multiplayer

[reef bison]

[sharp mason]

can you tell me why ?

[reef bison]

it didnt work very well for me because once the root motion ends the server is still moving the character on their side so re-enabling smoothing will cause a small rubberband that was bothering me

it probably wont matter for you because you wont enable it again

[nova wasp]

yeah I think ideally it's "done with" after this

[sharp mason]

oh i see

[nova wasp]

I suppose the client could tell the server with an rpc "I am 100% moving them this way" and the server could correct... itself

as long as it started in a somewhat similar location (validation not needed, but it might help to keep it not too far off)

It would be difficult to make a perfect 1:1 simulation on client and server if they are moving quickly but sending results should be somewhat doable... it might force delaying things a bit on other sides though

or it might be good enough

[reef bison]

even if the client tells the server the correct direction or the exact trajectory you still have a rubber band on the end because of latency

the movement stops first on the client and it will reenable the server corrections, so it will rollback to the server's position relative to the latency of the client from when they started the attack

i tried a lot of different solutions but couldnt come up with anything that didnt look very brittle

like adding a delay to the server correction so it could catch up with the client

i might give it a try again in the future

[sharp mason]

for this case i think it wont be a problem, but i can see the problem in the future when not using death root motion animations

well for now it worked like a charm

@nova wasp @reef bison thank you very much guys, i guess the next problem will be as you said to enable it again, but that is a problem for another time

[reef bison]

np

i just thought about 2 approaches that i didnt try that might work

the knockbacks in my game are very small so instead of playing it on the server i can just "teleport" the enemy to the final destination and play it on the client, so there would be no corrections at the end since the server is already in the final position

or i could speed up the server version

or i can make the server stop sending corrections to the predicting player while the rms is playing, once it ends both versions will be in sync

none of these sound robust so i will need to do some tests with different scenarios but thats gonna be far in the future

[sharp mason]

well i hope that i can fix this the same way i fixed my player dash using compressed flags and SavedMove_Character

[loud drift]

So I've been trying to enable multiplayer with steam. Changed the ini, added the steam plugins required, even compared my setup to another devs working setup. Have been having the most ungodly time trying to figure this out, I'm looking for someone that has genuine first hand experience with this that would sit down and work with me on it. This is not me asking for a handout I've done the work over and over. I think that I may just be missing one piece left of the puzzle just can't figure out what it is. Anyways thank you

[nocturne quail]

any diea why is UAccessoryItem::OnRep_bEquipped() not guaranteed for late clients?
first late client don't see accessory attach to weapon, but second late client can see, again third late client can't see it...
its like random desync...

void AWeapon::UseAccessoryItem(UAccessoryItem* AccessoryItem)
{
    if (!IsValid(AccessoryItem) || !AccessoryItem->SupportedWeaponTypes.Contains(WeaponType) || !WeaponOwner->GetInventoryManager()->FindItem(AccessoryItem))
        return;

    if (!HasAuthority())
    {
        ServerUseAccessoryItem(AccessoryItem);
    }
}

void AWeapon::ServerUseAccessoryItem_Implementation(UAccessoryItem* AccessoryItem)
{
    if (!IsValid(AccessoryItem) || !AccessoryItem->SupportedWeaponTypes.Contains(WeaponType) || !WeaponOwner->GetInventoryManager()->FindItem(AccessoryItem))
        return;
        
    AccessoryItem->UseAccessory(this);
}

void UAccessoryItem::UseAccessory(AWeapon* NewWeapon)
{
    AttachmentOwner = NewWeapon;
    OnRep_AttachmentWeapon();

    if (NewWeapon && NewWeapon->HasAuthority())
    {
        if (NewWeapon->GetEquippedItems().Contains(AccSlot) && !bEquipped)
        {
            UAccessoryItem* AlreadyEquippedItem = *NewWeapon->GetEquippedItems().Find(AccSlot);
            AlreadyEquippedItem->SetEquipped(false);

        }
        
        SetEquipped(!IsEquipped());
    }
}

void UAccessoryItem::SetEquipped(bool bNewEquipped)
{
    bEquipped = bNewEquipped;
    OnRep_bEquipped();
    MarkDirtyForReplication();
}

void UAccessoryItem::OnRep_bEquipped()
{
    if (!IsValid(Weapon))
        return;

    if (bEquipped)
    {
        UseActionText = LOCTEXT("UnequipText", "Unequip");
        Weapon->EquippAccessory(this);
    }
    else
    {
        UseActionText = LOCTEXT("EquipText", "Equip");
        Weapon->UnEquippAccessory(this);
    }
}

is it a race condition between two onreps?
OnRep_AttachmentWeapon(); and OnRep_bEquipped();

[lost inlet]

you cannot rely on OnReps being called in any sort of deterministic order. in addition, if its value matches the value from the CDO, you won't get an OnRep at all

[quasi tide]

You can change that behavior though 👆. For the DOREPLIFETIME macro, use the version that has the _CONDITION_NOTIFY afterwards. Then the last arg is the rep notify behavior. So you can use REPNOTIFY_Always in this case.

[nocturne quail]

Tried already but no luck, behaviour is same

some late clients can see update and some not, for some late clients weapon is not valid

[quasi tide]

It won't change the order of the onreps - just makes sure that they always fire

[nocturne quail]

switching to fast array, maybe it will solve it

inventory items are already fast array

this equipment state I think also needs fast array, I found they are very reliable in unreal replication

[dark edge]

race condition with Weapon

just have both the weapon and accessory equipping onreps check for the other.
Weapon shows up 2nd, it looks for accessories to attach.
Accessory shows up 2nd, it looks for Weapon to attach to.

[nocturne quail]

it will work but I am switching to fast array which will absolutely eliminate these desyncs

[dark edge]

All you are guaranteed is that the replicated state will eventually be consistent between server and client. You need to design your code such that it doesn't matter which order the state variables are updated in.

Fast array of.... what

[nocturne quail]

making a seperate EquippmentComponent which will work with inventory component

so inventory replicate items, equippment replicates the state of equipped/unequipped [slots]

[dark edge]

You're still not guaranteed any ordering

You'll still have the same problem, how to handle an accessory being equipped before a weapon is.

[nocturne quail]

equippment state for weapon/accessory both will be handles by the the equippment component

I had this issue before with character cosmetics, I solved it by fast array, but it was not done cleanly like in a dedicated component

had to refactor the whole inventory system once and forever

[nova wasp]

as mentioned before you are still early-returning on things not being valid with 0 ensures, logging or ability to breakpoint the line

if things can randomly be null there then you must re-apply the same thing once the intended pointer arrives

having random separate properties = more and more permutations of random properties arriving

if you ask "why doesn't this work"

and I see this

my first question is going to be... why should I ever trust that you actually made sure this wasn't just returning here

remember: we can only see what you show us

as I explained last time please at least add ensures/error logging in here

there is no need to guess ever

but yeah I agree with adriel

if the order is random, the last one must apply it (which means both must check)

[nocturne quail]

I maked sured if weapon was invalid by logging it was not a guess, and it was invalid for some late random clients

[nova wasp]

remember: we can only see what you show us

[nocturne quail]

now I have a centralized EquippmentComponent which does handle this using fast array, and works fine

[nova wasp]

nice to hear... sorry if I was digging up an old question

I didn't know it all changed

[nocturne quail]

I mentioned it before that weapon was invalid

[nova wasp]

I don't see where but fair enough

was it like a week ago or something?

[nocturne quail]

its fine to dig old questions, for me every opinion is a lesson to learn from

[nocturne quail]

a day ago

[nova wasp]

ah fair, I missed it

[tardy fossil]

is it normal for a actors rotation on a client to be off by like 0.001 degrees? even with net.QuantizeActorRotationOnSpawn=0? is that just an unavoidable FP precision thing?

[nova wasp]

I would assume it's probably just quaterion/rotator conversion not reproducing the exact result both ways here but I can't see what is actually being applied

[tardy fossil]

oh.. also the server is using a map file to read position / rotations for actors its spawning.. and i just noticed the ones that dont completely match up have angles with wayyy too many decimal places, like 45.675346772888 lol i think i just need to get the server to round to the nearest 2 decimals

[nova wasp]

scene components do not use euler rotations internally

they are all quaternions

the euler rotation that shows there is to make it human readable

it is not the real final value

a lot of the gameplay framework is converting back and forth

[tardy fossil]

ah yea, and it seems like the server converts back to rotator to send spawn rotation

too bad GetActorRotation isnt a virtual or i could just have that return the original euler angles it got from the map file

[nova wasp]

you CAN send quaternions over the network but it's generally easier to send euler rotations as they will tend to be much easier to compress as most gameplay actors will tend to change yaw only etc and have zero tilt/pitch

[nova wasp]

you could probably do something evil and zero the rotation to fool it into packing it down more and then use a separate property

you should have a decent amount of control in actor channels but I'm more familiar with the iris route here

another option could be to sort of... scew up the rotation on the "ground truth" in the exact same way just to make them similar

effectively quantizing on both sides and hoping it looks okay is doable for some things

[tardy fossil]

i think quantizing on both sides is the way to go tbh.. the angles dont need to have lots of precision.. just match up

[nocturne quail]

using fast array we really need to call this manually?

    if (UActorChannel* Channel = Character->GetNetChannel())
    {
        Channel->MarkDirty();
    }

I think it automatically handles it, no?

AI is confusing me 😄

[nova wasp]

for what?

adding an element? changing one?

I can't see what the fast array is here, I can't seer where you add an element to it

read the fast array header for examples

I don't think you need to ever manually mark the channel dirty but you cannot ignore following how to use a fast array

read engine examples IMO

they will show how you mark the array dirty directly for adding, removing and updating elements

If you find it's not being sent down maybe ForceNetUpdate() on the actor but I don't really think you should have to

one engine example : FActiveGameplayCueContainer::AddCue

note how they call MarkItemDirty

they add to the tarray in a funny way but MarkItemDirty is the important part

[nocturne quail]

EquipItem() calls SetSlotItem(), which directly modifies the EquipmentSlots.Items array and calls MarkArrayDirty()

[nova wasp]

I can only see the code you show me

[nocturne quail]

void UEquipmentManagerComponent::SetSlotItem(EEquippableSlot Slot, UEquippable* Item)
{
    if (!GetOwner()->HasAuthority())
        return;

    if (FEquipmentSlotEntry* ExistingEntry = EquipmentSlots.FindEntry(Slot))
    {
        ExistingEntry->Item = Item;
        EquipmentSlots.MarkItemDirty(*ExistingEntry);
    }
    else
    {
        FEquipmentSlotEntry& NewEntry = EquipmentSlots.Entries.AddDefaulted_GetRef();
        NewEntry.Slot = Slot;
        NewEntry.Item = Item;
        EquipmentSlots.MarkArrayDirty();
    }

    UpdateCacheFromSlots();
}

[nova wasp]

FindEntry?

is this an array?

I can only see the code you show me... I have to go though

[nocturne quail]

TArray<FEquipmentSlotEntry> Entries;

[nova wasp]

I would suggest wrapping the add/remove calls in the fast array struct to avoid mistakes

[nocturne quail]

FEquipmentSlotEntry* FindEntry(EEquippableSlot Slot)
{
for (FEquipmentSlotEntry& Entry : Entries)
{
if (Entry.Slot == Slot)
return &Entry;
}
return nullptr;
}

[nova wasp]

I believe this is incorrect... see my previous message

reading the engine header for fast array serializer is more or less mandatory here

the engine is trying to help you

let it help you

fast arrays are frustrating because you don't get a lot of direct feedback on what is wrong when things fail but you have direct examples of how to use this

[nocturne quail]

True, and I like them for some reasons
reliable in replication
don't have issues reading a sub container inside another replicated container

[fallen fossil]

Where should I store local data from Game state? I want to store Set of actors, so its not replicatlbe and I am thinking of storing in gamestate anyway :d

and Rpc add remove each one of them

[jovial ravine]

I think you would store them in GameInstance.

[dark parcel]

Store it on the component that is responsible for loading / saving.

It doesnt matter if you store it in a repkicated object, just make the data not replicated.

[nocturne quail]

Server:
1. Pickup creates Item (Outer = Pickup)
2. Inventory::AddItem creates NEW Item (Outer = Character)
3. EquipmentManager::EquipItem stores Item in Fast Array
4. Item->MarkDirtyForReplication() increments RepKey
5. EquipmentManager::ReplicateSubobjects sends the Item to clients
6. Fast Array replicates Slot → Item reference

Client:
1. Receives Item via ReplicateSubobjects (Item object created on client)
2. Receives Fast Array with Slot → Item reference
3. OnRep_EquipmentSlots called
4. Item pointer is VALID because it was already replicated!
5. AttachGearVisual called successfully

this rep work flow is correct?

[nocturne quail]

for everyone in future will have this issue, the fix is very simple!

It was actually an issue because `onrep` gets called before `beginplay`...
I was initializing the slots tmap on beginplay
when onrep get called for late clients, the map is empty and they gets no visual update!

solution is to initilize the tmap in PostInitializeComponents()

[nova wasp]

I appreciate you sharing your solution

[slate phoenix]

Hello everyone,

How can I disable these warnings? Normally, HISM instances don’t replicate and that’s why this warning appears. However, I implemented my own replication system.

It doesn’t affect the game, but these warnings keep appearing in the logs, which makes it hard to track other logs properly. Sometimes it keeps spamming, sometimes it stops.

How can I disable this warning?

[fossil spoke]

If you added your own replication system, how on earth could we possibly give you guidance?

You could turn off LogNetPackageMap as a log category...

But thats just masking the issue.

[slate phoenix]

When I close the actor's replica section, the warnings disappear, but then the replica section stops working.

[nova wasp]

if you actually want to solve the problem you can see why the log is created and try to make that codepath not happen

[nova wasp]

what does "close the actor's replica section" and what exactly is your own replication system?

[slate phoenix]

Actually, it might make more sense to ask this:

I have an actor that controls my HISM instances. In the Details panel, there is a Replicates checkbox. When I disable it, the warnings disappear, but then I run into replication issues.

Is it possible to handle replication while Replicates is set to false?

[nova wasp]

you are using unreal's replication system

you are not making your own replication system if you are using unreal's replication system

[fossil spoke]

🤦‍♂️

[nova wasp]

it's fair to say you are working on netcode I suppose but this is not a new system for replicating objects and properties

[fossil spoke]

Might be a language barrier maybe?

[nova wasp]

yeah I assume so

you can absolutely still replicate information about something without replicating that specific object if the information to get the same result is on both sides

[fossil spoke]

The HISMs themselves do not support replication.

As Megafunk mentioned just now, you would setup a mechanism for replicating changes to HISMs over the network yourself.

[nova wasp]

you could send information about where the instances are on a separate object maybe?

I don't know what the goal is here I suppose

[slate phoenix]

Sorry, my English is not very good. I’m writing using translation. I’ll take proper notes and review what you said.

And I will write to you in a clearer way (if I can’t, I will try again).

[nova wasp]

that's okay, I get it

I think it would help to describe what you are trying to make (what are these HISMs doing?)

[nocturne quail]

Thanks

[slate phoenix]

I'm building a sandbox MMO similar to Growtopia. The world is made of blocks (50x50x20 grid). Each block type is rendered using HISM for performance. When a player breaks or places a block, I use Server RPC → Multicast RPC to notify all clients, and each client updates their local HISM instances. The block data is stored in a MySQL database via a Node.js backend. So the HISM actor itself doesn't need to replicate - only the block changes need to be synced, which I'm already doing through RPCs. The warning appears because the HISM actor has Replicates enabled, but I'm not sure if I can safely disable it without breaking my RPC chain.

3D game

[fossil spoke]

Of course he is

[nova wasp]

you probably want to send RPCs with a way to point to each HISM instance locally

for example you could send their coordinates and map to that on the local client

you don't need to send the RPC to the HISM instance, it could be something on the gamestate or whatever

it just needs to represent the information

[slate phoenix]

Actually, I'm already sending RPCs through the PlayerController (BP_TopiaCharacter), not through the HISM actor itself. When a player breaks/places a block, the PlayerController sends a Server RPC, which then calls a Multicast RPC. Each client receives the coordinates and block type, then updates their local HISM instances. The WorldGenerator actor only handles initial world loading and HISM management. So the RPC chain doesn't depend on the HISM actor's replication. My issue is just that having Replicates=true on the WorldGenerator causes these HISM warnings, but I need it enabled because some initial setup functions run through it. I'll try disabling Component Replicates on individual HISM components instead.

[reef bison]

a multicast rpc sounds like a bad idea, what about clients joining after a block is added/removed?

[nova wasp]

why keep the setup functions on them?

and yeah an RPC here is going to require having a way to track which version each client received

people who connect later will not see these rpcs unless they are given them somehow and you need a way to track that

it would probably be easier to use some kind of compressed byte buffer on some manager actor for each chunk imo

some things will probably be better off as normal "actors" in the world but for the regular blocks I don't see why it can't be like a fast array or something

[reef bison]

i use HISM instances for my folliage that can be added/removed, the HISM itself isn't replicated, i have a fast array with all objects for a specific chunk and i replicate this array instead. then each client "listen" for changes in the array and update their own local HISM

[nova wasp]

that's a great way to do it

[slate phoenix]

Our game logic is a bit different, but think of it like this:
There is one main server, and players can create worlds (like a public lobby that anyone can own). However, I cannot use child processes inside a single server in Unreal Engine, so each world runs as a separate process on different ports.
For example, if a player creates a world named "oyun123", anyone can create similar worlds.
Now, when player A joins the world "oyun123", I load all the blocks into RAM and keep them there. If players B, C, and D join later, I make sure they don't receive missing blocks while loading.
To handle this, I use both:

RPCs during gameplay
Additional validation while streaming/loading blocks
This ensures that players who join later do not experience any missing blocks.
Also, there can be 20–30 processes running at the same time, and each process represents one world.
My original question was just about suppressing the HISM replication warnings in the logs, since the HISM components don't need to replicate but the actor needs Replicates=true for RPCs to work.

I will research what you said; perhaps it will be helpful to me.

[dark parcel]

30 process running on listen server, what?

[summer spade]

Hey folks, anyone here a wizard with GAS and EOS in C++? I'm hitting some nasty network lag on my UE 5.6 project and can't quite figure out where my setup is going wrong. I've been looking at the Network Profiler but could really use some advice on optimizing GAS replication over the EOS NetDriver. Happy to share code snippets or logs if anyone has a few minutes to help me troubleshoot!(im not sure where the error is so posted it in the cpp channel)

[fossil spoke]

You are best to describe exactly whats happening.

Show relevant code.

[nova wasp]

why is it suspected of being network lag? what is the apparent behavior here?

[summer spade]

Okee here is a code snippet of a lunge ability (like apollo in deadlock), when i test it out using a eos lobby(using the dev auth tool) , the dsync is too high , but today i tested out the lobby with my friend and the dsync was even with the movement sometimes , so im not sure if its a ping issue or what is a better way to move forward

[nova wasp]

what does "the dsync is too high" mean

remember: we can only see the information you tell us

we cannot assume

[summer spade]

like the network correct and rubberbanding is too high

[nova wasp]

a short video might help

[summer spade]

will get one

[nova wasp]

take note GAS does not handle predicted cmc movement for you

it predicts abilities in a different component and the character saved moves sent to the server are not always going to be in the same packet as the information for this gas ability starting

https://dev.epicgames.com/community/learning/tutorials/ywD1/unreal-engine-best-practices-for-networked-movement-abilities-cmc

[summer spade]

i think i did setup my cmc with new flags and custom modes properly , but i havent thought about how it interacts with the packets

[summer spade]

this makes sense

[nova wasp]

see my previous message: we can only see the information you tell us

[summer spade]

yeah im recording a video😅 , sorry its hard for me to explain in text

[nova wasp]

I feel like leaving out you have a custom cmc setup earlier is not helping

that is context we kind of need to see if it is part of this lunge thing

have you tested this with network emulation in PIE?

what bandwidth rate, ping, loss etc was the EOS server seeing for the connection?

what tickrate were all peers running at?

https://dev.epicgames.com/community/learning/tutorials/ywD1/unreal-engine-best-practices-for-networked-movement-abilities-cmc#55rootmotionsources this section about root motion sources is probably relevant here

[summer spade]

yes i do have a custom cmc that didnt have any issues with PIE network emulation (even at bad preset) but im running into the error while i use rootmotion tasks in abilities
ping is 30-40 ,network connection tick is 0.02ms,outrate in bytes is around 3000.framerate on client was 75-80fpps

sorry bit new to solving and debugging network so thank you for your patience

[nova wasp]

I asked about the framerate on both client and server here

you have also not described if this is a listen server or a dedicated server setup

[summer spade]

its a listen server

server framerate is 80fps

[nova wasp]

is this one single dash or like 3 in a row here?

it looks like it fails to stay consistent at the beginning and end but I can't tell

is the ability happening at the right time on the server with the same state?

the end position of the first one seems to be way off which gives the impression that the timing there and target state is not being consistent

you are going to need to use debugging tools to compare what happens over time to see who was too early or late here.... I would suggest just visual logging the states over time during this and when each side decided to do something

[summer spade]

its three dashes in a row , but in a time frame , and its changes distance bassed on the timing of the input relase , yeah it doesnt stay consistenat end, so i should check if its the same state as it is in the server too. will try using visual debugging

[nova wasp]

also lots of early returns with no ensure/logging in the code so I can't even say if this ran on the server

null ptrs should not silently succeed

you do not want randomness here

    AZeroLockCharacter* Character = Cast<AZeroLockCharacter>(GetAvatarActorFromActorInfo());
    if (!Character) return;

stuff like this is very very bad because you will have a hard time following when this fails

you can't breakpoint this line directly afaik

at least ensure or log

    AZeroLockCharacter* Character = Cast<AZeroLockCharacter>(GetAvatarActorFromActorInfo());
    if (!ensure(Character))
    {
       return;
    }

you could castchecked to make it just crash, log etc

it is incredibly tedious to find bugs caused by these being null if it just goes on as if nothing happened

also not sure why the CMC is being set to movement mode falling directly... what was it doing before?

[summer spade]

cmc was set to flying at the start so player doesnt have gravity , should i not directly set it to a different mode?

or is it better to use a tag to change it so it replicates to the server to?

[nova wasp]

you are going to need simulated proxies to be aware of this somehow

I'm not sure if root motion movement can disable gravity here or not but if if not movement mode flying makes sense I suppose

also this is not related to the actual coding problem but I would recommend against naming things in your game after existing games directly if you intend to sell it

[summer spade]

im not selling it , its just a passion project , to learn more about networking

[nova wasp]

that's fair

[summer spade]

let me try and test out the things you have recommended to see if i can pin point the issue or better solve it myself,thank you for your help

will let you know how it goes

[nova wasp]

if it helps there is a replacement for the cmc (Mover) but it's very new and experimental

which makes it considerably easier to add new movement code in a predicted way

the CMC is solid though if you actually correctly using saved moves though

definitely read that article

you can indeed consider just... not correcting movement during this dash as well

if it proves difficult to do consistently

[summer spade]

sounds good, i have heard about mover will check it out too

[grand kestrel]

You'd be crazy to use it for production atm. Probably fine for a learning project

[thin stratus]

https://tenor.com/view/crazy-rubber-room-gif-10524477174166992043

[nocturne quail]

Server:
1. Pickup creates Item (Outer = Pickup)
2. Inventory::AddItem creates NEW Item (Outer = Character) Item->MarkDirtyForReplication() increments RepKey 
3. EquipmentManager::EquipItem stores Item in Fast Array
4. Item->MarkDirtyForReplication() increments RepKey
5. EquipmentManager::ReplicateSubobjects sends the Item to clients
6. Fast Array replicates Slot → Item reference

Client:
1. Receives Item via ReplicateSubobjects (Item object created on client)
2. Receives Fast Array with Slot → Item reference
3. OnRep_EquipmentSlots called
4. Item pointer is VALID because it was already replicated!
5. AttachGearVisual called successfully

step 4 in server is necssary?

according to my understanding the inventory already does ++repkey , and item should be available for clients

but late clients doesn't see it if I don't do ++repkey twice

UItem* UInventoryComponent::AddItem(class UItem* Item, const int32 Quantity)
{
    if (GetOwner() && GetOwner()->HasAuthority())
    {
        UItem* NewItem = NewObject<UItem>(GetOwner(), Item->GetClass());
        NewItem->UniqueID = Item->UniqueID;
        NewItem->World = GetWorld();
        NewItem->SetQuantity(Quantity);
        NewItem->OwningInventory = this;
        Items.Add(NewItem);
        NewItem->MarkDirtyForReplication();
        return NewItem;
    }
    return nullptr;
}

void UItem::MarkDirtyForReplication()
{
    ++RepKey;
    if (OwningInventory)
    {
        ++OwningInventory->ReplicatedItemsKey;
    }
}

bool UInventoryComponent::ReplicateSubobjects(class UActorChannel* Channel, class FOutBunch* Bunch, FReplicationFlags* RepFlags)
{
    bool bWroteSomething = Super::ReplicateSubobjects(Channel, Bunch, RepFlags);
    if (Channel->KeyNeedsToReplicate(0, ReplicatedItemsKey))
    {
        for (auto& Item : Items)
        {
            if (Channel->KeyNeedsToReplicate(Item->GetUniqueID(), Item->RepKey))
            {
                bWroteSomething |= Channel->ReplicateSubobject(Item, *Bunch, *RepFlags);
            }
        }
    }

    return bWroteSomething;
}

basically two different rep paths

InventoryComponent                    EquipmentManagerComponent
       │                                       │
       │ ReplicateSubobjects                   │ ReplicateSubobjects
       │ (checks ReplicatedItemsKey)           │ (checks Item->RepKey)
       │                                       │
       ▼                                       ▼
    Inventory Items                        Equipped Items (Fast Array)

[nocturne quail]

reworked the whole system, now only depending on Fast Array delta replication

no ReplicateSubobjects any more

[limber gyro]

if im playing with 2 players in the editor and 1 is the server and the other one is a client and i call "UGameplayStatics::GetPlayerController(GetWorld(), 0)" i will always return the PC of the server right?

[nocturne quail]

it will return the pc of the player it is called on

[limber gyro]

ty

[dark edge]

Barring any PIE shenanigans that returns the 0th PC on the local machine that calls it

for a client it's their own, for server it's probably the hosts but I wouldn't rely on it

[tulip kiln]

what is the best for replication . Launch character or * Apply root motion constant force* ?

[nova wasp]

for what?

[fossil spoke]

Bandwidth maybe?

Eitherway, seems like a bit of a silly question.

[charred island]

is there any way to instruct unreal to choose a new spawn point if its first choice presents a collision with an already spawned player?

i have 4 spawn points and very often it will just choose the same one for 2 players

[nova wasp]

I would suggest looking into the gamemode restart flow if you are using that

I can't think of an immediately obvious step that could look for collisions off of the top of my head but you should have a large degree of control over the start position with overriding things

it would be simple to add if there wasn't one there already that does what you expect

[charred island]

i had some code to handle this at some point but i think it was buggy

[knotty marlin]

the default behavior of FindPlayerStart in GameModeBase already does this, I believe it tries to spawn at an open APlayerStart first and if all of them are occupied it spawns at an occupied one as close as possible without colliding. You could override the process in your gamemode in any one of a few different places like FindPlayerStart or even SpawnDefaultPawnFor and implement your own logic.

[charred island]

yeah thats what i had to do

for whatever reason the default behavior was not properly finding the next open spawn point if one was occupied

maybe my collision settings on my pawn are wrong or something and it isn't detecting the overlap

i just did a distance check to all pawns and it works now, but i'll try to come back later and figure it out properly

[crisp shard]

i've asked this in the AI section a bit, but im curious on people's takes for pooling AI in a "mmo" like open world, w multiple zones that would only be active if players are near (active would mean the zone's AI would be spawned and active as actual pawns / controllers).

personally, the spawning/destorying is happening so rapidly that i dont think spawning and estroying is actually the right idea and i get a noticeable hitch for both actions at scale. im using currenlty an async load and queue spawning, but this isn't enough to make things smooth enough.

im thinking pooling would be the move as long as i stay within a "max" budget which i'll hvae to figure out , but that's only if the idea of pooling is a fathomable one. as well with this question, what is the typcial way to handle controllers when pooling AI pawns? do you pool the contorllers too?

[nocturne quail]

just disable AI controller when they are not in range I guess

• teleport will also help, just teleport AI to another location near players 😄

[crisp shard]

yea just made a little something.... have no idea if it will work but will update if it does work.. lol

essentialy doing the most minimal work possible to pool these, but have around 600 for the start test. i made a subsystem that handles the pooling, and it basically intilizes each class type and it's pawn in a Tmap to make the pool and i have zones that "request" the ai when needed, and when needed would be when any player is in vialbe range of that zone , which currently im doing a distsquared cehck every 15 seconds in the world subsystem on all the registered players. which is not ideal, i'd like to figure out a simpler approach than that but it's fine i guess to see if things work. then, each spawer subscribes to this timers event which is a delegate glboabl proximity check for all the zones (there's around 10 zones right now). and they all do their own checks based on their home location points and their activation radius. if not active already (no players were in zone recently) it will "activate" the zone by grabbing it's own spawner/zone config and this bascially grabs the necessary AI from the pool by class type and amount. then those pawn are woken up / enabled. my current enable / disable logic is quite extreme, im not sure if it's all necessary but i essnetially tried to turn off everything possible and disable anything that could poassibly do anyting and set dormancy to intial when sleeping them (which might not be extrmee enough but should work). and smae for controller bbut the controller i never "unpossess", i just leave them paired, whcih im hoping is simple enought o work. but yea if not active it activates, when players leave zone and it's active, it'll start a deactivation timer which slowly deactives it and if players reenter, jsut stops that funciton.

not sure if that explains the current jist but i am about to see if any of thisworks lol

[formal path]

just pool the characters

[dark parcel]

How to check if a player state is locally controlled? I don't want to route through controlled pawn as they may not have a pawn yet / never.

oh nvm, it has ptr to the controller.

PlayerState->GetOwningController->IsLocalController

[short arrow]

Playerstate has a GetOwningController() function? That's interesting because the owner of the playerstate is the controller

so usually it's just Playerstate->GetOwner()->CastToPlayerController

[dark parcel]

@short arrow

class AController* APlayerState::GetOwningController() const
{
    return Cast<AController>(GetOwner());
}

it does, yeah.

which is the same thing

[short arrow]

we have the technology

[ashen badge]

Hey everyone, I’m looking for help with a Steam multiplayer join issue in UE5.7.

Setup
• Advanced Sessions plugin
• Steam App ID configured correctly
• Online Subsystem appears to work (Steam overlay + game activity)
• Using Steam Sockets P2P (no dedicated server)
• Implemented standard fixes recommended in DefaultEngine.ini

It works fine when running in the editor as a standalone game, but fails in the packaged build launched via Steam. Although the host can successfully create a session and can be found by a client, the client joining the session results in a PendingConnectionFailure error.

Has anyone else run into a similar problem, especially when it comes to 5.7 specifically? Any help or pointers would be greatly appreciated.

[nocturne quail]

Why is fast array callback not logging about client?

Playing as client in Editor

LogInventory: Error: [Inventory] SERVER: TryAddItem: Item_Bagpack_C_0 x1
LogInventory: Error: [Inventory] SERVER: InternalAddNewSlot: Item_Bagpack_C x1 (ID=323F37464BB7C6C1543E63909A28CB1C)
LogInventory: Error: [Inventory] SERVER: EquipItem: Item_Bagpack_C_0 to slot 7
LogInventory: Error: [Inventory] SERVER: InternalSetEquipped: Item_Bagpack_C bEquipped=1
LogInventory: Error: [Inventory] SERVER: RecalculateCapacity: Weight=500.0, Slots=500
LogInventory: Error: [Inventory] SERVER: RecalculateCapacity: Weight=500.0, Slots=500
LogInventory: Error: [Inventory] SERVER: TryAddItem SUCCESS: Item_Bagpack_C_0 added with ID 323F37464BB7C6C1543E63909A28CB1C
LogInventory: Error: [FastArray Container] PreReplicatedRemove on SERVER: 0 slots removed
LogInventory: Error: [FastArray Item] PostReplicatedAdd: Item_Bagpack_C x1 (ID=323F37464BB7C6C1543E63909A28CB1C) bEquipped=1 WeaponSlot=255
LogInventory: Error: ═══════════════════════════════════════════════════════
LogInventory: Error: [FastArray Container] PostReplicatedAdd on SERVER: 1 slots added
LogInventory: Error: ═══════════════════════════════════════════════════════
LogInventory: Error: [FastArray Container] PostReplicatedChange on SERVER: 0 slots changed

void FInventoryContainer::PostReplicatedAdd(const TArrayView<int32>& AddedIndices, int32 FinalSize)
{
    const bool bIsServer = Owner && Owner->HasAuthority();

    UE_LOG(LogInventory, Error, TEXT("═══════════════════════════════════════════════════════"));
    UE_LOG(LogInventory, Error, TEXT("[FastArray Container] PostReplicatedAdd on %s: %d slots added"),
        bIsServer ? TEXT("SERVER") : TEXT("CLIENT"), AddedIndices.Num());

    if (!Owner)
    {
        UE_LOG(LogInventory, Error, TEXT("[FastArray Container] FAILED: Owner invalid!"));
        return;
    }
}

[silent valley]

#online-subsystems message

[nocturne quail]

This issue is resolved [Closed]

[fleet swallow]

Thanks @silent valley! We've fone through all of this already with @ashen badge and we still get that "Pending Connection Failure"

[silent valley]

Did you package as Development or Shipping?

[fleet swallow]

Team confirms Development

We have the correct fallback driver + ICE renabled) and it still gives the exact same “Pending connection failure”. We’ve already eliminated the top 80 % of causes (wrong NetDriver, ICE disabled, missing steam_appid.txt, lobby/presence broken, SDR status bad)

[daring folio]

Hey peeps, if someone has a good grasp on the Network Prediction Plugin that could help me I am trying to figure out why the server is running very far behind the client leading to high lag for remote players. I am using mover with the NPP backend, with average/bad network conditions for testing if I press jump on 1 client at frame 50 (for example), this jump will be played by the server at its frame 50 like half a second later, and then another half a second this jump will be seen by the second client, I just don't know where exactly that "lag" is coming from, you can see this in the GASP example too so nothing that seems specific to my project

[fiery olive]

Hey guys, the game i’m working on is supposed to be a 4 player coop (on one machine) but we recently moved to 5.7 and now the game crashes consistently when all 4 players play. I get these errors in the output log, I also have my code. Any suggestions?

[rocky totem]

Does anyone know why UFUNCTION(Server, Reliable) only generates const pointer params?

[verbal ice]

Can't modify the data in an RPC

[rocky totem]

ahhhhh i guess that makes sense. spent 5 hours trying to figure out why my replicated setters were only generating a const pointer variant

[kindred widget]

Never seen this happen with a pointer. Arrays are always const in RPCs. Clarity that it's read only. You can't affect the thing on this machine anymore at the point of an RPC.

Oh. Scrolled. 👍

[rocky totem]

yeah I was asking in the cpp chat and we couldn't figure out why it was generating code that didn't match the UFUNCTION it was attached to which was breaking everything

[gilded vapor]

Was the definition class a data asset?

[rocky totem]

A primary data asset yes

[gilded vapor]

I think that's why its getting flipped to const~~

Similar to how arrays have special-handling

[sinful tree]

The large set of warnings/errors relates to GetOverlappingActors somewhere in your Character blueprint. We'll need to see how that's being called to diagnose that.

In your animation blueprint, you should be pulling the character movement node from the reference you've set rather than trying to pull it off the intialize section as per my attached image.

What ends up calling Spawn Players and what class is it in? The errors relating to a local player already exists indicates you're trying to call this function multiple times somehow.

[fiery olive]

Spawn players is set up in a game mode it’s just hooked up to an event begin play

[sinful tree]

Do you only get those errors when you transition to new maps? Ideally you'd only create new local players if you needed them and they do persist through level transitions.

Oh, I'd also change the "As Third Person Character P2" reference in your animation blueprint to a validated get, just to make sure its not trying to access the velocity or character movement component without it being valid first.

[fiery olive]

I think I get it on level start and the start of sim but that’s because my main menu is also a level.

[sinful tree]

So I'd remove the creation of local players in the Spawn Players function.
I'd maybe change it to look like the attached image. This way you're looping through the number of players present in the game, and spawning a character for each of them at the different player starts.

The creation of the local players doesn't need to happen on begin play of the game mode since transitioning to a new level would end up calling for it again. You'll probably want to handle creating the local players before launching into the actual game from your menu so you can move to the level and the local players will already exist.

[meager path]

Hey! Is there any way to change the "perspective" from which i can inspect actors in the outliner?
If I detach the controller and select my character (which is on the server) i see this in the details panel (Should be aoutonomous proxy).
And If I inspect a replicated actor (which the server owns but has some fields that are not replicated and clients fill out), everything is empty (pic2). And they should be looking like this (pic3) for the server

[thin stratus]

Yeah you can select the world it shows in the settings of the outliner

[meager path]

i think im blind

[halcyon ore]

top right of your pic

It has a world dropdown

[meager path]

OOH the cogwheel

damn, thanks a lot

[charred island]

is there any method for splitting the world into multiple servers covering different areas? (kind of like server meshing for star citizen).

[nova wasp]

the engine is slowly adding "remote objects" as a feature but short answer: no

you could code an entirely custom netcode system yourself if you wanted to but that is your problem

[sinful tree]

Yea, nothing natively.

[nova wasp]

the engine has many built in features that help prioritize network objects near players and filter them by distance etc

but all on one machine

I would suggest not trying to make an MMO if you are new to networking

[charred island]

yeah, i'm just trying to think of how i might scale this very high

is there any method for calling CUDA code?

[nova wasp]

I'm not sure what this has to do with CUDA

[charred island]

oh, just scalability

[nova wasp]

technically yes there are compute shaders

[charred island]

i have a bunch of very similar computations that i could write CUDA code to handle lots of objects

[nova wasp]

how many and have you tried just... tasking them out?

[charred island]

not yet, it just seems like an obvious optimization for my purposes

[nova wasp]

which are unknown and beyond my understanding

have you tried just... a for loop in a parallel for? you are going to pay a lot for gpu latency etc afaik

[charred island]

have not, no, did not know that was a thing

yeah even on CPU that would be pretty big

[nova wasp]

you are going to want to have a basic understanding of what happens in the frame in order to make performance decisions

[charred island]

lol, i have been working on that

[nova wasp]

what are you doing that requires this perf? remember: we can only see what you type into here

[charred island]

one sec

[fossil spoke]

Why would you jump straight to something like CUDA as a vector for optimization? I think you need to slow down a bit and try and research and understand what mechanisms exist within the Unreal ecosystem first before going out left field.

[charred island]

its just what i'm used to

but yeah

[fossil spoke]

Ive seen so many people get frustrated with Unreal, because they try to fight it instead of working how its designed to.

[nova wasp]

if you are used to this stuff why is the idea of a parallel for a new thing?

[charred island]

because i know i could do it on GPU even faster, but yes

[nova wasp]

you have yet to describe exactly what is going on here besides "I want to make star citizen"

well, server meshing tech specifically

[charred island]

i was trying to get a video

but basically:

[fossil spoke]

If you have to ask a question like that, you probably arent ready for attempting to develop that kind of game...

[charred island]

i have a hybridized physics system where part of the motion is calculated via normal physics, and then the other part of motion is calculated from kepler equations. all objects in the solar system are calculated by traversing the tree from the player as a root, figuring out the relative location/velocity of the edges, and then position things based on that

[nova wasp]

oh yeah, the space thing

[charred island]

i have it seamlessly recalculating the kepler orbits based on input from the physics system to create perfectly smooth motion

but recalculating new orbital parameters, and then calculating relative position/velocities based on these parameters could get pretty heavy

[nova wasp]

could get?

[charred island]

they would be very easy to parallelize, though

[nova wasp]

you have neglected to mention the number involved here but this seems like it would be pretty cheap to do in parallel

[charred island]

i haven't tested it heavily, but yeah, thats on the todo list

[nova wasp]

a for loop over some simple soa thing will easily smash though thousands of these

[charred island]

i just finally got it seamlessly handling the motion

and re: our conversation the other day

i think i found a solution to handling this over the network, too

since i now have it smoothly recalculating the orbital data every frame, i can now just send over the orbital parameters

no need to send a tree at all

everyone will have the same tree of static objects that is synced on server time

[nova wasp]

https://www.forrestthewoods.com/blog/tech_of_planetary_annihilation_chrono_cam/ yep (not orbitals specifically here, but the idea of sending curves)

[charred island]

and all objects just calculate their motion relative to those

[nova wasp]

I think that makes sense to me

as long as they can derive the same result is should work fine

[charred island]

yeah, this way there aren't any dynamic nodes in the graph other than the actors themselves

so no issues with syncing

@nova wasp https://www.youtube.com/watch?v=onOq3Oi7slI

here's some older footage

that netcode worked pretty well, but it wasn't using the hybrid physics system

so everyone was forced into the same frame of reference

[nova wasp]

I feel like the prioritization could easily be just the orbital bodies near the player here to reduce the number of things to pull down

[charred island]

oh for sure

[nova wasp]

as for the raw ship movement that will be a more typical game netcode problem

[charred island]

but the server still has to do it all

[nova wasp]

it's just doing some math eqaution for every orbiting body? that seems pretty trivial to me if you allocate it correctly

[charred island]

every edge on the orbital "tree", yeah

and then additionally recomputing the orbital parameters every frame

for the pawn

[nova wasp]

have you profiled this?

[charred island]

nope, lol

[nova wasp]

okay, you implied earlier you were familiar with this

[charred island]

i may be overestimating the perf impact, but the math is pretty hairy

once you get into "newtons method" it starts triggering some warning bells for me lol

[nova wasp]

math is not expensive, memory access is

[charred island]

well, good thing on that front that its very memory light

should fit in cache, if not registers

but yeah

[nova wasp]

the issue here I guess is the hierarchy which I don't really understand but this doesn't seem like it needs to be expensive

unless every body is talking to every other body etc, I don't know

[charred island]

i get you though, stress testing it is what needs to be done

there's just several more pieces that need to be implemented before i can even do that

i can stress test static objects (no orbital parameter recalc, just location/velocity calculations)

but dynamic objects like player pawns will be a little different

[nova wasp]

I assume the dynamic stuff like ships and whatnot that get pushed around during gameplay are replacing their offset to the frame?

as for those it's regular unreal replication here

[charred island]

they're pushed from one coordinate system to another

so i cant naively send them over

[nova wasp]

the important part is reproducing the position on the other side

you don't need to use actor transform replication if it doesn't fit

[charred island]

yeah

again, i think i have this part covered

i can just use the same technique in the video i posted

basically just does a delta of the position and sort of rubber bands it to the correct location

[nova wasp]

with enough buffering you can make almost any shitty situation smooth

[charred island]

originally wrote all of that netcode in blueprint

haha

[nova wasp]

I mean you can definitely prototype a lot in there even with the lack of control

you more or less get onreps and enough arbitrary data to at least try simple things

[charred island]

yeah it worked fine

but i refactored it along with 90% of my systems in the last couple of months after not touching it for a year

my dream would be if i could scale all of this to handle things on the individual projectile level in the hybrid physics system

because then projectiles would actually follow orbital paths and whatnot

could do trickshots around the planet

[unique forge]

1 quick question

you guys start play as listen server

or clients

to test like flashlight or camera rep

[halcyon ore]

I mean, depends.
Is your game gonna be listen, or dedicated.

But, I always prefer dedicated.
Then fixup any later issues for listen/ standalone.
(hard to add replication when there is none sort of thinking)

[unique forge]

horror multiplayer with some levels, idk lol

[olive briar]

I’m normally using a game engine that’s different from Unreal Engine and its multiplayer system, and I’d like to ask you a few questions because I’m very inexperienced with multiplayer—especially when it comes to servers and clients.

First of all, my understanding is that the main purpose of servers is to receive incoming data, validate and correct it, and then send data back. In other words, they mostly handle data tables, APIs, and calculations. The client, on the other hand, processes the data received from the server and renders the visuals accordingly.

For example, if I use League of Legends as a reference: the servers would only handle things like player positions, items, gold, turret health, game time and logs, minions, player hitboxes, and all related calculations. Meanwhile, on the client side, the visuals are synchronized based on the positions received from the server.

Also, for latency, I assume there is client-side prediction for the player.

--I translated from ai beacuse my english is not good :d (except this sentence)

and I apologize for the length

[halcyon ore]

50/ 50.

Yes the server receives data, and does stuff with it.
(that be validating, apply damage, healing, killing, etc)

But, its not data tables, and API's really.
The server on Unreal Engine is a full copy of the game/ map.

It doesn't edit data tables.
it edits its own local copy of the heros stats, positions, and etc.

Every end has a copy of the actor (or anything), and they share info back and foruth to keep them in-sync

[lost inlet]

listen server = 1st PIE window is ROLE_Authority, rest are clients
play as client/dedicated = all PIE windows are clients, emulates dedicated server setup

with a single PIE client, a listen server won't be discernible from singleplayer

[unique forge]

whats better for replication and stuff

lol

[lost inlet]

I don't know what you mean

a single PIE client in listen server mode won't be testing replication at all

[unique forge]

so play as client is the better solution for every stuff for replication blabla

[olive briar]

In that case, for example, if a player is going to buy an item, it would first appear on the client side as if the purchase is possible (or this doesn’t have to apply to everything). Then the client sends a purchase request to the server. The server checks the player’s gold, processes the transaction, and returns the result. Depending on the situation, it either sends the update only to that player or broadcasts the change to all players.

so this is hard

https://tenor.com/view/dog-crying-meme-doggo-crys-megan-soo-crying-dog-gif-5276199764143986284

[lost inlet]

most of the time yes, but if you plan on supporting listen servers then you should test stuff with that setup with multiple clients now and again

[halcyon ore]

Pretty much.
The general idea at least.

But, stuff like variable replication will handle most of that for you.
No need to keep manual sync, UE does that for you.

Server handles buy request.
If it can be bought
-100 gold var

• 1 item array entry

Default replication will handle telling players of the -100 gold, and +1 item entry.

[olive briar]

I didn’t know exactly how it works in Unreal Engine, because I had only seen it done manually through APIs.

And doing it manually with just APIs is really frustrating.

[halcyon ore]

yeah, a ton of it (at least for a basic game like League)
Is auto handled by generic variable replication.
Cuz, at the end of the day, League is like 10 vars if you really think about it.
(health, coin, items array, level, and team or something)

[olive briar]

what exactly is automatic in UE?

[halcyon ore]

Variable replication.
By default it will keep it in sync for you.
You make the health var replicated.
All server has to do is change the var.
and, it will automatically be told to whoever is valid for it.
(in a game like league, that'd be everyone)

No need to manually keep the health sync'd (like you may be thinking?)
You just change it, and by default UE does it all for you.

[olive briar]

ok very thx for you spend your time 🙏

[dark parcel]

You cant just test in one. The mechanic has to be working on both cases.

[dark parcel]

How? You already got the right idea.

[halcyon ore]

I don't think they realized UE handled vars and stuff for them.
So, they thought they'd need like 50 RPC's to keep each 1 var in sync, for late joins, constant changes, and etc.

[halcyon ore]

Or, like I said.
Test the case your gonna use.
A multiplayer horror game is not gonna be on a dedicated server machine.
One of the players will be the host.

[dark parcel]

You dont use rpc to sync variable.

Thats the most common mistake.

[halcyon ore]

I know, thats the point.
But, they didn't know that.

[unique forge]

quick question

i tried with play as client to test

flashlight with spotlight blbabla

when 2 players, both works, but more than 3 it just break

like

player 1 und player 2 can enable flshlight and spotlight can see blabla

but when more than player 2, then player 2 and 3 cant use the flashlights and spotlight but can still see the spotlight from host/server/player 1

is this even right or just editor bug

[sinful tree]

Is your flashlight a separate actor from your player? Are you sending RPCs to the server to turn the lights on?

@unique forge

[unique forge]

well

i googled

3 pie wont work

like 3 clients

idk

[sinful tree]

Ok.... So.... How are you communicating to turn the lights on?

[unique forge]

a

its weird

i start play as client yk

client 1 works flashlight spotlight blabla

client 2 too

but when client 3 comes in, then it doesnt work for him

[sinful tree]

You're not answering the question. I understand its not working, but no one will be able to help you if you can't explain or show what you're doing to turn the lights on in the first place.

[unique forge]

Idk tbh

I don’t even know why it’s working

You got a plan or method how flashlight works?

normally with repnotify etc

[sinful tree]

Player pushes button > Sends RPC to Server > Server sets flashlight boolean repnotify variable
RepNotify > Turns light on or off based on boolean value.

[nova wasp]

also make sure you are not accidentally only using one player controller or something on the server

and yes if you can't describe what is being done here in terms of whether it's an rpc etc then it is not possible to help

we can only see the information put into here

[dark parcel]

Post the code

[dark edge]

Trying to "idk tbh yolo" your way through multiplayer game code is like playing Dark Souls by spamming buttons. Yeah you might make something happen, but you'll get much better results by being deliberate and understanding what you're dealing with.

[dark parcel]

quiet tricky if people follow youtube guides instead reading the network bible.

[unkempt pollen]

Quick question, can clients in a server trigger custom events from the Game Instance?

[verbal ice]

wdym

what are you trying to do

[unkempt pollen]

I am making a function where it leaves a server and gets put in a offline map

It's also an local event

[dark edge]

Any network traffic has to go through something that's not the GameInstance since it's not replicated

What's the actual gameplay here, explain what the player experiences.

[unkempt pollen]

A player clicks a button on their UI to leave a multiplayer match, a fade out sequence occurs and then put into a offline map.

[dark edge]

Yeah ez

That can all be local depending on if you need the server to know they did that vs just disconnected and disappeared.

[unkempt pollen]

Thanks, just making sure 👍

[sharp mason]

Hi Guys, how can i test the Network insights on the editor ? i could only make it work using stanalone game, but when i launch it with 2 players one player goes directly to the map and the other goes to the menu. Is there a way to play using PIE ? do i need another setup something to play listen server using Standalone game ?

[nova wasp]

yes, startup the editor with network profiling in the startup args

-NetTrace=4
-tracehost=localhost
-trace=net,default,task

etc

[hazy vortex]

Should I choose GASP or ALS for a multiplayer game? I plan to buy the animations and combat systems from the market. I thought ALS would be better in this respect, as I considered it a more established system. I'd like to hear your opinions as well.

[sharp mason]

do i put them on DefaultEngine.ini under [ConsoleVariables] ?

[nova wasp]

nope, these are startup args for the exe

I use Rider EZArgs but you just need to run the editor exe with these args

[sharp mason]

oh i see, now it works, thank you very much

[nova wasp]

it is definitely a bit more confusing in a PIE profile as you could see multiple server connections collected

I have found sometimes it only works after completely stopping and starting pie/the trace

[sharp mason]

Is there some guidance about package sizes or what should i be aim at ? i checked Networking Insights on documentation but is more a how to use and not what do i do with this data

[nova wasp]

honestly I don't have a really great idea of the exact bandwidth you are aiming for for the average game

this gives you the tool to at least see what is being sent and received and proportionally how much each field is taking

it's kind of up to you to decide which information is more relevant for the players to receive

for example you might find something needs to be sent very often because it changes a lot... maybe you can consider sending it less often by reducing its network update rate or even changing how it works to be cheaper for bandwidth

for example if too many things need to be sent to a connection some of them will have to be skipped

if you skip sending information too much things that are more sensitive to constant updates like the posiitons of moving things, characters etc might start to get larger spaces between receiving info

unreal has built-in bandwidth rate limits that are quite low that you can tweak a bit but I am not sure how they factor in to like, steam connections etc

If you want an example of a common networking situation

consider something like sending the transform of an object... this is quite expensive to send raw

what you can do for actor replication is actually reduce the quality of the information so that it is less precise, but takes less bits to send

by default most of them already will be quantized a bit

[sharp mason]

fair enough, its my first multiplayer game so i am trying to make less mistakes and also figuring out what i should be aware of, thanks for the tips

[nova wasp]

I personally do not really know of a really great way to decipher which objects were omitted from a packet due to bandwidth limits

I think there probably is a way that I am just not aware of

generally speaking it's easy to see when the bandwidth starts capping out visually as you will get stuff that actually starts "lagging"

[cinder quartz]

What's a way to integrate gun recoil safely for online play?
Obviously it's not needed to be fully replicated, but I wanna make it work correctly regardless of connection latency or client performance... So the angle getting constantly modified by a lot of factors at all times kinda worries me that a simple RPC can't sort by itself. I already had to fight HARD to make camera pitch replicate somewhat ok enough for basic gameplay purposes.

[dark parcel]

@cinder quartz wdym by a simple rpc cant sort by it self?

Remember the only way for client to communicate with server is through server rpc and nothing else.

[nova wasp]

what does safety mean?

I would suggest simulating locally and predicting for something that is this sensitive to timing

you have not given any details about what this rpc does or means (when does it apply?)

[cinder quartz]

By that I meant more so, any extra steps or considerarions.

[cinder quartz]

As for this, the idea is replicating the effects of recoil on the player controller, with rough visuals and most importantly sending the correct position and orientation when spawning the bullet.

I think it should works but I am a bit concerned by whether or not potential performance issues or latency drops could cause problems if I do the current method I do where I just grab a transform, calculated in the client, and sent to server to make the bullet.

[dark parcel]

really depends on how you want to design the game.
You should be able to describe the system in plain english before even diving into the coding implementation.

latency will always be an issue

easiest mode is to just trust the client and have some checks.

let client inform the server, hey I shoot, on this direction.

[cinder quartz]

Well that's simple. When the player fires, the gun rapidly applies forced aiming upwards and sideways with a strength based on the type of gun and how low they have been firing.

Then if they stop firing, this forced influence is removed up to a maximum value.
However, if the player tries to counteract the recoil mid firing, this effect is diminished accordingly. So if you fight the recoil and stop firing, the reticle would not try to move much.

[dark parcel]

so which part are you struggling with?

I would just let the client tell the server which direction it ends up firing.

if they cheat, then oh well.

[cinder quartz]

Eh cheating may not be a huge issue for my use case

[dark parcel]

the recoil just get calculated locally right after firing. Client only need to tell the server the final direction it's shooting.

Yeah then just network the result to the server and have the server replicate to other players.

[cinder quartz]

I do wanna learn about the ideal method for most of what I do though

So that if I sometime I have to work on a competitive game thing I am not as aimless

[dark parcel]

well think about ways to prevent people from cheating or at least make it harder.

just with plain english, forget about implementation in unreal.

For one, you can do a bit of math to see if the shooter shoot with almost no recoil.

anyone that is "too good" can be flagged.

[cinder quartz]

I suppose I could make a margin of error based on input value changes

Could be fun to figure out sometime

[dark parcel]

what error and what inputs?

what use is to track people mouse

just look at how close the shoots are to each other. If near perfect all the time, they are probably cheating.

Also client can just send RPC with some manipulation, so checking the time between the shooting RPC can also be done.

https://www.youtube.com/shorts/uRqo7L1CARE

[dark parcel]

isn't visibility already replicate? it could be fighting the boolean OnRep.

[unique forge]

no idea why

[dark parcel]

/** Whether to completely draw the primitive; if false, the primitive is not drawn, does not cast a shadow. */
    UPROPERTY(EditAnywhere, BlueprintReadOnly, ReplicatedUsing=OnRep_Visibility,  Category = Rendering, meta=(AllowPrivateAccess="true"))
    uint8 bVisible:1;

Yeah it's replicated.

[unique forge]

what should i do?

[dark parcel]

Just send a server RPC with a boolean param, which set visibility of the replicated scene component (the spotlight)

delete your bFlashlightOn.

that's what I would do.

hmm but not sure if you have access to OnRep_visibility in blueprint.

cuz if that's the case, then you can keep your boolean but don't make your scene component (spot light) replicated.

@unique forge Yeah, just do this.
Keep your boolean and logic, but don't make the spotlight 1 replicated. Untick replicate in the detail panel, that should fix it.

[unique forge]

a

[dark parcel]

?

[unique forge]

just noticed

[dark parcel]

im not sure what you noticed

[unique forge]

client 2 (player 2) window spotlight is normal

but others see something else

the spotlight yk

[dark parcel]

what changes have you made? worry about the transform later.

just make sure that the spotlight is on and off accross all machines.

[unique forge]

yk why this is happening, when i add my level assezs from other level, then it doesnt show

when in this level is no assets or something else

then it does show

thats the only thing i noticed

what causes this mystery

[dark parcel]

I dont know what adding level asset from other level mean.

also u are not answering the question, so it's really hard to help.

[unique forge]

what you said about spotlight1 replicate off

i did

this is works on level basic

[dark parcel]

k and?

[unique forge]

its tricky and complex

no idea why

in this level there are like postvolumes, lightmassimprotancevolume etc

[dark parcel]

can you print something on the OnRep function

then take another recording.

need to see if the OnRep is actually being called on the problematic machine.

[unique forge]

I just think this level is haunted or something

[dark parcel]

could be transform issue, how is the spotlight set up?

it might be facing different way on Client 2.

I saw your screen shoot and it's an issue already.

[unique forge]

okay

camera/spotlight problem?

[dark parcel]

for one the camera comp transform will only update on the local machine, it's not replicated nor replicate to other machines.

[unique forge]

okay

[dark parcel]

I would attach the light to the capsule component.

you will have to replicate the pitch though.

[unique forge]

later

i went to basic level again, on there it works butter

on this level, it just doesnt work

could it be maybe rebuild light?

[dark parcel]

I have no idea at this point. how are the characters spawned?

[unique forge]

the auto receive input on both are disabled

[dark parcel]

@unique forge print string this

btw how is the light become On and Off?

[unique forge]

the on off is the material

youll see on video

bad video for showing on off

that glow light is the On material

incase if you still dont know, there are these 2

[dark parcel]

I can't replicate the issue on my end.

yeah looks like transform issue.

Shouldnt attach the light on the camera, at least for the one to be networked.

[unique forge]

did you tried with 3?

for me only works for 2 players (server and client)

[dark parcel]

yeah I did, you can see on the screen

[unique forge]

but 3rd joins in, its a chaos

[dark parcel]

1 server and 2 clients. All work fine.

[unique forge]

okay

[dark parcel]

try testing on another level maybe.

[unique forge]

is must be level issue or something, this level is haunted

i tell you

[dark parcel]

🤷‍♂️

[unique forge]

i must have added something in this level that causes this problem

but no idea what causes this

[dark parcel]

check if you got anything on the level bp and if the game mode is the same.

other than that, no idea.

[cold mesa]

Testing out Chaos Mover - I must be missing some setting right? The positions are way off on simulated proxy.
Using Resim on simulated proxy basically gives the best result (haven't tested that much since it makes no sense in my mind, Simproxy should interpolate which i figured it always did for Resim anyway? We have the auth state, why would we resim?)
Tried setting Interpolate manually when initalizing the chaos mover comp and that got rid of the overshoot issue for simproxy (I really dont think i understand how the physics prediction works for simproxy, how can it ever overshoot? it should be behind)
Also tried the example version of using a Network Physics Settings comp and asset, setting the simproxy to interp and setting up input decay. The Correction limits seem plausible, but daaaaamn everythings so out of sync, we are talking 100s of uu at times with network simulation settings set to average.
Setting the pawn to replicate movement will get it in the right spot, but with horrible effects and that defintely can't be intended right? then we're replicating the movment twice and just clobbering it.
Granted it's WIP but i still feel like i must be missing somthing basic here cause it can't be THAT WIP that it doesnt even really work for networking when that kinda is it's intent?

[cold mesa]

This sorta ends up being the end state after a move where blue/red is Auth and Autonimous Proxy and Blue is Simproxy, and it just stays there

[ruby lodge]

I am currently deep diving a bit into Client prediction and rollback again since a long time. Is this correct for a change in movement state?

[meager spade]

blueprint doing this is not a great idea

this isnt really that predictive, and tbh, this is fragile

also owning client never changes the state

on "rollback"

and mixing replicated props with RPC's like this

is a recipie for disaster

[old dirge]

i'm having an issue where whenever i enable Component Replicates on a component in my character the client can no longer move or do anything (veggified) while the host can the same stuff, but when i turn it off it goes back to normal then i can move and do the same stuff that the host can do. any help would be lovely?

[ruby lodge]

In case of lag, the client prediction part does its job ig

[ruby lodge]

On Rollback it resets the state to what the server had

Could you explain how else to approach it in blueprint?

That would help a lot! I have seen others do client prediction somewhat like that, but they rarely also do rollback

[mild canyon]

Does anybody have any recommendations for making a reliable timer for a round based gamemode. Currently my replicated one runs into an issue where the clients seems to lag behind by 1 second due to a multicast.

[halcyon ore]

Just replicate an end time.
Profit.
No need to constantly update 1 second left, 5 minutes left

[mild canyon]

so I would run local timers for each client. I just multi cast when it starts and ends?

[halcyon ore]

You really only need/ care about end, you don't care about start (at least networking wise)
Cuz, you just gotta say 20 seconds left

[mild canyon]

Yeah that makes sense. I did leave out, I do use a kind of coutndown feature. Where I ahve a stopgate that prevents player from moving before the round has started

Preferrably I need this to be 1 to 1, do you have any recommendations in a case like that

[halcyon ore]

Same thing.
Replicate the end time.

When is play allowed, essentially.

[mild canyon]

ok. I think I understand... I will give it a shot to remake this. Thank you for your help.

[halcyon ore]

Yep

[ruby lodge]

The OnRep variable has the SkipOwner condition to not be called twice on the client in the top event. But as it's being called only on the client in the rollback event, its only set for him

[meager spade]

but why do others need toknow then

whats the point in replicating it?

only server and owning client needs to know the mode

i wouldnt even replicate them

just do the rpc's

[dark parcel]

https://www.youtube.com/shorts/H60Ir7dUtlI?feature=share

Are there any reason to have the hitbox like this? Networking related?

[gilded vapor]

Maybe its the mesh smoothing into the actual-location due to latency?

[lost inlet]

yeah I would bet. this looks like sv_showhitboxes 1 so the hitboxes are showing the server location

but source does have a pretty robust lag compensation system

and source engine networking is a subject that a lot on yt get extremely wrong

[quasi tide]

and source engine networking is a subject that a lot on yt get extremely wrong
Agreed

[robust shell]

this might be a very basic question but why can't I see my other characters when testing with multiple instances? are there more stuff I need to do other than ticking replicate on the char mover blueprint?

[karmic viper]

hi everyone is here anyone that could sugest / help me please? im struggling with getting item from inventory, any ideas how can i do it? i have a simple set up like that on picture 1 and on host it runs perfectly but not on client, client can have item but when interacting (check for item) it just gives false all the time any idea how to get the item? the idea is for example to check if player have a key and once he do it unlocked the doors. please help!

just to mention, the client have the item in inventory so it looks like the server cant retrieve it from client? i might be barking at a wrong tree (photo 2 and 3 is host and last one client )

[karmic viper]

you need to turn on listen server to see each other

[sinful tree]

You'll have to show how the client's inventory gets populated with an item, as that too would need to happen on the server.

[karmic viper]

thank you 🙂 i have manage to find out finally 🙂

that was my solution, preaty much i forgot check "is locally controlled"

[sinful tree]

This isn't right either.
Checking whether something is locally controlled or not should not play a role in the server or client determining whether a certain player can interact with an object.

If "Check For Key" is run on a client, then you also have the problem that you're allowing the client decide whether something should be taken from their inventory or not, while also calling to run on server something else - essentially, a client could call that RPC at any point. It also makes little sense that this RPC exists on a BP_Door object as it would most likely not be something owned by a client, and thereby, they wouldn't be able to call that RPC in the first place.

If it's actually running on the server already, then it would return false for most characters since only the server's player character would be locally controlled, and that also means that "Open Server" shouldn't be an RPC as a client could technically call that without first checking for a key.

Ideally this should be more along the lines of:
On Client (or listen server):
Interaction Event (key press?) > Determine the actor they want to interact with (a chest?) > RPC To Server to begin Interaction with an input of the Actor they want to interact with.

On Server:
Validates whether the input Actor on the RPC is valid, and can be interacted with by the character then passes the rest to the actor itself to then handle the interaction. In this case, you'd then have the server run Check for key > Remove Key > Begin Opening

[magic smelt]

what environements do you guys use to test multiplayer? Ive used sandboxie and it seems to always bug out with two steam accounts.

[quasi tide]

We just test it with a closed group and ourselves

[dark parcel]

A pair of pc and laptop

With friends on something major. Distributed via steam.

[ruby lodge]

Because all players need to have the correct value to know how fast the player is moving or do you mean for the others its sufficient to know from the replicated location?

Just saw your messages now

Generally, assuming the variable is not replicated anymore. The client prediction, server check and rollback is good like that then?

[delicate shard]

Hello 👋 I had a quick question regarding the "Load Stream Level" node for a multiplayer game. Is it possible to use this node only locally? It works fine on the clients, but whenever the hosting client calls it, it loads the map for everyone. Any way to prevent this?

[knotty bane]

If you execute it on the owning client you should be good, otherwise you run it as server authority

[knotty bane]

I've done it with running a VM on the same computer, so similar to sandboxie, now i just use a cheap scrap pc as a second device

[tardy fossil]

might not be that easy.. server needs to load it too for collisions unless movement is client authoritative

[twin vessel]

Cmc question:

if we receive a correction and in the same frame we also receive an acknowledgment for the move right after the corrected one, then our last acknowledged move is no longer the one from the correction, but the one after.

This makes it so now we freed the saved move right after the one that got corrected.

Now, when we replay the moves on client, we start from the first saved move, which is no longer the one after the correction, but the one after that.

This means that we will end up in a non correct position and trigger another correction, since we now are missing a move.

What am i missing here?

[delicate shard]

I’ll try this after work and let you know

[delicate shard]

It might be, I’m using smooth sync plugin instead of UE’s default movement replication, collisions only need to be client sided. I only replicate the players location/rotation

[sinful tree]

Is everything pretty much client sided? Actors placed in those streaming level also wouldn't be replicating.

[delicate shard]

For these levels, yes

this persistent level is mainly hub worlds, kinda of like a lobby where you can edit your loadout. Think of it like old gen Monster hunter villages/gathering hubs

but also as mentioned im not using the default replication system, I'm only replicating the player's location and rotation in the persistent world, so anything collision and else don't matter here

If I were to see the other players in another world thats not loaded, they would only be appearing floating

[sinful tree]

Are the separate levels supposed to be actual play areas or is it part of the hub?

[delicate shard]

different hub worlds

the play areas will be in another persistent level and I'll travel all the players at once, when they ready up

[sinful tree]

Why the need to separate them and trying to make it so only clients that are present in them have them loaded? Surely it wouldn't be gigantic if it's not used for play?

[delicate shard]

in the scenario where all players are on a different part of the map, I don't need all the levels to be loaded into memory if they're not there

like I mentioned this works fine for clients (shown in screenshot), but is there a way to turn off replication for the listen server momentarily while I call the Load Stream Level node so that it only executes on their side and not all clients?

[sinful tree]

Unless those levels are absolutely huge, which by your description of what they would be used for would probably not be, using level streaming just so all players don't need to load what they're not present in seems overcomplicated for minimal gain, especially if these levels are static and would be the same for anyone else that also goes to the location of that level.

From what I understand of level streaming, it is not possible to have the server load a level and it not also load on clients -- hopefully someone else may be able to pipe up about this. You could have clients hide the level so it isn't visible to them, but they would still end up loading it if the server loads it.

[delicate shard]

Yeah I figured that might be the only way, but still exploring to see if I can just load it client side for everyone first, this will determine if I'm limited in map size for the future

[meager sage]

Hey all,
Is is legit to add a const FSstring& to reliable RPCs that are called one-shot? (Like they should not be repetitive)
The point is to add flexibility and respond to the strings

[nova wasp]

what does "legit" mean here

what is in the string

what are the range of different values? (well, what is in I guess)

if you send a string it is forced to serialize it as the entire string, this is expensive but totally fine if it's rarely sent

but it depends on the actual thing sent

[meager sage]

Hmm I can limit how long the string can be?
Yes to paint the picture, it's multiplayer support for a UI indication system. So for example I want to be able to broadcast adding UI locally and then if the string has some other instruction, I can grab it and apply ity on the Indicator UI

[nova wasp]

that does not actually describe the range of unique values

it sounds like you want to send a range of different "instructions" somehow and parse it on the other side

[meager sage]

Exactly

[nova wasp]

how many unique instructions are there?

[meager sage]

So I thought I would leave it open because it's for a plugin and I want it to be flexible. But what I'm getting from you here is that it might be better to make some kind of enum instead?

[nova wasp]

I think one way to let users add more but keep it fairly cheap to send would be using gameplay tags or fnames

not sure why a string is needed here but you have not actually describe what these look like yet

[meager sage]

Oh FName is actually lighter to sned? For some reason I was sure it's the other way round

[nova wasp]

why would an fname be more expensive than a string?

in Iris they are sent as a network export too, in the old network backend I forget if they ever export the string to a mapping

in the old network backend gameplaytags would be cheaper here

[meager sage]

I'm not sure a string is needed at all, I just assumed it's the standard way because I saw some built in networking functions that take strings like level transitions and all that

[nova wasp]

the reason they take strings is because they are designed to point to random assets

and they are also ancient 90s unreal code

which isn't to say it's bad I suppose, just generally speaking you don't need to rely on strings here... we have much easier things to use

[meager sage]

Oh damn I see 😅

[nova wasp]

it sounds like you want the UI indicator to point to a specific UI element

[meager sage]

I have so much to learn, thank you so much 🙏 🍺
I think I will start by looking into FNames and GameplayTags to see how they are sent etc. My fear is coupling it to gameplay tags becasue I want to sell it, and some users might not be familiar with them.

[nova wasp]

I'm not sure if it's like the "type" of UI element or a specific UI element in a list of them etc

[meager sage]

So I believe I got it figured out, I do send in the RPC the actor to point to and a soft ref to the asset to load (I guess this is like sending a string already?)

[nova wasp]

GameplayTags here would work pretty well because it would let users add more and even show them in the editor, it's very common to send GameplayTags around because they are cheap to send (a limited number of them so they get turned into a smaller ID)
the only downside is GameplayTags must kind of be defined ahead of time, so for things that are truly dynamic dynamic information is needed

[nova wasp]

sending assets HAS to send the stable path name even if it was a pointer to the asset

at least once

but the cool part is it can resolve the path on the other side and have a smaller shared ID that the server can send (at least in iris)

this happens basically magically for you in the background but I am more familiar with Iris... I am not sure how soft refs get sent but I would assume they need to send by path as well

If you want you can measure how your packets look with network insights
you need to start the game or editor exe with these startup args:

-NetTrace=4
-tracehost=localhost
-trace=net,default

[meager sage]

I'm inexperienced with Iris ufortunately. Hope my solution will be robust enough, I think I will end up with FNames or GameplayTags eventually. They are only optional too, so..

[meager sage]

Oh cool, thank you, and then what should I lookout for?

[nova wasp]

I'm honestly a bit of a weirdo in this area due to being one of the few who has used iris in production more so than the current networking backend... which means I am a bit biased in the information I give

Iris has a few major upsides but it is totally fine to use the regular networking backend as from the gameplay code perspective things will be largely the same
(and some downsides, Iris serializers are a bit of a headache when you need them but you will VERY RARELY need to make one)

[nova wasp]

generally how large your rpc is

it it isn't being split into multiple packets etc it might not be a big deal

to be clear there's no way to avoid needing to send large data, but you can get a good idea of how large it is in proportion to typical unreal rpcs

and you can see how each part of the rpc is serialized per-member

along with receiving data for replicated propreties and packet header info

as for "what would be good or bad" it's going to depend on bandwidth constraints and if you can avoid the problem or not

sometimes sending giant mega rpcs is required

[meager sage]

So I read the wrong data in a prev message, it's actually 253 bytes with the current string version (I know it depends on the length of the string but I see it's actually not the most significant part. I also pass an FName alongside it and they weigh the same.) Either way I will move to an FName or GameplayTag, but would you say it's okay to stay within the 200-250 byte size?

[nova wasp]

that's a realtively large rpc but 253 bytes will easily fit into a single packet

also consider that users could spoof sending this info

you don't want to like, parse the string and execute random stuff if possible

[nova wasp]

can you give an example of what a string looks like? you can obfuscate things etc if needed to not reveal project info

in your rpc I mean

[meager sage]

Yeah it's just something I started to play around with just now, it's definitely possible not to use a string. I just need something generic on the receiving end to extend the function. In this specific example, I added the string "AsyncDistanceNode" which I check for equality in an example blueprint, to show that it can be used to create some async node that has some handy distance based exection pins. It's really not a must to use a string, it was just my first thought becuase I saw it used before

[meager sage]

I hope it's okay with you to check out the insigts. Don't hesitate to tell me to go look for help elsewhere, I can feel that I might be going too far. Sorry about that.
If you could tell me what you think should be my main focus I would appreciate it, I'm sure that I'm not reading this info correctly

[nova wasp]

okay, so this chart shows "incoming" packets on the client

you can change that in the dropdowns in the top and by default it makes two conveneint tabs, one for server and one for client iirc

in the editor you might see more than one connection for each client/server listed as they might have seen multiple PIE sessions

what is the definition of this rpc?

[meager sage]

    UFUNCTION(Server, Reliable, BlueprintCallable, Category = "World Indicator|Multiplayer", 
        meta = (DisplayName = "Broadcast Add Indicator (All Players)", AdvancedDisplay = "5"))
    void Server_AddIndicator(const UWorldIndicatorDataAsset* IndicatorData, USceneComponent* TargetComponent, 
        FName SocketName, bool bAutoShow = true, bool bAddUnique = false, const FString& Message = "");

Like in code?

[nova wasp]

also you might want to see what happens if you send the same information multiple times

so for example, in this case

UWorldIndicatorDataAsset is being sent over the network as a path to the object because it is an asset

sending USceneComponent* TargetComponent is going to require that it is network addressable somehow, which I am not sure of here (if it's receiving okay on the other side, it's fine)

[meager sage]

I mean I saw it works in a shipping build but only in LAN

[nova wasp]

what does LAN mean here?

you could test this in the editor and you would immediately see if it worked

[meager sage]

I used 2 pcs on the same network connected them with the ip address

[nova wasp]

that's fine, it won't change how serialization works

the packets don't care about what the ip is

[meager sage]

Oh okay, yes I also thought it's weird that it works. Maybe it depends if the target component belongs to a replicated actor?

[nova wasp]

for dynamic things, I think yes but I'm not 100% clear on if the component itself must be replicated. Generally speaking I would recommend against randomly making components replicated without a good reason

unless they truly change state in a way that isn't simple to represent

also just for sake of argument, you can see sending booleans is basically free lol

these are 1 bit each!

so this is actually sending a network GUID for "export" , which I think means that ensuing rpcs of the same value will be sent as a smaller id

if you can it would be interesting to send this rpc with the same target pointer twice, just for demonstration

basically what I am expecting to see is that the second RPC will be VERY small due to having a shared ID which is just a tiny number that the client and server agreed upon

[meager sage]

Yes it looks like exactly what you've described

[nova wasp]

now unfortunately in the default replication setup the FName is still sent as a string

but tbh... 200 bits isn't bad if this is not sent every frame?

[meager sage]

Nope, it's meant to be sent manually when an indicator is added/removed. So my assumption is that it will not be used very often. (Added/removed in the sense of created destroyed, not hidden)

[nova wasp]

of course you can't really predict what socket names people will use but you could make some simple setup where the information is sent is which "kind" of socket is used

[nova wasp]

kudos for actually getting network insights going as well... I often say "profile it" in here but rarely do I see people actually try to do so

[meager sage]

Haha thank you so much, it's indeed my first time using insights. I'm beyond grateful for your time and help

[nova wasp]

Network insights only really shows you the raw packets without much context for the current sort of... limits and priorities so it's more useful to find outliers or the overall "pattenr" of sent information at least for me personally (I'm sure I am missing something)

if it helps I would HIGHLY recommend using regular unreal insights as well with named events on at least once just to see the "shape" of how a game frame looks

you don't need to understand everything unreal is doing, but seeing what order things happen in will helpa avoid a LOT of confusion and help visualize their cost proportionally

for example insights will show the rpcs being received, and when they are sent out

insights is not just useful for finding why something is slow, but to just understand wtf is happening in general

and to be clear the editor will be a bit of a weird case (PIE single-process multiplayer ticks each world back-to-back and will tend to have more slower extra stuff enabled) but it will largely be equivalent to the running game in a general overview sense

[meager sage]

For sure, I watched an insights walkthrough by Ari Arnbjörnsson a while ago and my eyes were bleeding. But even this small experience you gave me will help me understand it better, I should try to watch again soon.
I'm gonna call it day for now, will be back on it tomorrow.
My plan is to move the message string to a gameplay tag because think I saw that after the first call it becomes completely zero cost (I will need to verify that).
So I will be left with mostly the socket name which is also optional and consts 0 when not passed. So if the user can make a scene component just to house the indicator and avoid the socket name we get a pretty nice and optimiszd rpc call (at least after the initial call) >> Maybe I can force them to do so 😈 .

Thank you truely!!! 🙏

[nova wasp]

fwiw gameplay tag replication being nicer might require some extra setup without iris
https://tomlooman.com/unreal-engine-gameplaytags-data-driven-design/#networking-replication

and this REQUIRES the server and client to have exactly the same gameplay tag list which is going to be sensitive to tags being added after startup

[meager sage]

I'll read it before attempting anything, thanks again

[jovial ravine]

I was trying this recently and couldn't get it to only load for one client. A.i. says: No, you cannot natively load a streaming level for only one client in Unreal Engine’s default multiplayer setup. The engine hard-codes level streaming replication, meaning if the server loads or unloads a stream level, it automatically replicates that state to all connected clients.

[delicate shard]

yeah it seems like it

[nova wasp]

just load a level instance?

[delicate shard]

you lose connection to the session if you do

oh you mean create the level like that

[nova wasp]

this will generally be easier to get going with the server being aware of the collision primitives with the default movement setup

[delicate shard]

This is what I was trying

I'm not using default replication, I dont need server collisions

[nova wasp]

be specific

[delicate shard]

I'm only replicating the players position and rotation in space

[nova wasp]

I am referring to the default character movement component which relies on local prediction with all peers doing partial forward prediction with a small rollback sim on the server

[delicate shard]

yeah I disabled that, I'm using smooth sync plugin

[nova wasp]

if you do a custom setup for sim proxies that are in an "unloaded" cell locally that might work okay

but you will probably want to override the onreps directly

[delicate shard]

my only issue is load stream level loading the map for all client when called from the listen server

[delicate shard]

I haven't tried with an instance

could I get accurate coordinates each time?

[nova wasp]

what does this question refer to?

what would be "innaccurate" here

[delicate shard]

if you spawn the level at a specific location and rotation, for 2 different players, will it align perfectly? in the scenario they both have it loaded

I giess I could set some sort of object in my persistent level and spawn it at that location

[nova wasp]

to what degree must this be perfect on both sides? perfect bitwise determinism here is going to require controlling floating point math down to every single step of anything that can possibly affect the result of the frame... yes you can send a transform to another client and expect it to be 1:1 if you don't quantize it

[delicate shard]

Essentially trying to replicate the load stream level behaviour. My maps have a set physical location in the persistent level

The idea behind loading these levels locally only is to prevent having 4 levels loaded in memory in case all players are finding themselves in a different location. This is just for the "lobby" part of the game. Players can travel to different areas during loadout phase essentially.

[nova wasp]

Even using a non-instance level doesn't appear to induce a disconnect... I am really curious what is actually disconnecting here

I must be missing something that only exists in a fully built cooked game?

[delicate shard]

I thought you meant open level at first, my bad

[nova wasp]

was there an actually measured performance issue for the server here?

[delicate shard]

not now but planning ahead

[nova wasp]

not sure where I ever specified "open level". to be clear I am just messing around with these BP calls

in c++you can have a lot more control over a level streaming object depending on how much painful boilerplate you want to add... I am not sure why a disconnect is induced here

[delicate shard]

theres no disconnection with these nodes, like I said I thought you meant open level node at first

[nova wasp]

openning a level is not the same as loading a streaming level

generally speaking an "open" call is tantamount to basically starting a new level from scratch with it as the persistent level

[delicate shard]

yeah I know

I simply wanted to know if there was a way to prevent the listen server from executing the LoadStremaLevel node on all clients. But I'm guessing there's not, so I'll work around it

[nova wasp]

it does not seem to do this

are you doing something to induce updating the replicated level visibility?

[delicate shard]

yeah I'm building a system that will be basically load the level if a player travels to it and no one is there yet or set the visibility to true if already loaded in. Then when they move out, check if there's remaining players, unload the level or hide it locally if there's still players

[nova wasp]

that describes what you want to make, I am asking what you are actually doing

it helps to know the goal but I am wondering what specificly happens to load the level on the server, for example

[delicate shard]

I was just messing around with it but rewriting the component at the moment

initially I was using the LoadStreamLevel node but the server would spawn the map on all clients, like shown in the screenshot

[nova wasp]

you can set bClientOnlyVisible to true on the level object created by level streaming

[delicate shard]

yeah I know, I was trying to prevent the level being loaded entirely in memory for all clients

[nova wasp]

otherwise it the client and server will pass info back and forth about the levels loaded and their status (for normal level streaming

[delicate shard]

I will handle the loading/unloading of the levels on the server only

and handle visibility client side

[nova wasp]

do they need to be normal streaming levels? level instances seem to pretty much just do what you want

[delicate shard]

Yeah I'll expirement with it tomorrow, but I'm having the streaming setup for the actual gamemode where the gameplay happens and will have replicated/synced enemies and such

[nova wasp]

replicated objects need to be handled on the server the normal way

[delicate shard]

yeah

[nova wasp]

so you can't have any streaming setup for the "actual gamemode"

clients can absolutely only receive information for nearby objects, but the server would have to have them loaded for clients to receive replicated data

[delicate shard]

yeah during gameplay/combat gamemode the players will be restricted to the same "zone" before loading the next area in a way

I'll figure it out then, but thanks for the tips!

[tidal cloak]

Hello. I got a weird situation.
I have a game instance subsystem which handles backend messages. when a message comes from backend server it fires an appropriate function in specific class. Think of it as subsystem listens for message and based on message it broadcasts a specific delegate.

Now i have a player controller that says "Hey subsystem if "abc" message comes execute a function named "xyz" that is defined in me".

Now suppose there is a listen server host. a client connected to it. then due to connection error client disconnects , client goes to starter map.

now in subsystem invocation list the object ptr is stale. so on message from backend bound function does not gets called.

I Thought of rebinding function to delegate on PC's beginplay. but the thing is on network error i think player controller survives when going back to starter map. because begin play does not gets called.

TLDR : Host and client -> client disconnect -> Goes back to starter map -> in subsystem delegate invocation list loses ref to PC.

[thin stratus]

PlayerController doesn't survive. But you should also not keep any World references in your GameInstance to begin with. If your PC needs to know about the mesage, then the PC should Bind and Unbind it.

[tidal cloak]

So maybe Beginplay not firing could be UE bug. I need to find a event that fires in PC on network failure level switching.

[thin stratus]

PC won't fire BeginPlay on SeamlessTravel if the PC Class is the same on Source and Target GameMode.

But that's the only case.

[elfin holly]

I think there is something wrong with the Steam Sockets plugin with 5.7.4 ? Anyone else confirm they are unable to build with it enabled?

[jovial ravine]

I was using Load Stream Level by object reference or by name and couldn't get it to work. It would spawn the levels for all clients. I don't think I tried instance. If it works that's awesome. I used this:

[nova wasp]

the words you are quoting say "level instance"

load a level INSTANCE

[jovial ravine]

Thanks!

[nova wasp]

a level instance is similar to a streaming level but they let you load more than one out of the box

they are a bit more convenient for stuff like this

[slate phoenix]

Hello everyone,

I’m planning to implement a trading system between two real players (not NPCs). However, I’m not sure where I should implement it. Should I create a component or an actor?

At first, I considered putting it in the Controller, but I don’t want to do that because players might manipulate it. Instead, I thought about spawning an Actor when one player sends a trade request to another, and destroying it when the trade is finished.

I’d like to hear your opinions. What would you recommend?
(Note: I’m not asking how to build the trading system, only where it should be implemented.)
I’m developing a MySQL-based MMO sandbox game.

[haughty ingot]

@meager spade

[ashen plume]

limited mmo and ue5 experience, but my gut says spawn an actor, for my understanding of how authority and relevancy works, which is lost with components or controllers on players

[dark parcel]

Component lives inside actor.

[dark parcel]

Wdym by player manipulating? How is it possible that the player manipulate the trade unless a bad design choice was commited?

The items are in the server. It doesnt matter what the client says.

[slate phoenix]

Since the PlayerController runs on the server, a player could more easily access or exploit it if they wanted to hack.

[dark parcel]

The controller exist on the server and a client gets a copy. This is true for all replicated actor. Can you be more specific on "access and exploit"?

The only way for client to communicate to server is with server rpc and nothing else.

[slate phoenix]

A PlayerController and an Actor are not the same. If I implement it in the PlayerController, each player would have their own trade system, which could be easier to manipulate. The same applies to using a component.

That’s why I’m considering using an Actor. When a player sends a trade request, it wouldn’t be tied to a specific player—instead, an Actor would be spawned and the trade system would be handled within that Actor.

The PlayerController is mainly for player control, not everything. Please don’t shift the topic elsewhere.

Honestly, I wasn’t sure, so that’s why I asked this question. I’ve been thinking about where it would be better to implement it or how I should design it, but I couldn’t decide. That’s why I’m asking which approach makes more sense. Just answering this question is enough.

Maybe my thinking is wrong.

Thanks for the suggestions everyone. I decided to handle all trade logic on the backend (WebSocket + MySQL). The UE5 side only handles UI and RPC bridging — no trade storage or logic on the client/server. Since the backend validates everything (item ownership, swap via atomic DB transaction), the Actor vs Component debate doesn't really apply to my case. I'm going with a Component on the PlayerController just for sending RPCs and managing the trade UI. Appreciate the help!

[meager spade]

websocket and mysql

looks at rest api as more secure

thats how our backend system works

just a bunch of rest api calls

[delicate shard]

yeah that's how I'd approach it as well. Have server validation when a player selects an item to trade to make sure they have it in inventory and such

[limber gyro]

Im having a bit of an issue with clients missing when they are suposed to hit, this doesnt happen in the server, most likely because the server is seeing the actual game and not the delayed version.

My question is, whats the term used to describe the fix for this? projectile server prediction? i think the server has to store some seconds of data so that it can rewind acording to the clients delay, how do you do this? does unreal have anything for this out of the box? could some one point me in the right direction please

[haughty ingot]

Generally the advanced solution for this is for the client to send a packet of data containing what they’re claiming they hit, and at what time. The server receives it, rolls back a frame buffer they have containing all relevant information about the game state at that time stamp, and then validate the hit on their end. Confirming that the players were in normal positions, and the hit player was in fact at (or relatively near) that position at the time the client says they shot.

It’s not a perfect solution, interpolation introduces some variation, but that’s generally the technique

Valorant has a good video out about it somewhere, they run some animations on the server to get very precise hit confirmations

[limber gyro]

so its a client authoritative hit with a server check

[haughty ingot]

Not client authoritative, never client authoritative. The server is not changing any game state based what the client tells it without confirmation. The client is just sending the data it’s claiming to have happened, there’s a lot of complexities because you have to account for round trip ping time in the rollback

Clients will just generally do some VFX predictably

A common trick is that hit markers are the actual confirmation the client gets back from the server, confirming the hit

[limber gyro]

lets say i store all the nescessary info like animations and meshes and what not, how do u do the collision check manually in the server? is there like a function to check for hits against meshes in the projectile that we can call? or do you have to code that yourself?

[haughty ingot]

Actual projectiles are harder, this is more for hit scans

Projectiles are typically done by spawning a local client instance, and then a server instance. When the client finally receives the server’s instance, it will interpolate its client local projectile to the servers

[limber gyro]

isnt that what the projectile movement component does?

[haughty ingot]

Hit scans are simple because you’d already have the hit position that was claimed and if you’ve kept track of animations on the server you can rollback the hit boxes and try to confirm it was possible

No not at all

Unreal Tournament has the code for this in the Epic repo

[limber gyro]

but if im just interpolating then im not really check if the client hits right?

[haughty ingot]

You’re interpolating the client’s local projectile to the authority projectile, to try to have to client’s visual accurate to what is actually happening on the server

The server’s projectile is still the authority

[limber gyro]

ok so i dont check i change the projectile position and the player adapts

doesnt seem like a very good solution

[haughty ingot]

It’s a solution that’s good enough for most AA and AAA titles including Unreal Tournament

[limber gyro]

ye but unreal tournament doesnt have very fast projectiles, and the ones that were faster like the rocket launcher had a very big splash radius

overwatch for example fakes fast projectiles with hitscans

[haughty ingot]

Yeah but I explained how hit-scans work too

[limber gyro]

tracer for example shots small pelets that travel very fast, but they arent actual pelets, its a bunch of traces and the pelets are just vfx

[haughty ingot]

Then I’m not really sure what you’re asking, I explained a way to deal with hit-scans, and then a way to deal with actual projectiles. You can use either or

[limber gyro]

ye, i understand that and thank you for the explanation, its just that the projectiles in my game are faster than unreals but they are not hitscans, so they are in between, so i might have to change a bunch of stuff, not rly happy about that

i was hoping there was a better solution

[haughty ingot]

Unreal Tournament forced open the actor channel for their fast traveling projectiles

[dark edge]

Depending on how crazy you're trying to get with it I'd just have client say "hey I hit that guy" and server do a sanity check.

If you're not making counter-strike then don't bother with rollback.

sanity check not being "did they hit that guy on the server" but "could they have presumably hit that guy on the server"

[limber gyro]

well im trying to make something with competitive play in mind, altough thats probably just a dream

also, wouldnt that be the same as just increasing the hitboxes a lot?

[dark edge]

Kind of, yes. It'd rely on client saying they hit them but yes a cheating client could effectively have bigger hitboxes.
If you have rollback then you can just say "projectile X hit hitbox Y at time Z" and roll back and check, although that'd be gobs more work. It's really no different from hitscan though. All a projectile IS is a short hitscan per frame that moves through the world if you think about it.

My projectile system has zero differentiaition between hitscan and projectile, hitscan is just a projectile with an effectively infinite speed. But they are also not actors, just structures.

If your projectiles are pretty much deterministic, meaning given the same launch position and launch velocity, they trace the same path through the world, then server and client can agree on that. What they WONT agree on is when and where colliders are, especially the colliders that are on other characters. By definition, at a given time snapshot, your screen, the server, and the enemies screen ALL disagree on where your projectiles and their hitboxes are. That's the fundamental problem that you have if movement is predicted (which it should be for a snappy game).

[limber gyro]

adding a delay = ur ping*2, wont that fix most of the desync issues?

[dark edge]

Do you want shots to be delayed?

[limber gyro]

well, if the delay is not noticeable and it can fix the issue in a simpler manner

[dark edge]

You'll have to delay all movement too and then you're back to not making a shooter

[limber gyro]

if the spawn time is syncd i assume that you would only have issues in weird network conditions

[limber gyro]

didnt think about that

[dark edge]

Even if the spawn time is perfectly synced, where you see another player, where they see themself, and where the server sees them are by definition different places.

Basically at synced real time T, you see the player at T + YourRTT/2 + TheirRTT/2, the server sees them at T+TheirRtt/2, and they see themselves at T

because movement is predicted.

[limber gyro]

how much work is it to code those rollbacks? im trying to figure out if its worth the hassle

[dark edge]

If you have to ask, a lot

[limber gyro]

figured

[dark edge]

If the confirmation can be "the shot coulda hit the other characters capsule" then it's easier

[limber gyro]

would make for great portfolio work tho

[dark edge]

but if you need to roll back to the EXACT same hitbox locations driven by animations, pretty hard.

[limber gyro]

ye but thats not a good aproach, for something like that i could probably just get away with making my collision really big and lowering character movement a bit

[dark edge]

It's a plenty good approach for some games. You are severely underestimating how hard this is.

[limber gyro]

hwo realistic is it to have the client decide when to hit and then have something like game guard to deal with cheaters?

[limber gyro]

i am not, thats why i didnt start coding the thing

and i am exploring every option

[dark edge]

Dunno, probably easier, def. less load on the server.
Big games spend millions on this and still have cheaters.

[limber gyro]

capsule option might be the better option here tbh

from what ive read saving all the info nescesary eats up ram quickly

[cinder quartz]

Need some feedback for my logic.
I have a weapon that fires shots via 3 functions
FireWeapon which kicks it off first detecting the input
ShootProjectile which checks, client side, for ammo and attempts to spend it, and issues new shots to fire if the weapon is auto and the player doesn't let go of the input
Then ShootProjectile_Server which sets up and spawns the projectile, then controls the timing by grabbing WorldTimeSeconds and setting the time for the next shot beforehand.

Here are my two isses: First, I need to add a condition for the process where a shot is not attempted if the player has a specific GameplayTag
Then, I realized that grabbing server time seconds may not be reliable given the desyncs that may happen.
Some advice on this could be helpful...

[crisp shard]

if refering to this node, then im pretty sure this is sync'd on client / server

by pretty sure, im 99.99p

[cinder quartz]

I used World time directly, not game state... I will use game states and update then

[limber gyro]

When using steam for sessions, when u join a session you also join a lobby, is this by default or am i missing something here?

[lament flax]

no, it can have severe latency if you want precise time

check this for examples to implement a more precise clock for clients https://vorixo.github.io/devtricks/non-destructive-synced-net-clock/

[shrewd ginkgo]

I have a multiplayer issue with ragdoll physics. When my character dies, I enable ragdoll simulation, but the corpse ends up in different positions on each client.

[kindred widget]

You can't really help that. If you want to sync it a bit, you could replicate a couple of bone locations, but unless you want to pay some expensive net costs, it'll still be off slightly on each machine. A bone or two replicated and positioned will get you in a similar ballpark though.

[nova wasp]

some people do a setup where they replicate a sort of invisible anchor where the clients all try to chain the ragdoll to that

to keep it roughly in the right area

replicating the entire pose would be absurdly expensive but TBH if you did it in separate pieces it's definitely doable if you don't need a billion of them a frame

and quantized it down a bit

if it was okay for it to teleport a bit to the final pose you could definitely replicate the full pose even if it's just the physics asset relevant ones like major limbs

[lost inlet]

The game state one is fine for most use cases

It used to update fairly infrequently resulting in pretty bad clock drift, but that was changed at some point in the latter UE4 releases

[lament flax]

it was written in 2022

i assume it took into consideration the fixs you are talking about

[lost inlet]

Well it does work with a frequent update rate

You don't always need to account for half RTT

[lament flax]

.

engine clock is fine in most cases

if you are not at 0.XX precision

[lost inlet]

It was set to update the replicated clock every 100ms, previous default was 5 seconds

[lament flax]

damn

[crisp shard]

Interesting, I may give that a go. But yea the one I mentioned is supposed to be synced, as close as anything else out of the box that is

Just read the above as well, nice

[limber gyro]

In the game mode is the function Logout called when you leave for the mainMenu for example?

assuming the main menu has a diferent game mode

[sinful tree]

In multiplayer, gamemode only exists on the server so if it's called, it's called only on the server. When a player leaves the server, regardless of reason, OnLogout would be called and it gives you reference to the playercontroller that left.

[limber gyro]

i forgot about it being in the server only, guess im gonna have to send an rpc

it would only work on the player hosting the session

[sinful tree]

Can't rely on that for all reasons a player may leave... Like... If tehy lose connection, they wouldn't be able to send the server an RPC.

Or the server to the client for that matter.

[limber gyro]

if they lose connection wont logout get called the same?

[sinful tree]

Yes, but you're saying you'd need to send an RPC.

[limber gyro]

im trying to leave a steam lobby when they leave a custom game, so if they lose connection the they are probably gonna timeout from the lobby

im assuming that case wont be an issue

[sinful tree]

What is the RPC for tho?

[limber gyro]

to tell the client to leave the steam lobby

[sinful tree]

If the client is disconnected, they wouldn't get that RPC.

[limber gyro]

the only way i can see it not working is if they have some sort of packet loss or connection gets interrupted for a few seconds and they go back to the main menu with a disconect but dont disconect from steam

[sinful tree]

Right. Which can absolutely happen. There's no reason to send an RPC to handle that.... Ideally you'd just have the client manage it themselves without the server having to tell them to do it.

[limber gyro]

well, i had it like that before but its ugly, i had to set a bool when the client joined a custom game and then when they left check that bool and make them leave the lobby

theres gotta be something more elegant

[sinful tree]

OnLogout on the game mode is mostly to handle cleaning up anything after a player leaves on the server end of things. Like, maybe it resets the server to a waiting mode after all players have disconnected, resetting scores, or other things that need to be handled with a player leaving the game that may change based on the type of gamemode used. All of this can be done without an RPC from the client.

There is a NetworkError and TravelError events in the GameInstance that you could leverage on clients to handle connecting or disconnect problems, which you could then call DestroySession on. You can have code on the client when you choose to manually disconnect to leave the game, like if you were to go into the menu and select "Leave Game", that can also just call DestroySession which will cause them to leave the game and if everything is set up right, return to your main menu.

[shrewd ginkgo]

I'm simulating the character's physics at the moment of death, but the body ends up in different places for each player. It doesn't have to fall exactly the same way; I just want to make sure the body stays in the same spot for all. How do I do that?

They should die with the ragdoll effect and end up in the same spot.

[tardy fossil]

the skeletal mesh ragdoll needs to be root component for it to replicate properly.. could just spawn in a new actor with the skeletal mesh as root or try and change it on the players pawn actor

another way is to set all bones except pelvis to simulate physics so it doesn't detach from the players collision capsule, but i've never been able to make that look decent

[chrome bay]

There are lots of ways to do it, what fits you depends on the game etc - but nothing about rag-dolling characters will stay synchronised out of the box.

[shrewd ginkgo]

All I want is for them to fall in the same spot. As long as everyone sees the body in the same place, it doesn’t matter how the body is lying on the ground.

[shrewd ginkgo]

How can I do that anchor thing

[nova wasp]

create an anchor on the authority that also applies to clients local version of the ragdoll. Replicate down the anchor position which is far cheaper

I believe it could be a typical constraint

[wary spade]

Hello everyone , can anyone help me with this
https://forums.unrealengine.com/t/separate-server-and-client-execution-for-listen-server-host/2718290

[fallen fossil]

Ayo, is it ok to RPC Data assets as pointers? would it be better and smaller packets than TSoftObjectPtr path? 🤔 , they always exist on client side sooo, technically ok? or will fail if not loaded

[nova wasp]

GetNetMode

[nova wasp]

same thing

they must be sent as package exports either way afaik

[fallen fossil]

"Package Exports" ?

[nova wasp]

sending an asset pointer will force it to be loaded in most cases

[nova wasp]

a string that is turned into a local ID for more effecient bandwidth later

which is how most assets are sent

[fallen fossil]

I knew they do that internally, but I though it some more effective way, okey thanks

[nova wasp]

fwiw you can just network profile it, I am not 100% sure if soft class paths change what they do based on loading state honestly