.

[blazing bear]

what's wrong with the player only able to sprint when stamina is greater than 0? what do you want it to be like?

[queen mortar]

No I meant that player Should only be able to sprint when stamina is greater than 0. Currently I have stamina float that updates the progress bar in UI.

[blazing bear]

oh so you don't have a stamina system currently? just UI?

[queen mortar]

Here is the setup I have.

[blazing bear]

where is Drain Stamina called?

[queen mortar]

When player starts sprinting, a timer is set on the server that runs this function each 0.1 seconds

[blazing bear]

as for having the player only sprint when stamina is greater than 0, that's obvious how, it'll be just a branch

not sure how it'd be done in BP honestly... is there nothing about a networked stamina system online?

[queen mortar]

The setup I had before had a timeline that updates the stamina variable on each timeline update. I was worried that that would be sending too many informations over the network in a short time, thats why I switched to timer but now the progress bar isnt smooth since I'm directly setting the value.

[blazing bear]

what would be ideal is to have both the client and server set stamina inside a ticking function, and if there is a mismatch between client and server, then fix it

if the client sets stamina on its own, without an rpc, then updating the stamina UI becomes easy

[queen mortar]

But shouldn't the stamina be set on the server though?

[blazing bear]

Well it's not smooth because the client is waiting for Stamina to be replicated every single time, to update the bar

[queen mortar]

It's not really about mismatch. It's not smooth because in Drain stamina function, I subtract 0.01 from stamina. If I wanted it to be smoother, I should subtract less, but then the draining process would take longer. In that case it's best to use some sort of interpolation but that would end up setting the variable and calling on rep function every frame or so...

[blazing bear]

I mean, I'm speaking Ideally, which is doing it inside a custom CMC class, thus no, Stamina isn't a replicated variable, it's updated both in the server and client at the same time, but when there is a mismatch, the server is favored because they're the authority

[nimble onyx]

Does anybody know how i would make a ragdoll replicate? ragdoll works on owning client but does not seem to update on the others

[quasi tide]

You're doing a client rpc. That will only go to the owning client. Use an onrep variable and inside the onrep, do your death logic.

[nimble onyx]

its multicast

but ill try the onRep

[quasi tide]

That isn't multicast

[nimble onyx]

wtf unreal rolling back my things thank you

didnt event notice that

[quasi tide]

Multicast would still be incorrect anyway.

[nimble onyx]

still alot better xD

[quasi tide]

Unless you're destroying the actor after some time from dying I guess.

[nimble onyx]

thats handled in my respawn event

[quasi tide]

Then multicast would be fine depending on how your game plays out.

[nimble onyx]

would there be a way to make him not stiff as a board ?

[quasi tide]

That depends on your settings and all that. Not really multiplayer related.

[nimble onyx]

ok

[torpid crest]

is it possible to use world partition in multiplayer?

I have a problem where, when the client teleports to another part of the map and then returns, all actors with physics have fallen through the landscape. My guess is that it's because that part of the landscape was unloaded

[blazing bear]

check the physics asset, there you can configure how the mesh will ragdoll

[nimble onyx]

it works perfeclty on the owning client just not for the others

sorry if i pinged

[blazing bear]

you're not replicating ragdoll physics right? is the Replicate Movement property enabled?

[nimble onyx]

yes

[blazing bear]

each machine should set simulate physics to true

[nimble onyx]

ill check real quick

[blazing bear]

also disable any sort of collision on the dead actor

and set the collision profile to ragdoll

simulate physics to true should be done after disabling collision just in case

[nimble onyx]

where would i find these

machine as in the abp

[blazing bear]

I mean machine as in the players

you need to do this stuff on a repnotify

[nimble onyx]

Ok I Got it Thank You

[rugged oriole]

im trying to do the exact same thing

the ragdoll replicates, just desyncs

physics dont sync, and ive heard too many ppl say not to replicate physics

but i need the ragdoll physics to be replicated here 😭 idk how to go abt it

have u come up with a solution

[nimble onyx]

What I did was Make custom events in the player bp SVR_Sim , In My Controller I Call It From My Bound DIspatcher From My Component OnRep_Dead, And My Death Event

and this is my end result

[rugged oriole]

alr I'll take a look

[rugged oriole]

just a question, if you were to move your ragdoll around, would it sync across all clients?

[nimble onyx]

dont know thats not a feature planned for

[rugged oriole]

ah

the ragdoll for me works, but like when the player kicks it around n such, it'll always desync

might need to constantly set the bone rotation and position in tick with a repnotify

[quasi tide]

Don't do that

It is a massive waste of resources.

[rugged oriole]

what would your solution be?

[quasi tide]

Unless, for whatever reason, it is actually gameplay critical that dead bodies be in the exact same location and pose for everyone.

[rugged oriole]

true it isnt critical

[quasi tide]

To just not do it. No one actually cares.

It is purely cosmetic.

[nocturne quail]

my linux server is crashing repeatedly, what can cause this?

Signal 11 caught.
Malloc Size=262146 LargeMemoryPoolOffset=262162
CommonUnixCrashHandler: Signal=11
Failed to find symbol file, expected location:
"/mnt/d/ARMAServer/LinuxServer/ARMA/Binaries/Linux/ARMAServer-Linux-Shipping.sym"
[2025.05.06-19.45.56:739][265]LogCore: === Critical error: ===
Unhandled Exception: SIGSEGV: invalid attempt to read memory at address 0x00000001b71d1660

[2025.04.06-19.45.56:747][265]LogCore: Fatal error!

0x0000000006d49392 ARMAServer-Linux-Shipping!UnknownFunction(0x6b49392)
0x0000000008245716 ARMAServer-Linux-Shipping!UnknownFunction(0x8045715)
0x000000000747d11c ARMAServer-Linux-Shipping!UnknownFunction(0x727d11b)
0x0000000006efbe05 ARMAServer-Linux-Shipping!UnknownFunction(0x6cfbe04)
0x0000000006d03470 ARMAServer-Linux-Shipping!UnknownFunction(0x6b0346f)
0x00000000079012de ARMAServer-Linux-Shipping!UnknownFunction(0x77012dd)
0x000000000790226a ARMAServer-Linux-Shipping!UnknownFunction(0x7702269)
0x00000000078f536b ARMAServer-Linux-Shipping!UnknownFunction(0x76f536a)
0x00007fc8cc61a1ca libc.so.6!UnknownFunction(0x2a1c9)
0x00007fc8cc61a28b libc.so.6!__libc_start_main(+0x8a)
0x000000000380c029 ARMAServer-Linux-Shipping!UnknownFunction(0x360c028)

[2025.04.05-19.45.56:771][265]LogExit: Executing StaticShutdownAfterError
Engine crash handling finished; re-raising signal 11 for the default handler. Good bye.
Segmentation fault (core dumped)
Segmentation fault (core dumped)

[lost inlet]

well you don't have symbols, so who knows

[nocturne quail]

i am testing it on wsl, no idea how to attach debugger to it

[lost inlet]

well at the bare minimum have the symbols

at least that gives a readable callstack

[nocturne quail]

can i launch the linux server from visual studio?

[lost inlet]

you can attach to WSL sessions, but either way you need the symbol file

[nocturne quail]

what i am doing now is compiling in vs and launching in wsl

i have the symbol files for non packaged

which should also crash

[tardy fossil]

gotta enable that in project settings -> Packaging

[lost inlet]

^

[nocturne quail]

if i build in developement server for linux, it never crash

seems like there is something not supported in shipping

but than it should not package if there is some not supported things exists

[wanton sleet]

is there a function to get all player characters ?

[exotic wasp]

i would just disable collision with it tbh

[nova wasp]

the PlayerArray of the gamestate

and then get each of their pawns

but you can just iterate over the actors as well, depending on timing the playerstates might not see their pawns quite yet if its inside of network reading

<@&213101288538374145>

guards, arrest this man at once

[nimble onyx]

xD

How Would I Replicate An Aim Offset ? Bc This Get BaseAim Roation Aint Working For me when i look down i get a positive num for some reason it works fine when ran in standalone but not as clients

[nova wasp]

the pitch should be replicated in RemoteViewPitch

this looks like you are replicating your own rotation though

you may need to override pawn functions to get the new rotation

[nimble onyx]

i got it thanks

[hollow lintel]

How might a world subsystem check for authority? Is GetWorld()->GetNetMode() < NM_Client the way to go, or is there a better way?

[fossil spoke]

Yes, if you read the comment on the ENetMode enum it should say anything < NM_Client is some form of a Server.

The implication of that being "authority".

[hollow lintel]

yeah that's just what I was looking at. 😄

In my head I'm like "great, it's a server. Does that mean it has authority?" Which, I've never seen a scenario in UE where the server doesn't have authority. But that enum doesn't mention authority, so I thought I'd double check.

[quasi tide]

Client spawned actors - server won't have authority unless you link them up and give the server authority.

[dark parcel]

0o you can link it up?

Interesting

[hollow lintel]

Will client-spawned actors even exist on the server to know about them?

[quasi tide]

No. Which is why I said the other parts of the message.

[hollow lintel]

Continuing with this world subsystem: I'm making an RPC that can be called via said subsystem. Is there a way to find out what client called the RPC on the server? I'm guessing I can get other info from that like the player controller, character, etc.

[blazing spruce]

Hi, I have an ability on my pawn class that when enabled/disabled I need a load of other actors to react to it and do some shit, I'm using an interface to do this but most of the stuff I need done on other actors only needs to happen for the player that's activated the ability, so for example, one of the things that happen when the ability is activated is the ambient sounds will fade out.. once IsUsingAbility is true it calls GetAllActorsWithInterface and calls the event on the interface passing in IsUsingAbility, my BP_Ambience class implements the interface so it can call an input on the Metasound to fade the sound out and back in again when the ability is disabled.

To make that sound fade out only for the player activating the ability I can pass the pawn through the interface and do a IsLocallyController check beforehand however once I add the pawn class reference to the interface it creates a dependency. Is it possible to make it so I can have it run only for the player activating the ability without passing the pawn class through the interface to avoid the dependency it would create or do i have just have to pass the pawn class through to be able to do what I want?

[fossil spoke]

Subsystems arent replicated

So they dont support RPCs

[hollow lintel]

shit.

[nova wasp]

if you want a singleton that is replicated the simple thing is to either use a manager actor or a component on some other replicated actor

replicating subsystems is not impossible but not really simple

https://github.com/daftsoftware/GameStateSubsystem

[hollow lintel]

I think I'd rather just create the actor. But this feels very much like a Unity engine thing to do 😄

[nimble onyx]

Thank You Everyone That Helped Me I Now How Fully Replicated Physics Interaction System

[rugged oriole]

why is the IsMoving check being replicated even tho it's not a replicated variable 😭

for example, if any client moves, it's set to true for all clients

[fossil spoke]

@rugged oriole Because Velocity is affected by replicated data.

IsMoving is not being replicated.

It is just changing as a result of data that is replicated, but is abstracted away.

[rugged oriole]

wdym

[fossil spoke]

In the same way Velocity is.

[rugged oriole]

how can I prevent that

[fossil spoke]

What do you mean? Movement of a Character is something you typically will always want replicated.

[rugged oriole]

velocity is being replicated?

im confused

[nova wasp]

the velocity of the character movement component is affected by actor transform replication from the owning character afaik

[fossil spoke]

If this confuses you, you probably need to focus on SinglePlayer for a bit.

^

A Characters movement is replicated by default.

Therefore, it will update its Velocity based on any movement.

[rugged oriole]

how would I go about doing an only local isMoving check then

[fossil spoke]

And since you are querying the Velocity for setting IsMoving, this means its value is derived from something else that is replicated.

[rugged oriole]

without using velocity

[rugged oriole]

doesnt rlly make any sense, im trying to do a camera shake but it plays for all users, and the only way it could do that is if both the IsMoving check and IsSprinting check is true on all users, ok if velocity is being replicated and IsMoving is being set to true bcuz of that, that doesn't explain why IsSprinting is being set to true when it has nothing to do with character movement component and isnt a replicated variable

either that or im a complete idiot and it's client start camera shake plays that plays all clients

fixed it, forgot i can use Run on owning Client for custom event 👍

[rugged oriole]

Get Local Role also saved my life

[thin stratus]

Why is that an RPC though?

I haven't read through all of it, but aren't you just playing some ScreenShackes or so based on the Character moving?

[thin stratus]

If that's all you want, you'd simply use a branch with "IsLocallyControlled" inside your Pawn/Character.

That will be True for the local instance of each player.

The LocalRole stuff above is somewhat redundant, as you arleady have the IsLocalController check there.

The Controller wouldn't even be valid on SimulatedProxy

[pallid stone]

Got a question on the net cull distance squared. The default is 225000000.0, and the square root 15000. Is that 15000 unreal engine units (150m), or 15000 meters?

[chrome bay]

UE units, 150 meters

[pallid stone]

Thank you. Does this mean that a client "forgets" an actor exists, or it just sits there not really recieving any updated information?

[chrome bay]

Client destroys the actor completely, if you enter relevancy range again a new one is spawned

Well, for an actor spawned at runtime anyway. For actors that are part of the map, they just sit there and get no updates

[magic vessel]

So the server keeps track of these actors that go out of range of relevancy? What happens in Listen Server architecture?

I'm fairly confident with a lot of multiplayer concepts, but relevancy in these cases has always been a bit unclear to me

[chrome bay]

Every frame the server evaluates whether an actor is relevant for a connection or not, and closes/opens the channel for that actor if it needs to.

Relevancy is per-client. Server of course doesn't have any relevancy concept, the actors always exist server side.

[prisma snow]

I assume that the check is skipped for actors that are always relevant

[chrome bay]

yee

and there is a timeout (default 5 seconds IIRC) to stop actors flitting in/out of relevancy frequently if you are right on the border - since opening/closing channels and spawning/despawning actors relatively speaking is costly.

[prisma snow]

yeah makes sense

[rugged oriole]

I have no clue but when i added a pin off of simulated proxy it, it played on both screens

I’ve tried using the Is Local Controller check and got the same result

[pallid stone]

This, is excellent information to know. If you wanted the client to keep a copy but not necessarily destroy it, what would be the best method?

[chrome bay]

Don't think you can

[pallid stone]

OK thank you

[bold wolf]

I am trying to use max draw distance settings of bps
but cant to it
it is not even working on server

[hollow lintel]

Didn't want to spam, but my problem seems like a multiplayer issue as much as a GAS one
edit: apparently the forward button in discord spams for you

And my follow-up comment:

Actually going through the debugger, one of two things happen:

1. Attached to PIE: local role is sim proxy inside UE source ASC code (handle gameplay event)
2. Attached to server: rider let's me know that the breakpoint won't be hit, and sure enough it isn't.

I would have thought that PIE local role would be auto proxy

[main bluff]

     UPROPERTY(BlueprintReadOnly, ReplicatedUsing = OnRep_CurrentStackCount, Category = "Item")
    int32 CurrentStackCount = 1;

    UFUNCTION()
    virtual void OnRep_CurrentStackCount() {};``` Will it work when I will override this virtual method in the child class? What be called in the child class when CurrentStackCount will be replicated, CurrentStackCount will  call this class function or  child class  function?

[dire burrow]

    UPROPERTY(ReplicatedUsing = OnRep_MousePositionVector, BlueprintReadWrite, Category = "Movement")
    FVector2D MousePositionVector;
    UFUNCTION()
    void OnRep_MousePositionVector(); // OnRep function for MousePositionVector    
    virtual bool UpdateHeadRotationPure() override;

^this works well assuming your OnRep function fires 😄

[boreal scarab]

Hello!

What's a good checklist to go down to figure out why my materials are not replicating, but the meshes are?

Looks like this:

[dark edge]

show your code

a material doesn't replicate, a reference to a material asset can

[boreal scarab]

Ahh, okay which.. code specifically? My code is pretty.. I'm doing async loading, I have lots of interfaces, logic is pretty split

[dark edge]

why does the right one have the black and white squares, where does that data come from?

is it just a texture or is it some runtime stuff

[boreal scarab]

Its runtime, its being set here:

void AZBlock::UpdateDisplay()
{
    UMaterialInstance* BlockMaterial = DisplayData->BlockMaterial.Get();
    ensureAlways(BlockMaterial);

    // TODO: Dunno if we should be creating the DIM every time.
    UMaterialInstanceDynamic* NewMID = UMaterialInstanceDynamic::Create(
        BlockMaterial,this);

    if (NewMID)
    {
        BaseMesh->SetStaticMesh(DisplayData->BlockMesh.Get());
        BaseMesh->SetMaterial(0, NewMID);
    }
}

[dark edge]

ok so what's DisplayData

where is the array of bools or whatever represents that state

[boreal scarab]

UCLASS()
// TODO: This is used for Blocks in general and should be named as such. Not just elementals.
class ZONECONTROL_API UZBlockDisplayData : public UPrimaryDataAsset
{
    GENERATED_BODY()

public:
    // "AssetType:GameplayIdentityTag"
    virtual FPrimaryAssetId GetPrimaryAssetId() const override;

    // Returns the GameplayIdentityTag.
    FGameplayTag GetBlockTypeTag() const
    {
        return BlockTypeTag;
    }

    static FPrimaryAssetId GetFPrimaryAssetIdFromBlockTypeTag(const FGameplayTag& BlockTypeTag);
    
public:
    // Primary material for the block.
    UPROPERTY(EditAnywhere, BlueprintReadOnly, Category = "Visuals", AssetRegistrySearchable)
    TSoftObjectPtr<UMaterialInstance> BlockMaterial;

    // Primary mesh for the block.
    UPROPERTY(EditAnywhere, BlueprintReadOnly, Category = "Visuals", AssetRegistrySearchable)
    TSoftObjectPtr<UStaticMesh> BlockMesh;

    // GameplayIdentityTag. Describes what this block is trying to be.
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "Tags")
    FGameplayTag BlockTypeTag;
};

And that gets loaded asynchronously before the board is spawned

[dark edge]

is a board one block or many blocks?

[boreal scarab]

many blocks. I make sure its all loaded first here:

void AZGameMode::NotifyGameBoardDependenciesReady() const
{
    bool bIsDependenciesReady = true;
    bIsDependenciesReady &= bIsGameBoardDataReady;
    bIsDependenciesReady &= bIsBlockBlueprintReady;
    bIsDependenciesReady &= bIsElementalDataReady;

    if (bIsDependenciesReady)
    {
        OnGameBoardDependenciesReadyDelegate.Broadcast();
    }
}

[dark edge]

What is the fundamental data a client needs to draw the same board

I'm guessing it's the BlockTypeTag right?

that indicates if it should be black or white?

[boreal scarab]

Yeah it comes from my BlockState object
class ZONECONTROL_API UZBlockState : public UObject, public IZBlockStateView

which looks like this:

 */
UCLASS()
class ZONECONTROL_API UZBlockState : public UObject, public IZBlockStateView
{
    GENERATED_BODY()

public:

    // ~IZBlockStateView impl
    UFUNCTION(BlueprintCallable)
    virtual FGameplayTag GetBlockTypeTag() const override;

    UFUNCTION(BlueprintCallable)
    virtual int32 GetBlockId() const override { return BlockId; }

    virtual FOnBlockStateChanged& GetOnBlockStateChangedDelegate() override { return OnBlockStateChangedDelegate; }
    // ~End IZGameAssetProvider interface.

    UFUNCTION(BlueprintCallable)
    void SetBlockTypeTag(const FGameplayTag& NewBlockTypeTag);

    UFUNCTION(BlueprintCallable)
    void SetBlockId(int32 NewBlockId) { BlockId = NewBlockId; }

    UPROPERTY(BlueprintAssignable)
    FOnBlockStateChanged OnBlockStateChangedDelegate;

protected:
    // Set on construction and never changed. Used to reference the owning player.
    UPROPERTY(BlueprintReadOnly, Category = "Info")
    int32 OwningPlayerId = -1;

    // Set on construction and never changed. Use this if a handle is needed.
    UPROPERTY(BlueprintReadOnly, Category = "Info")
    int32 BlockId = 0;

    // Use EBlockSide to position these values.
    UPROPERTY(EditDefaultsOnly, Category = "Gameplay")
    TArray<int32> PowerValues;

    // GameplayTags for various mechanics.
    UPROPERTY(EditAnywhere, BlueprintReadWrite, Category = "Gameplay")
    FGameplayTagContainer GameplayTags;


    // Can be BlockType.Elemental.* or BlockType.Tile.*. Used to identify the type of block.
    UPROPERTY(ReplicatedUsing = OnRep_BlockTypeTag, EditAnywhere, BlueprintReadWrite, Category = "Gameplay")
    FGameplayTag BlockTypeTag;

private:
    // Replication
    UFUNCTION()
    void OnRep_BlockTypeTag();
    // End Replication
};

[dark edge]

yeah you're right this is a mess.
Anyways, you need to have some replicated state somewhere.
Somehow, somewhere, there needs to be 81 bools or tags or whatever that represents the state of the board

you could have a block be a replicated actor with a replicated tag, you could have a map of block ref to tag, you could have an array of tags, whatever.

but you need that data to be replicated, then everything can run the same from there

[boreal scarab]

ohhh!! that's probably what it is, my ZGameState has the block states

private:
    UPROPERTY(Replicated)
    // TODO: Put this in docs somewhere:
    TArray<TObjectPtr<UZBlockState>> GameBoardState;

but I don't think I'm implementing DOREPLIFETIME

[dark edge]

OnRep_BlockTypeTag is where you should apply that, unless that's some leftover code

do you want a bunch of replicated blocks with replicated data, or just some single replicated aggregate?

Assuming this worked, and is correctly called when the tag is changed, this should be what triggers the block to change material LOCALLY

[boreal scarab]

Well the idea is that the array of block states that GameState has is the source of truth. Its the only place that block states can be modified, and then the UI simply updates when its state updates

[dark edge]

why even have the tags on the individual blocks then

choose one

either way, that's how you should do it.

OnRep_SomeState -> do what needs to happen when that state changes (swap material, set parameter, whatever)

[boreal scarab]

that's a good point yeah, no need to cache the tag in two places. The blocks have a view into their data anyway

[dark edge]

yeah all that gives you is the possibility to have a conflict, with no real benefits

[boreal scarab]

yeah agree (i think that was leftover from an older design I had and I just hadn't relooked at that). So really all I need to do is make sure that the states are replicated properly in GameState, and remove the tag from the block object

[dark edge]

At a certain point you might not need block objects at all.
Gamestate:
TArray<staticmeshcomponent> BlockMeshes
TArray<int32> BlockIDs
etc. Might end up going that way

but if you don't have thousands of them no real reason to go down that road IMO

[boreal scarab]

Yeah I ended up splitting that out into a ZGameBoardManager which is a UWorldSubsystem. Just to kind of keep GameState as sort of a data only management thing, and the board manager is more of a "frontend" type object for lack of a better word

[dark edge]

yeah i wish subsystems could replicate, that'd be dope

I'll need to tackle it eventually for my project, I have many subsystems, one per domain (mechanical, electrical, fluids, signals) and one per device type (engine, cockpit, wheel, propeller, gearbox, etc), each of which maintains an array of their things

[sterile plaza]

Anyone know things about COND_Custom? I set a property to use that and I implemented PreReplication and call DOREPLIFETIME_ACTIVE_OVERRIDEit is always replicated.

Documentation here is pretty straightforward. https://dev.epicgames.com/documentation/en-us/unreal-engine/replicate-actor-properties-in-unreal-engine?application_version=5.5#custompropertyreplication

[boreal scarab]

hehe, i did not know they didn't replicate! I don't think that changes anything for me but that's good to know.

BTW is this design like.. I know its a lot of plumbing but feels clean in terms of responsibility? Blocks are gonna have abilities and can there's gonna be a lot of logic for capturing and using those abilties. So I figured having the GameBoardManager be the thing that takes UI reqeusts and authoritatively checks if they're allowed, modifies the state, and then UI updates just work because its connected to the view via a onChanged delegate seemed clean?

[meager heart]

Does anyone have any suggestions on how I can optimize these 10k+ actors for networking? I've implemented distance relevancy and made them not replicated until interacted with but I still spend a ton of performance on networking

The current implementation is probably not very performant by itself, but at rest a lot of time is still spent sending networking data that I don't quite understand

[boreal scarab]

uh.. actually I think alot of my issues are coming from learning how UE replicates things.

This:
TScriptInterface<IZBlockStateView> BlockStateView = nullptr;

Is supposed to be just a lightweight view into the ZBlockState. It simply has getters. So the idea was to do this:

UCLASS()
class ZONECONTROL_API AZBlock : public AActor
{
    GENERATED_BODY()

public:
    AZBlock();

public:
    UFUNCTION(BlueprintCallable)
    void Init(TScriptInterface<IZBlockStateView> InBlockStateView,
        TScriptInterface<IZBlockGameplayInterface> InGameplayInterface);

And pass it NewBlock->Init(BlockState, GameState); those objects. And then it can simply update its own UI by asking questions through the view.

But uh.. basically all of my interfaces are nullptr on the client. Does unreal just not support this kind of paradigm? Like.. maybe interfaces aren't allowed on clients or something?

[nocturne quail]

why i'm getting this error when joining the session?

Client log

[2025.02.07-18.56.57:809][236]LogNet: Error: UEngine::BroadcastNetworkFailure: FailureType = ConnectionLost, ErrorString = Your connection to the host has been lost., Driver = Name:GameNetDriver Def:GameNetDriver IpNetDriver_0
[2025.02.07-18.56.57:813][237]LogNet: Error: UEngine::BroadcastNetworkFailure: FailureType = ConnectionLost, ErrorString = Your connection to the host has been lost., Driver = Name:GameNetDriver Def:GameNetDriver IpNetDriver_0

Server Log:

[2025.02.07-18.56.47:046][199]LogNet: UNetConnection::Close: [UNetConnection] RemoteAddr: 127.0.0.1:50720, Name: IpConnection_2147474725, Driver: Name:GameNetDriver Def:GameNetDriver IpNetDriver_2147482471, IsServer: YES, PC: BP_PlayerController_C_2147474719, Owner: BP_PlayerController_C_2147474719, UniqueId: NULL:DESKTOP-164L35-CBF269C49A7AE146D61D3456BC606496, Channels: 34, Time: 2025.04.07-18.56.47
[2025.02.07-18.56.47:049][199]LogNet: UNetConnection::SendCloseReason:
[2025.02.07-18.56.47:055][199]LogNet:  - Result=MissingLevelPackage, ErrorContext="/Memory/ZCXHBH5BHUVE8BAYIGC9SJYIV,/Memory/ZCXHBH5BHUVE8BAYIGC9SJYIV"
[2025.02.07-18.56.47:059][199]LogNet: UChannel::Close: Sending CloseBunch. ChIndex == 0. Name: [UChannel] ChIndex: 0, Closing: 0 [UNetConnection] RemoteAddr: 127.0.0.1:50720, Name: IpConnection_2147474725, Driver: Name:GameNetDriver Def:GameNetDriver IpNetDriver_2147482471, IsServer: YES, PC: BP_PlayerController_C_2147474719, Owner: BP_PlayerController_C_2147474719, UniqueId: NULL:DESKTOP-164L35-CBF269C49A7AE146D61D3456BC606496
[2025.02.07-18.56.47:068][200]LogNet: NotifyAcceptingConnection accepted from: 127.0.0.1:50720
[2025.02.07-18.56.47:074][200]LogNet: NotifyAcceptingConnection accepted from: 127.0.0.1:50720
[2025.02.07-18.56.47:101][201]LogNet: NotifyAcceptingConnection accepted from: 127.0.0.1:50720

when joining a session hosted using small map works fine

session with a big open world map don't works

with big map i am getting this error as well

UNetDriver::TickDispatch: Very long time between ticks. DeltaTime: 21.22, Realtime: 21.22. IpNetDriver_2147482471

[formal solar]

Would a low framerate on listen server (mostly from CPU) increase ping for connected clients?

[chrome bay]

yes

since whatever you're sending is taking more time to process, and more time for the results to get back to you

[formal solar]

Thank you. Might be worth trying to cut down CPU fram time before delving into ping compensation and all the fun stuff

[nimble onyx]

is it easier to do replication *In c++ Or Blueprints

[quick lagoon]

hey im having some kind of issue with my multiplayer spawning, for some reason when additional players spawn in other players get their pawns unpossesed and i dont know why. i am very new to multiplayer/netcode stuff so i dont really have a clue as to why it has started to do this, it was working fine about 30 minutes ago

so like in this example you can see that after the second player in the smaller window loads in the first player to join gets their pawn unpossessed

[boreal scarab]

is this allowed?

    UPROPERTY(ReplicatedUsing = OnRep_BlockStateView, BlueprintReadOnly, Category = "Gameplay")
    TScriptInterface<IZBlockStateView> BlockStateView = nullptr;

[quick lagoon]

I've gone through every instance of my players being possessed and unpossessed and I think this intial spawning code from my game mode might be causing it, but I dont see why it would be doing it, it should just run through the list of connected player and spawn pawns for them

could it be my play settings? like whether im using standalone or listen server?

[boreal scarab]

Welp I figured it out. Whether this actually reacts to changes properly remains to be seen but.. I just needed:

void AZBlock::OnRep_BlockStateView()
{
    UpdateDisplay();
}

bool AZBlock::ReplicateSubobjects(UActorChannel* Channel, FOutBunch* Bunch, FReplicationFlags* RepFlags)
{
    bool bWroteSomething = Super::ReplicateSubobjects(Channel, Bunch, RepFlags);

    if (BlockStateView.GetObject())
    {
        bWroteSomething |= Channel->ReplicateSubobject(BlockStateView.GetObject(), *Bunch, *RepFlags);
    }

    return bWroteSomething;
}

As well as I needed to guard UpdateDisplay to early out if BlockStateView was nullptr because that function apparently can get called before the view is fully replicated

[quick lagoon]

Ok I believe I figured it out, i wasn't check if the player controller already had a pawn so it would get overriden

[exotic wasp]

you can get by with some basic stuff in bp but its really best done in c++

replicating a property in c++ though is several steps whereas you just tick a box in bp

[nimble onyx]

So far its going quite well replicating in cpp to me it feels easier and some i might need to do it in c++ just because im replicating some physics

[exotic wasp]

ah yeah c++ lets you override functions that run after replication which is needed for physics stuff afaik

[quick lagoon]

ok another issue, im having an issue where my client players are not considered to be locally controlled? how do i change this

[sinful tree]

Pawns should only be considered locally controlled if they're possessed by the local player's controller. Any other pawns would not be considered locally controlled.

[quick lagoon]

But like when I have the player Pawns print their own locally controlled status they say its false

[dark parcel]

Why do you use plural? A player can only possess a pawn.

If I'm playing with 6 players where each person have their own pawn.

My pawn will return IsLocallyControlled true.

Rest are false.

[quick lagoon]

So does islocallycontrolled only apply to the player connected to/hosting the server?

[dark parcel]

?

[quick lagoon]

Ohhh I see what you mean

[dark parcel]

Can you reword your question.

What does player connect to / hosting the server even mean.

You just describe both client and server.

Not that locallycontrolled have anything to do with it.

Everyone is running their own instance.

You run your unreal.exe

I run mine.exe

We both execute codes

They don't happend in one instance. I hope this give you perspective.

[quick lagoon]

It does to an extent, yeah

[dark parcel]

So if you run is locally controlled check in your machine.

You will get your possessed pawn return true

While the rest of the pawn to be false

And the same thing when I run the same function.

I will have my possessed pawn return true

While the pawn that belong to you in my world return false.

You can test this easily

Create a text component.

On tick, write true if is locally controlled is true and false otherwise.

Then just play the game with 3 players.

[quick lagoon]

Yeah I think I understand, at least somewhat

[dark parcel]

Just do the above exercise and you won't have to guess anymore.

Understanding roles and locally controlled actor is very important to split codes

[quick lagoon]

I literally started doing multiplayer stuff like a week ago so I guess I still haven't wrapped my head around the framework of it all

[nocturne quail]

How can i pass the string as FUniqueNetIdRepl ?

void AARMAGameModeBase::PreLogin(const FString& Options, const FString& Address, const FUniqueNetIdRepl& UniqueId, FString& ErrorMessage)
{
    Super::PreLogin(Options, Address, GameplayStatics::ParseOption(Options, "480"), ErrorMessage);
}

error C2664: 'void AGameModeBase::PreLogin(const FString &,const FString &,const FUniqueNetIdRepl &,FString &)': cannot convert argument 3 from 'const FString' to 'const FUniqueNetIdRepl &'

FString UGameplayStatics::ParseOption( FString Options, const FString& Key )
{
    FString ReturnValue;
    FString Pair, PairKey, PairValue;
    while( GrabOption( Options, Pair ) )
    {
        GetKeyValue( Pair, PairKey, PairValue );
        if (Key == PairKey)
        {
            ReturnValue = MoveTemp(PairValue);
            break;
        }
    }
    return ReturnValue;
}

[sinful tree]

FUniqueNetIdRepl::UniqueIdFromString()

UniqueIdFromString ( FName Type, const FString& Contents)

?

[nocturne quail]

testing it now thanks

[nocturne quail]

this is a protected method, is there any public accessor to it?

[dark parcel]

If the code is short. Can't just copy paste to a new function?

[nocturne quail]

linked in many headers

void FUniqueNetIdRepl::UniqueIdFromString(FName Type, const FString& Contents)
{
    // Don't need to distinguish OSS interfaces here with world because we just want the create function below
    FUniqueNetIdWrapper UniqueNetIdWrapper = UOnlineEngineInterface::Get()->CreateUniquePlayerIdWrapper(Contents, Type);
    SetUniqueNetId(UniqueNetIdWrapper.GetUniqueNetId());
}

[quick lagoon]

im still experiencing a discrepancy between players with regards to the locally controlled player thing, I have this multicast event begin play node where if the controller is a local controller it will add a hud to the player's viewport, whereas if not it just prints out some text. When I run the game in listen server mode the first player that joins (I assume they are the "host") gets their hud no problem, but the second player does not and prints the text (in the 2nd screenshot the right window is the "server")

[sinful tree]

Player Controllers don't replicate to all clients. Only to the owning client. This also means it's kind of pointless to multicast on them.

[quick lagoon]

huh ok, hooking it into the normal begin play seems to at least paritally fix it

[sinful tree]

Checking if it's a local controller on begin play should be enough to ensure that local players only create the hud. Ideally you would probably use the HUD class to create and add your widget which is created for locally controlled player controllers, so then you don't even need to do a local check at all.

[quick lagoon]

Are you talking about the default hud class in the game mode?

[sinful tree]

Yep

[nocturne quail]

ended with this. 😄

void AARMAGameModeBase::PreLogin(const FString& Options, const FString& Address, const FUniqueNetIdRepl& UniqueId, FString& ErrorMessage)
{
    ErrorMessage = GameSession->ApproveLogin(Options);
    FGameModeEvents::GameModePreLoginEvent.Broadcast(this, UniqueId, ErrorMessage);
}

[left marsh]

How can I make the Collision Profile of a collider replicate? For a Character's capsule.

[dark parcel]

Multicast is server telling everyone that have a copy of the actor to run the function.

I think you need to read the pinned material first. Get to know what's replicated or not.

All controllers exist on server but clients only know their own controller.

[dark parcel]

You need to have a sync value that dictate the profile.

I used enum to describe the character mobility

OnRep->

Switch (mobility)
If pushable
Set collision profile to pawn, etc.

Or i guess you can replicate a struct that contain all the info that you need to replicate.

[nova wasp]

yeah you would want something specific to the game imo that describes either being default or the pushable mode

each thing can define what the default is (or even from the cdo)

[torpid lantern]

So right now my dedicated server (Dasv4- 2 cores) craps out with 200 NPCs in play.

The NPCs are replicated characters with an AI controller, State Tree, and a GAS component.

Game has an isometric camera and I can really shrink the net cull. So far I've kept what I believe to be "heavy" logic away from tick. I could add more cores to the hardware, but I'd prefer not to. What are my options to get this up to the 1000 range?

I'm considering maybe Mass, or perhaps moving away from "Character" - but I think I'd need my own movement system, right?

[nova wasp]

so have you profiled it?

also no more cores will not save you here unfortunately as the cmc is entirely single threaded
there is an experimental async one that I have not seen used

I would say with such a low count you should be able to just manually tick them round-robin style and see heavy gains

[torpid lantern]

I have not profiled it against the prod environment. I'm hosting in PlayFab and I haven't quite figured out how to remote profile.

[nova wasp]

you could even profile locally and see roughly similar numbers? why even ask about performance if you haven't clicked the insights button?

it's very easy to profile in modern unreal, you do not need to use the awful csv profiler

insights also works remotely but I'm not sure how that works with playfab

[meager heart]

Just re-pinging this because it's been a bit, looking for some help with improving networking performance with lots of actors

[nova wasp]

step #1 is to not represent these as replicated actors... this should be sent in a compressed way

I think you could keep the "distance based" property of net cull distance by having like one actor represent a set of these squares rather than 1 per square

I have 0 info on how these change either, I have no idea what you are trying to do

also would help to enable push model networking as just polling these will be insanely costly

[hollow lintel]

purely academic question: in theory you could simply replace all the networking in the CMC with RPCs as opposed to the saved moves that it does and now you have lockstep movement, yeah?

[nova wasp]

how do you think saved moves are sent?

[hollow lintel]

-______________-

curses

foiled again

[nova wasp]

saved moves are confusing but they really are just rpcs sent over

changing WHAT they do and how they are read back is something though I guess

and how they work on sim proxies is mostly just the actor transform+ velocity onrep and some other properties (which can come in at random frame times btw lol)

so there is absolutely a world where you send the inputs to sim proxies as well to get a "rollback" setup

[meager heart]

Hmm yeah that definitely makes sense, currently like you said each square is its own actor (technically the mesh itself is a child actor component which is intentional but is probably less performant, this is why I'm trying to get a base test without any sort of interaction)

I'll look into push model networking, I've never heard of that, so thank you!

one actor represent a set of these squares rather than 1 per square
Is there another method of doing this, outside of refactoring these actors to basically combine them into one? That's the only thing I can think of initially, outside of an engine system that I don't know about

[nova wasp]

"Is there another method of doing this" you still haven't described what "this" is

[knotty frost]

How do I prevent this from happing. I tired changing my collision setting and that didn't seem to fix it.

[meager heart]

Sorry that was meant to be a quote, formatting got messed up when I copy/pasted it - I was referring to your suggestion:

one actor represent a set of these squares rather than 1 per square

So I'm asking if there's another way of combining these actors for networking/replication outside of just actually making them one actor

[nova wasp]

I am asking about what the squares are supposed to be replicating for the game

[meager heart]

Honestly I'm just worried about improving the standing performance of all these actors, I'd like the static existence of all these actors to not sacrifice 20ms of CPU time every frame, and then I'll see if I can optimize their actual function

Sorry if that's unhelpful but ultimately I'm just trying to understand why these actors themselves are sending so much data without doing anything / being relevant

[nova wasp]

okay if this is just a proof of concept to see what happens I don't think I can say anything about how to make it better without knowing what the cubes are supposed to do

turns out having 10k actor channels that neat to check for dirty properties is expensive :U

[meager heart]

Let's just say the cubes are supposed to be static then and not do anything if that helps

[nova wasp]

if they are static then they should just be part of the level itself

[meager heart]

I didn't know that's what it was checking / sending data for, that makes sense

[nova wasp]

you may need to turn on more profiling events like named events to see more detailed info

[meager heart]

Ohhh didn't know that was there, thank you!

[nova wasp]

you did profile at least so you are headed in the right direction

a lot of people underestimate how you can see what is slow in your game by pressing 2 buttons lol

[meager heart]

Let me see what that says, hopefully that will point me in the right direction and then I'll try to see if I can group the actors like you said as well

[nova wasp]

(not everything is represented well in insights but it's going to show... something)

[meager heart]

yeah getting comfortable with profiling is super helpful lol

Honestly I work with it in my job and this is the first time I've ever thought about using it for my personal work - I like forgot I could use it

[torpid lantern]

It's more than two buttons, at least for me. I've profiled non-prod and I think I've traced mine to the CMC but honestly, I have no idea what I am looking for.

[nova wasp]

I would say my dumb guess for how to send them would be to represent each cube as a bit in an array replicated down as a single struct, where it only changes when cubes have something change on them
so the client represents them as meshes but they are not really network objects, but local only things that react to onreps from a manager actor

[nova wasp]

neither do I because I can't see anything

in general when profiling in the editor things will be a lot slower than a shipping build

you can make local builds just for profiling too, test builds are useful for this as they are mostly optimized in comparison to a debug build

basically you are looking for why the frame took a long time

if something is huge, you figure out why and if it needs to be

[meager heart]

Oh yeah this is much more helpful, I'd seen people use stat named events but honestly didn't know what it did lol, thanks again

[nova wasp]

stat named events makes things run tad slower but they emit events that the old CSV profiler would read, but insights can see them too

this is why it makes a giant "hey, stat named events are on" warning to tell you it will be slow

it actually makes blueprint assets show up as their names in the graph as well

as a rule of thumb 60fps is about 16.67777ms

so this netbroadcast tick time is taking... the entire frame

[torpid lantern]

Is that helpful?

[nova wasp]

actually yes... I am not sure if these are okay to share though as it might reveal some info about where it is stored on your pc

probably just file addresses though

[torpid lantern]

I see that, not concerned.

[nova wasp]

the one thing I would add to this trace would be tasks

but this is good because it covers the bases of cpu stuff

[torpid lantern]

I am using a .bat file to start the server with -trace=log,bookmark,frame,cpu,gpu,loadtime,file,net

Would I just add "tasks"? What does that give me visibility in to?

ps. In that trace, I spawn 200 NPCs around frame 2556. Prior is ~100 NPCs

[nova wasp]

yep that's a lot of cmc ticks lol

[torpid lantern]

Does Mass solve the CMC issue?

[nova wasp]

it is a for loop

[torpid lantern]

How the heck did you figure that out?

[nova wasp]

an ECS library ultimately is just querying for similar things and executing stuff in parallel

yes it makes stuff fast

but you have to give it stuff to do

and it cannot magically solve things that are inter-dependant without some help

one reason the cmc must run on the main thread is that is modifies things that the main thread reads

like physics bodies and actor transforms, things it hits

[echo pasture]

did they remove widgets staying in seamless server travel or is my own stuff just bugged

[nova wasp]

what you have here in this trace at a surface level

• general cmc traces to check floor that you kind of can assume is there with navwalking
• moving kinematic bones you might not need to move
• updating all animations every single frame

[torpid lantern]

I appreciate this. I'm very curious to learn how you pulled this from the trace.

[magic karma]

I don’t mean to interrupt anything but I’m so frustrated and confused that I know I’m missing something important. I’m trying to keep my clients and server very separated(I just need my server to run an occasional RPC) but it feels like Unreal is fighting me every step of the way. Like I’m trying to get my client’s controller to possess another pawn, but for some reason the possess node only runs on the server. Why does my server, who just processes data storage, need to know what pawn the client is controlling? Is there actually no way to possess a pawn locally?

[nova wasp]

I zoomed in and read the function names

this is a single cmc

the sweeps are quite cheep but updating the kinematic bone collision bodies is 2x the cost for... I don't know

if you aren't actively using the physics body every frame why update it I guess

it's a commonly used optimization strategy to just not udpate these unless needed

this is the physics bodies bound the the skeleton which I don't know if you use

[torpid lantern]

I only use them on death, when I enable ragdoll.

[nova wasp]

then don't turn on collision for the skeletal mesh

until then

[quasi tide]

Yeah - you lack the understanding of UE's entire networking architecture it seems. It is server authoritative. Meaning the server is the one true version of the game.

The server needs to know because it is the ultimate authority for the game.

[magic karma]

It shouldn’t matter what the true version of the game is. I’m able to spawn actors only on the client, and I’m able to call rpcs from things that exist on the client but not the server(I think). So why is player character possession any different?

[quasi tide]

Because in the eyes of the server, the player controller is the player. So they need to have network authority over what they possess. If it just blindly lets anyone have network authority of anyone else, it would be a disaster.

[nova wasp]

there is one case of something being locally posessed (spectating) but it is a unique case

[magic karma]

I get that player controllers are special, but it still doesn’t make sense to me why the server needs the character. And in my case other characters aren’t even spawned into clients. But assuming that the server does need to know about a client possessing something, what would be the cheapest way to show replicate that possession back to the client? I haven’t had a need to use conventional replication until now. AFAIK the lowest replication rate is twice a second? Is there a way to only tell the client about the possession when it happens? (Also what do I even replicate to the client to tell them about the possession)

[nova wasp]

and it requires a lot of c++ overrides

[quasi tide]

Literally just do a server rpc to do the possession. It is that simple.

[nova wasp]

the character can be local only and you could have a dummy pawn serve as the possessed thing

you can ignore possession and just forward input + camera

but why change this for no reason

[magic karma]

So if I call possess on the server that automatically gets replicated to the client?

[quasi tide]

Yes

[hollow lintel]

Can I have something like a plain UObject as a member of an actor contain all the implementation for my RPCs? I'm looking for something lightweight, and a way to declutter my character class, and RPCs are easily the biggest eyesore there.

[nova wasp]

yep

actor components can rpc

but they must be owned by something that can rpc the same way

this is good way to cut down on rpc spaghetti

[exotic wasp]

you can but I don't know why you would use uobject

use a component and then you can take advantage of GetComponentByClass

[nova wasp]

just make sure you actually mark it to replicate if using the subobject list etc

the docs do a decent job of showing how to do this iirc

[hollow lintel]

Lightest weight possible was what I was thinking. Megafunk mentioned components, but don't actor components have their own transform that have to update?

(even uobjects have way more bloat than I want. It'd be nice to just throw it all in a raw c++ class)

[nova wasp]

actor components are not scene components (scene components extend actor components to have a transform)

if being heavy is an issue describe the actual performance problem and scale at hand

if you don't want to use objects you may have a difficult time in the unreal engine gameplay framework

network "manager" objects that have state for everything at once (or an area) can be a good way to gather state for multiple things in one place I guess, but you lose out on controlling them as individual things etc

if memory overhead is an actual issue from just actor components I think you are doing some serious heavy-weight stuff lol

a single art asset will instantly dwarf the memory consumed by almost any component

unreal objects have a lot of pointless stuff taped on them for legacy reasons but individual actor components are not the problem

[chrome bay]

Also - turn on Deferred Kinematic Updates, assuming this is being done in PrePhysics

tbf it's rare you don't want some kind of collision on your characters skel mesh

If you're using Mover, you have to make engine changes to make deferred kinematics work.

[formal solar]

I get confused by all the 'CPU Stall - wait for even't entries in the profiler

[chrome bay]

IIRC that usually means that thread is stalled waiting for another to finish. If you have the "Tasks" channel enabled in the profiler you can see what it's waiting on by clicking on it, and you'll see an arrow with the dependency.

Assuming you're using insights that is and not the old profiler

[chrome bay]

UActorComponents don't. Replicated UObjects have additional networking overhead of their own.

There's a reason CharacterMovementComponent calls RPCs on the parent actor, not itself.

[formal solar]

I use the profiler from 'Session Frontend'

after saving a log with startfile/ stopfile

[chrome bay]

That's the old one. Use insights instead.
https://dev.epicgames.com/documentation/en-us/unreal-engine/unreal-insights-in-unreal-engine

[grand kestrel]

Insights is actually amazing

[nova wasp]

they should seriously just have the old csv profiler go "you probably want insights" lol

I can imagine there are still situations where it makes sense but... I do not know why you would when insights is around

[dire burrow]

So i have searched absolutely every where trying to figure out whats wrong .. but no mater what i try Bone Rotation simply will not replicate.. so i found a a odd bit of information that was talking about the variable needing to be to in GameState so i tried this dispite getting the feeling its not right .. but doing this i discovered it causes all the clients to recieve the same rotation ... now i get what replication in the game state is for .. but i just do not understand how its not replicated from my BP which deals with the controller input .. it doesnt need the controller varaiables as the variable belongs to the Fore mentioned BP.. Replicated/Replicate Movement is enabled for the BP and headrotation/mousepositionvector are set to replicate.. the Animation BP updates its own version of the varaiables from those replicated ones.. which still leaves me confused because the Bone Modify sets there own rotation.. so then i see that the varaiables can also be replicated in side the AnimBP... but does match any other guides.. as I mention (every single guide is a toggle switch 🤣) onRep Notify wont help as thats just saving the variable being fully replicated to my understanding.. if any one has any thing for me on this i would really appreciate it

[bold wolf]

I searched online but couldn't find any, does nanite work for multiplayer?

[nova wasp]

nanite is a way to render meshes

this does not affect how multiplayer works to my knowledge

[dire burrow]

oops the srv event is set to mc there but regardless even with run on server it still dosnt work

[lost inlet]

?

[nova wasp]

the possessed pawn/ player controller rule is for client -> server rpcs as they must be objects owned by the connection

[lost inlet]

Multicast RPCs work on any actor that is replicated. Client/Server RPCs work on any actor as long as there's an owner

[nova wasp]

the server can RPC any replicated network stable actor whenever

[bold wolf]

so this actor not owned?

[dire burrow]

ah wait

when i spawn the MGT_Commander BP i never gave it a instigator node

could that be causing all this?

[lost inlet]

Well the owner is the relevant one

[nova wasp]

that appears to be a server multicast rpc so if the server created it and it is replicated it should work as expected

assuming you actually call this on the server

and it is replicated

[dark parcel]

Did he said variable can be replicated inside anim bp?

Probably just misundertood

Since he paste a tutorial with incorrect title

"How to replicate variable in anim bp"

https://m.youtube.com/watch?v=R3mPtalY4o4&pp=ygUmSG93IHRvIHJlcGxpY2F0ZSB2YXJpYWJsZSB3aXRoIGFuaW0gYnDSBwkJfgkBhyohjO8%3D

Bad title, abp don't replicate.

@dire burrow

[dire burrow]

I think your onto something if you look at the Spawns owner node it states that it can be left empty but is used for replication

[dark parcel]

What ever you want to replicate need to exist outside your anim bp. And the anim bp read the replicated var

[lost inlet]

isn't this guy one of the known unreal shitposters?

quantity over quality

[dire burrow]

yeah i did try with code like mes guide that one ifirc is oriented to a animation montage

[dark parcel]

@lost inlet probably, I seen few of his stuff, kinda don't like it.

He does rpc inside GA too

[dire burrow]

hm maybe i should revist that... but yeah i dont know what to give the owner node of the bp spawn i guess giving it something would just give me peice of mind

[dark parcel]

For the record I am just warning the title of the video you watch. Not really saying you should learn from it.

what ever you need to replicate need to exist in an actor that is replicated.

[dire burrow]

yeah

i think i've watched it couple of times and dawn the same conclusion its not helping me with my issues

but yeah i just plugged in getparentactor to owner of the bp spawn

[lost inlet]

Code Like Me should absolutely be one of the channels you ignore

UAnimInstance isn't replicated in any way, so yeah, the only option is to read properties from the owning actor which is replicated

And if it's something that infrequently updates look into using delegates for that (event dispatchers for the BP term)

[dire burrow]

ok fingers cross this might be why... might be the case if you spawn and it doesnt have a owner actor it wont replicate by default

[lost inlet]

Actors without an owner replicate

Relevancy is the main thing to contend with

[dark parcel]

Just need to mark the actor as replicate

[dire burrow]

oh ok

[lost inlet]

As previously mentioned, client/server RPCs need an owner. Client RPCs have one intended recipient, the owner

You can only send a Server RPC as the owner of an actor

But multicasts and replicated properties have no such requirement

[dark parcel]

Can't help but to say multicasting is most likely the incorrect move for what ever you want to do there.

If you need to rotate the head bone then use replicated variable.

Client can send Server rpc then server will replicate the value to all other client.

Other clients can just set the incoming value with interpolation on tick.

[lost inlet]

As a general rule of thumb. Multicast = one shot events, property with OnRep = any persisting state changes

Though the OnRep is optional, that depends if you need to do something to react to the state change

[formal turtle]

is it normal for server gameinstance to have variables reset when using servertravel?

e.g valid pointer ref becomes invalid

/ struct becomes empty etc

[chrome bay]

depends what they are I guess

[formal turtle]

FStruct SelectedCharacterData
1.) character class reference
2.) 'get player name' fstring var

EventOnPostLogin fails to retrieve the values as it says array is empty

seamless travel is enabled

[chrome bay]

I mean nothing will just randomly get nuked, unless it's referencing an actor in a UWorld or something

[formal turtle]

think i figured it out

I dynamically swap the gamemodes thus OnPostLogin wasnt being triggered after switch

[exotic wasp]

I think you're arbitrarily moving functions somewhere they probably shouldn't be in that case

[prisma snow]

We have an Actor whose only purpose is to do RPCs, but it's an RTS game so the amount of player input is very limited compared to another type of game.

I would not overcomplicate yourself and just have a nice ordered section in your actor(s) for netcode if that makes sense.

[hazy scroll]

Hi guys, I have a strange replication issue. I tried so many thing but still not working.

The clients cant see the particles. Test on Client with print string: "Server: Before MCExplodeEffects" and "Server:MCEffects".
So the clients dont execute the multicast code, but the server calls the mutilcast correctly.

SR_Explode is called via this structure: CharacterBP{ Primary(Run on Server)->ActionShoot-> } GunBP {Shoot->HitTrace->} ExplosiveBP{ SR_Explode}

Without class advices:
Primary(Run on Server)->ActionShoot->Shoot->HitTrace->SR_Explode

[blazing bear]

You're saying multicast isn't even received by the clients?

[hazy scroll]

yes thats right. They printing nothing. And every important actor is set to replicates

[blazing bear]

Does the server existing actor firing the multicast exist on the clients as well?

[hazy scroll]

I thing so. I've put an print string on event begin play of the ExplosiveBP Actor. Server and both clients are printing. And the SR_Explode is firing later

[exotic wasp]

Dumb question, but is that actor replicated?

[hazy scroll]

yes thats why I am so confused

[blazing bear]

Running it on the beginplay is what's causing you the problem

[hazy scroll]

The SR_Explode is not firing on begin play only the print string to check

The SR_Explode is firing after I press the left mouse button

[exotic wasp]

https://tenor.com/view/omni-man-omni-man-are-you-sure-are-you-sure-invincible-gif-3935116808772397515

[blazing bear]

How is the actor spawned?

[hazy scroll]

In CharacterBP with the Event Primary(Run on Server) (first picture). And the use function looks like that (second picture)

Oh and the event use is not replicated(Forgot to deactivate it)

Otherwise it would be double run on server

[exotic wasp]

So that LineTrace is being run on the server?

[hazy scroll]

yes

and everyone sees the claymore after placing. But when I shoot it the clients dont see the particles

[exotic wasp]

Is it being destroyed on the server?

[hazy scroll]

yes

[exotic wasp]

So

My guess is that it's getting destroyed and that replicates

And the explosion effect never plays because the actor is destroyed

Most networking is batched every frame IIRC

Can't necessarily guarantee order of anything either

[hazy scroll]

Oh that was exactly the problemxD

thanks

[exotic wasp]

As to how to actually fix that I'm not sure

Could just play the explosion effect when destroyed

[hazy scroll]

Or the easy but bad way is to make a short delay or timerxD

[blazing bear]

Try using the OnActorDestroyed delegate

Otherwise you could: Make actor invisible --> Spawn particles --> Destroy

[exotic wasp]

That would likely still result in it being destroyed before doing anything

[exotic wasp]

This is probably the best way

Though how you would decide if it's destroyed because it's exploding or just normally is an issue

Someone with more experience here probably has input

[blazing bear]

Or just don't rely on the actor to spawn the particles in the first place

[sinful tree]

For something simple like this, instead of immediately destroying the actor, you could use SetLifeSpan and give a brief duration that could be slightly longer than your particle system plays back for, just ensuring that it should play and finsih on any clients as well and when the lifespan is up, it'll get destroyed as normal.

[thin stratus]

It's actually quite normal to destory after a delay.

You usually set some OnRep boolean or so that indicates that the actor is basically about to be destroyed.

In the OnRep function you turn off visibility and collision.

And the Server then destroy it after the delay or via lifespan.

If you want to be fancy and you have C++ access, use "TearOff" and "TornOff".

That cuts off replication and leaves it in the hands of the client to destroy the actor.

[final bluff]

hey yall!
I'm running into niche situation and curious how other people dealt with this.
Player A creates a session -> Player B starts to execute join session on Player A's Session -> Player A Destroys session (While Player B is joining)-> Player B joins in on Player A's Level

This leaves the 2 players together connected in the same level replicating to each other but Player A's Session is destroyed and Player B's Session is valid.
How could I stop Player B from finishing the join session execution if Player A has already destroyed the session? Or alternately is there a different way to handle this situation?

(Btw , when player B joins I do not see any issues , all tested replication and functionality still works as expected BUT i feel like there are forsure potential issues if moving forward because one player is in session and the other is not lol , Also if player A destroyed session , the whole point is to not allow other players to join and another player joining in on a "Closed Lobby" is an issue .

[thin stratus]

Are you aware of the fact that "Session" and "Server Connection" are two separate concepts?

A Session is simply a set of information. Such as the name of the server, how many and which players are in it, how to connect to the server etc.
The Server Connection can be established without the need for a Session. And a Session can be created and joined without ever connecting to the Server.

If you want to stop a player from connecting, then you have to stop the part where the Server is listening for incoming connections. Usually done by simply performing another HardTravel to any level.
You can probably also somehow stop listening via C++ without traveling.

Destroying the Session doesn't stop a player that is already connecting from doing so.
You can, however, use PreLogin/Login/PostLogin in the GameMode to potentially check if the Server still has an active Session and if not deny the player.

[echo pasture]

So I setup a test project for this and... seems to be true, widgets don't stay anymore during seamless server travel. Would be nice if they updated the docs

[echo pasture]

And I still don’t understand the alternative solutions either. GetMoviePlayer->PlayMovie doesn’t seem to work

[potent tundra]

hey guys, this is a multiplayer question but it also involves some physics

im trying to make a system where players can push a ball around smoothly. Ive been experimenting with the predictive interpolation and resimulation settings, applying the impulse on the client first and then again on the server, however this is pretty unstable and i can see correction happenning. I really need the balls physics to be smooth for all clients so, what do?

[final bluff]

yes! through testing i'm understanding that a Session and a Server Connection are 2 separate things. Still trying to fully understand the difference between them though and how users interact with the 2. Viewing the session as "simply a set of information" helps differentiate the 2 concepts.

Like you are saying , i need to stop the part where the Server is listening for incoming connections . I have tested and see how performing a hard travel could accomplish this. But for my scenario i need the host to stay in the same level , so im looking for a way to stop listening without traveling.

[echo pasture]

I know some devs on rolled out also needed this feature and they simply subbed in a deterministic physics engine. perhaps you could also sync the applied forces. also I've observed it to not be so bad, so in general i'm not sure

there's also ways to make it smooth itself even during correction

[prisma snow]

do you need super inmediate response?

[final bluff]

I see there is UGameInstance::EnableListenServer , and i have been able to successfully use this during runtime to change a level into a listen server. But in a previous post you mentioned using this could lead to some issues. If there is a clean way to Stop Listening while staying in the same level , and then go back to listening when needed WHILE still staying in the same level , then i think i will be a able to get my desired outcome.

[final bluff]

^

[echo pasture]

Hack workaround to get the seamless servertravel umg loading screen for 5.4 and 5.5 (since it was removed) for others interested:

it can also be done very easily in c++ as well, I just did in blueprint since it was faster

I just didn't wanna fuck around with MoviePlayer and Slate or Lyra loading plugin no more

[dark edge]

Have you messed with the default correction settings?

They are very strict by default

[brazen anvil]

When adding another variable that needs replicated as much as something like actor rotation, is there a preferred method of doing so?

[potent tundra]

No i havent, good shout. What settings in particular?

[dark edge]

Error per distance, error per angle, max restored error, etc

Turn on net.showcorrections 1 to see the error and the accumulation

[final bluff]

Thanks for the response. Kicking the player if server is not in session on PostLogin is working right now . Will continue to test it out 🙏

[dark edge]

If they're replicated

[fickle pewter]

Hello everyone. Can anyone help with advice, a tutorial, or a link to documentation on how to integrate a dedicated Steam server into a project? I want to implement a quick matchmaking system like in CS:GO or Dota 2. No matter how much I search, I can’t seem to figure this topic out.

[nocturne quail]

Can i use NULL SUBSYSTEM so the clients around the globe can connect to the server?

[lost inlet]

the null subsystem has no backing online service, it's just directly connecting to other machines through an IP

[nocturne quail]

this is what i want, will port forward and give them the ip address

ofcourse i have a virtually sets up teh vpn to hide my real ip and packet filter behind firewall

[nocturne quail]

so i have completely disable the steam platform right?

the ID 480 is origin limited thats why i will use null if it will work

[torpid crest]

Hey guys, I’m having a weird issue with server travel via Steam.

I have both the listen server and a client in the lobby. When the server initiates a server travel, the client joins the session but doesn't have an assigned player controller.

However, if the client closes the game and rejoins the session, they do get a controller 😄 Any idea what might be causing this?

[dire burrow]

man i still remain firmly stuck with head/spine rotation

[dire burrow]

I you using the get player controller by index?

i know that can cause some weird controller assignment issues

[torpid crest]

Player pawn

Should I spawn player pawn in player controller begin play?

[mystic estuary]

Don't use delays in multiplayer for initialization purposes. For this one, use OnControllerChangedEvent

[dire burrow]

some usefull tips regarding how not to deal with controllers in this
https://wizardcell.com/unreal/multiplayer-tips-and-tricks/#1-get-to-know-the-game-framework-objects-online-wise

[dark parcel]

never use delay for initilization. You can't assume player's ping.
Read what Tony said

[torpid crest]

so this one instead

[dark parcel]

in short your issue may derived from the fact that the pawn isn't possesed by the client yet but your begin play is already called.
So naturally your is LocalltControlled check will fail

for the most part OnBeginPlay isn;'t the right place to init anything on multiplayer either.

[dire burrow]

ah yeah he needs the "wait for me" stuff

[dark parcel]

if you need to do something when the player already have a character then use events that are called after possession.

For blueprint, perhaps Receive Controlled Changed

if you do C++, I use AcknowledgePossession event

[dire burrow]

yeah if you have say blueprint and it wants to talk to the character for example but its not ready yet .... i have seen this used in various things

[mystic estuary]

Yes

[mystic estuary]

That's extremely lame

[dark parcel]

Lyra did it for player state. I don't know why, but yeah altough I don't have anything remotely like this.
Maybe okay to do as last resort.

imo repnotify does the job most of the time.

[mystic estuary]

IMO it's alright doing that in places where you don't have a good hook or you're simply lazy to make one, but in base things like PC/PS/CH you have enough entry points to make it event based. I sometimes do that in UI when it's something like GS or a component that might not be replicated in time, but I never do that when it's related to some important initialization

[dire burrow]

yeah i personally switched to character manager interface so my other bps can be informed when the character initilized all its crap

but yeah i didnt know there was a specific event till recently so probably going to deprecate it in my stuff

but i feel like the C++ interface does perform faster then graphs.. it might just a be a placebo effect not sure

just hate the bitchyness of the syntax of UFUNCTION & CO + C++ bitchyness Class inhierentence layered on top of that

only made worse when you want to test replication

[prisma snow]

idk which kind of performance we're talking about but always profile if that's relevant

[dire burrow]

yeah i duno it just seems cleaner and when the multiplayer clients run.. looks kind of smoother then what i was seeing before in the animations department

I just want to change the entire animation bp to pure interfaces

no more sequence

i think the performance gain might be simply that A] my headrotation was a macro in the sequence and B] now its not perminently in the sequence being checked if is valid blabla bla .. now it just gets straight to work only when it needs to work

also i learned macros are slow

do not use them for functionality

[prisma snow]

???

[prisma snow]

That's super nice, but that's not necessarly related to performance, it's maybe just different logic

But in general the only way to assess performance is profiling and anything other than that is very subjective and likely even wrong.

[dire burrow]

yeah totally you will never know which exact functions are chewing the cpu and memory

with out doing that

something like chatbot gpt ha.. will probably sit there and do all the profiling for you in realtime in the future

but yeah just running my editor in full debug mode slows me to a crawl ... these days

which is a good enough indicator in its own right

like you know how debug mode sits there with the cpu graph/memory load etc

[lost inlet]

it'll be nice when VS's on-demand de-optimisation thingy will be more commonplace

[prisma snow]

I usually profile on a dev packaged game if I want close to real data

[dire burrow]

but yeah i am super bummed that we cant just simply (in graph/editor) replicate the animation bp's variables .. or replicate the controlling bp's variables ... its just like some bug stops this from happening with metahmans

they dont respond to the C++ replicated headrotation/mousepositionvector

its just a dead end

but yeah if these specific animation bps can be "replicated"

then why dont they? with out any thing else

why are they even there then in the animation bp?

if the anim bp cant "bReplicates"

[dark parcel]

for the 10th time, anim bluerpint don't replicate

so any variable you declare in anim bp will not replicate

doesn't matter if you mark it as replicate or not

[dire burrow]

So the variable it just doing that because it doesnt know any better but then to expose the replication properties?

[dark parcel]

This interface is also confusing, I don't even know how this get executed

[dire burrow]

its a interface another bp calls it

[blazing coyote]

anyone know how to make the cloth physics work only on clients and not replicate the stuff from the server? I can see lot of issues and hitches on the client when watching serwer cloth.

[dark parcel]

your bp shouldn't even know it's anim instance

it's the other way around

anim instance read from it's owner

[dire burrow]

its totally can do that though .. it just gets the mesh from the character then gets the animation instance then casts it to ABP_MANNY then runs the implemented interface event function

[dark parcel]

My suggestion is to try and replicate pitch first, then go back to your system afterward.

[dark parcel]

just because you can, doesn't mean you should.
This is terrible practice.

Just like widget, anim instance job is to read from it's owner.

the owner shouldn't need to know about the anim instance

[dire burrow]

ok i just figured that if i want all my movement code and various systems centralized into 1 kind of portable bp with out having to worry about screwing up my mh character all the time i would try and keep every thing inside 1 commander bp

[dark parcel]

What? No

[prisma snow]

yeah in general just replicate core game logic and drive the rest of the state (anims, visual effects, widgets, etc) from the gamepaly data locally

[dark parcel]

yeah, just replicate what's needed. I would simulate the result locally if possible.

[prisma snow]

yep

[dark parcel]

some things don't even need to be replicated.

[dire burrow]

yes exactly

[dark parcel]

e.g a character head looking at the closest character.

[dire burrow]

half of every thing i am doing with out a single shot fired which is what frustrates me with this

[prisma snow]

yeah most of the time it's not very important or can just be locally calculated, saving precious kbs of internet data and your sanity

[dire burrow]

as i said if i replicated the variables in the movement component with just "Replicated" it shouldnt of not worked for the head rotation

[dark parcel]

Why would it?

marking a variable as replicated doesn't magically rotate the head.

You have to implement the logic, such as setting the head to rotate to the value every frame with interpolation (for example).

[dire burrow]

the rest of the "replicated" stuff quiet happly replicated the leg ik / rotation etc and stuff

[dark parcel]

I think you are assuming a lot of things

why would leg ik be replicated at all?

it shouldn't be and if you are using the foot ik in third person template then it's not replicated at all

[dire burrow]

except that it literally does with the MC

[dark parcel]

they are simulated locally.

no machine tell other machine what state their foot is at

[dire burrow]

its a weird one i can see there foot ik working

[dark parcel]

why wouldn't it be. They are calculated locally

0 networking

it steps on the floor when it steps on the floor

also if you are doing MC for anything stateful, that's already incorrect

head rotation deffinitly not something that involve multicasting

if you see a tutorial that does that then just run away.

[dire burrow]

yeah defo not using Multicast

just replicated via C++

[dark parcel]

BP or C++ wouldn't be much different in this regard

except if you use OnRep

[dire burrow]

UPROPERTY(ReplicatedUsing = OnRep_MousePositionVector, BlueprintReadWrite, Category = "Movement")

the only reason this is so that is so the C++ side can see the gravitycontroller and talk to BP's (only the UOBJECT you precsribe)

or IOBJECT but not had much sucess with that side of it 💦

[prisma snow]

why are you replicating mouse position at all

[dire burrow]

so that all the other clients know about the head rotation in the animation graph?

[dark parcel]

that value is already replicated

use GetBaseAimRotation

that's basically the ControlRotation iirc.

aka, where the player is looking at.

Just assuming your game is third person.

[dire burrow]

so then all i should do is bypass local logic and allow server logic or something

[dark parcel]

You don't need to replicate the head rotation.

For each of the autonomous proxy, you get the value of their control rotation. Then you just make the head face that way.

If LocallyControlled
GetControlRotation -> Rotate your head
Else
GetPawn-> Get base Aim rotation -> RotateHead with Interp

why interpolate? because there's no way you can send update fast enough to a computer with high fps. So to avoid snapping effect, you interpolate the current rotation to the target rotation.

void AAGPlayerCharacter::AimOffset(float DeltaTime)
{
    if (IsLocallyControlled())
        AO_Pitch = GetBaseAimRotation().Pitch;
    /** Gets proxies characters in server that server don't control. */
    if (GetBaseAimRotation().Pitch > 90.f && !IsLocallyControlled())
    {    /** map pitch from [270, 360] to [-90, 0] because of compression */
        const FVector2D InRange(270.f, 360.f);
        const FVector2d OutRange(-90.f, 0.f);
        float L_AO_Pitch = FMath::GetMappedRangeValueClamped(InRange, OutRange, GetBaseAimRotation().Pitch);
        AO_Pitch = FMath::FInterpTo(AO_Pitch, L_AO_Pitch, DeltaTime, 15.0f);
    }
    else
    {
        AO_Pitch = FMath::FInterpTo(AO_Pitch, GetBaseAimRotation().Pitch, DeltaTime, 15.0f);
    }
}

[dire burrow]

yeah not it was a thing in gravity/sphereical dynamic mesh's that made me lerp the head rotation becase the cameras attached the head sort of thing

so yeah it just made the camera some how more stable its a long story but yeah just a side not that its what makes them able to run forward head locked then return to unheadlocked then able to look any where and go back to fps mode

its kind of a little hybrid in that respect

but yeah the faster you run etc the more the 3rd person zooms out for the the tiny sphereical world